Marketing communication has always depended on images, but the nature of the image has changed. For most of the twentieth century, the product photograph was the standard unit of persuasion. Today, a growing number of firms replace photography with #3D_visualization: a digital model of the product that can be lit, textured, animated, placed in any scene, and rendered into an image that is difficult to distinguish from a photograph. This shift matters for international marketing because a single 3D asset can serve dozens of markets at once, and because a #photorealistic image carries a claim of factual accuracy that a stylized illustration does not. This article examines how objective, photorealistic imagery communicates #value_proposition content across cultural boundaries. It develops a conceptual framework, the Layered Visual Communication Model, which separates a rendered marketing image into five layers: geometry, material and light, scene and context, human presence, and interaction. The central argument is that the first two layers behave in a broadly culture-general way because they encode verifiable product facts, while the last three layers are culture-loaded and must be adapted. This distinction gives international marketers a practical rule for what to standardize and what to localize, and it gives researchers a testable structure for future empirical work. The article is an integrative conceptual review. It synthesizes literature on augmented and virtual reality marketing, mental imagery, presence, uncertainty reduction, and cross-cultural consumer behaviour, published mainly between 2021 and 2025. It offers eight research propositions, a measurement agenda, a managerial playbook, and a discussion of ethical risks, including deceptive realism and the uncanny valley. The article is written for students and early-stage researchers who want a structured entry point into a young and fast-moving research area. Keywords: 3D visualization; photorealism; global marketing; value proposition; cross-cultural communication; augmented reality; consumer trust; visual persuasion 1. Introduction 1.1 Background Ask a shopper in Lagos, Lima, or Leipzig what a product is like, and the answer usually begins with what they have seen. Vision dominates product judgment in digital settings because it is often the only sense available. The screen carries no weight, no smell, no resistance under the thumb. Everything the buyer knows before purchase must arrive through the eye, and increasingly through the ear and the fingertip on a touchscreen. For years, firms answered this constraint with photography. A studio, a lighting rig, a camera, a retoucher. That answer had two weaknesses. It was slow, and it was local. A photograph taken in Milan showed a Milanese sofa in a Milanese room. When that image travelled to Seoul or Sao Paulo, the sofa travelled with a room attached, and the room was not always welcome. #3D_visualization solves both weaknesses at once. A product is modelled once as a digital object. From that single object, a firm can generate an unlimited number of images, videos, configurators, and #augmented_reality experiences. The room is no longer fixed. The lighting is no longer fixed. The only thing that stays fixed is the product itself, which is precisely the thing a global marketer wants to hold constant. 1.2 The problem The promise is easy to state and hard to deliver. If a firm renders one image and pushes it to sixty markets, it has achieved efficiency but perhaps not effectiveness. Visual meaning is not universal. A colour that signals purity in one market signals mourning in another. A domestic scene that reads as aspirational in one country reads as unaffordable, foreign, or simply strange in another. A smiling model who looks like the target buyer in Amsterdam looks like a foreigner in Jakarta. At the same time, a product is a product. A drill has a chuck size. A shoe has a sole thickness. A refrigerator has a door that opens in a certain direction and a shelf that holds a certain number of bottles. These facts do not change when the image crosses a border. If a rendered image communicates them accurately, it has done something valuable and it has done it in a way that does not depend on cultural decoding. The tension between these two observations is the subject of this article. Photorealistic 3D imagery is simultaneously a factual instrument and a cultural artefact. Marketers who treat it only as the first will produce cold, foreign, unconvincing communication. Marketers who treat it only as the second will lose the main advantage of the technology, which is scalable #objectivity. 1.3 Why photorealism, specifically Not all 3D imagery is photorealistic. A brand can render a product as a line drawing, a clay model, a flat cartoon. Photorealism is a particular choice with particular consequences. When an image looks like a photograph, viewers treat it like a photograph. They assume it records something that exists. They assume it was not invented. That assumption is the source of the persuasive power of #photorealism, and it is also the source of its ethical danger, because the assumption is false. A rendered image records nothing. It constructs everything. This is why the word "objective" in the title of this article needs care. A rendered image can be objective in the sense that every dimension, material, and finish it shows corresponds to the physical product a customer will receive. It cannot be objective in the sense of being an unmediated record of reality. Objectivity in rendering is a discipline, not a default. It has to be built into the pipeline, verified, and defended. 1.4 Research objectives This article has four objectives: To clarify what "objective, photorealistic imagery" means in a marketing context, and to separate the verifiable parts of a rendered image from the interpretive parts. To explain the psychological mechanisms through which 3D imagery communicates value, drawing on #mental_imagery, presence, and #uncertainty_reduction theory. To map where #cultural_context enters the rendering process, and to propose a rule for what should be standardized globally and what should be localized. To set out a research agenda that students and early-career scholars can act on, including measurable constructs and study designs. 1.5 Research questions RQ1: Which components of a photorealistic 3D marketing image transfer across cultures without adaptation, and which do not? RQ2: Through which psychological mechanisms does photorealistic 3D imagery influence value perception and #purchase_intention? RQ3: How does the pursuit of visual #objectivity interact with the aspirational and symbolic goals of brand communication? RQ4: What are the ethical and regulatory boundaries of photorealistic product rendering in international markets? 1.6 Contribution The contribution is conceptual and pedagogical rather than empirical. Existing research treats immersive and 3D technologies mostly at the level of the whole experience, asking whether an #augmented_reality feature raises purchase intention or lowers returns. That question is useful but coarse. This article argues that the image should be opened up and examined layer by layer, because different layers behave differently across cultures. That is the theoretical move. The practical move is a decision rule that a marketing manager can apply on a Monday morning. 1.7 Structure Section 2 reviews the literature. Section 3 builds the theoretical framework and states eight propositions. Section 4 explains the method. Section 5 presents the analysis, layer by layer. Section 6 discusses implications for theory, practice, and ethics. Section 7 states limitations and a future research agenda. Section 8 concludes. 2. Literature Review 2.1 From text to image to model The history of product representation in commerce is a history of increasing #vividness. Catalogue text gave way to engraved illustration, then to photography, then to video, and now to interactive 3D. Each step added sensory information and reduced the gap between what the buyer could perceive and what the buyer would eventually own. Interactive 3D is different from video in one important way. Video is a fixed path through a product. The camera moves where the director chose. A 3D model is an open space. The viewer chooses the path. This shift from a fixed to an open representation introduces #interactivity as a distinct variable, separate from vividness, and the two do not always move together. A low-resolution model that the user can rotate freely may be more useful than a beautiful film the user can only watch. 2.2 Photorealism, rendering, and what makes an image believable Photorealistic rendering is now a mature engineering field. Physically based rendering simulates the behaviour of light using models grounded in optics, so that a metal surface scatters light like metal and a rough plastic scatters light like rough plastic (Pharr, Jakob, and Humphreys, 2023). The important implication for marketing is that believability is not achieved by adding detail. It is achieved by being consistent with how light actually works. Viewers cannot articulate the rules, but they notice violations instantly. Two recent developments have lowered the cost of realism. Neural radiance fields reconstruct a scene from a set of ordinary photographs and allow new viewpoints to be synthesized (Mildenhall and colleagues, 2021). #Gaussian_splatting achieves similar reconstruction quality with real-time rendering speed (Kerbl, Kopanas, Leimkuhler, and Drettakis, 2023). Together, these techniques mean that a firm no longer always needs an expert modeller. A phone, a turntable, and good lighting can produce a usable asset. This matters enormously for smaller firms in emerging markets, and it is why #neural_rendering deserves attention in marketing curricula, not only in computer science curricula. 2.3 Immersive technologies in marketing The marketing literature on immersive technology has grown quickly. Rauschnabel, Babin, tom Dieck, Krey, and Jung (2022) define augmented reality marketing as a strategic subdiscipline rather than a one-off campaign tactic, and they propose a customer journey model with branding, inspiring, convincing, and keeping as the four functional goals. Their companion conceptual work with Felix, Hinsch, Shahab, and Alt (2022) separates augmented from virtual reality along two continua, local presence and #telepresence, and warns against the loose use of the term XR. Empirical work supports the broad claim that these technologies change consumer response. Kowalczuk, Siepmann, and Adler (2021) compared cognitive, affective, and behavioural responses to AR in e-commerce and found meaningful differences relative to conventional presentation. Gatter, Huttl-Maack, and Rauschnabel (2022) examined whether AR can satisfy the #need_for_touch, a question that goes to the heart of why 3D matters for tactile product categories. Reviews by Kumar (2022) and by Erensoy and colleagues (2024) summarize a body of evidence that is consistent in direction but variable in effect size, which is exactly what one would expect from a field that has not yet agreed on how to describe the stimulus. The metaverse literature adds scale to the discussion. Dwivedi and colleagues (2022) give a multidisciplinary account of the opportunities and risks of persistent virtual environments, and Barrera and Shah (2023) propose a conceptual framework for #metaverse_marketing. Hennig-Thurau and colleagues (2023) provide early evidence on social interaction in these environments. For present purposes, the metaverse is best understood not as a destination but as the extreme case of the same underlying trend: commerce mediated entirely by rendered imagery. 2.4 Culture and visual meaning Cross-cultural marketing has long argued between standardization and adaptation. The debate is old, but 3D changes its economics. Previously, adaptation meant reshooting. Now, adaptation means re-rendering, which is cheaper by an order of magnitude. When the cost of adaptation falls, the optimal amount of adaptation rises. That simple economic point is under-discussed in the literature and deserves emphasis. The dominant framework for describing cultural difference remains the dimensional approach associated with Hofstede, refined and re-validated in later work (Minkov and Kaasa, 2022). Studies applying these dimensions to immersive technology find that acceptance is not uniform. Irshad and colleagues (2025) report cross-cultural differences in how AR shapes the retail shopping experience, and Connell, Marciniak, and Drylie Carey (2023) show that the culture attached to a brand can matter as much as the national culture of the consumer. Dimensional models are useful but blunt. They describe national averages, not individuals, and they are weakest exactly where marketers most need precision, namely in predicting how a specific image will be read by a specific segment. This article therefore uses cultural dimensions as a sensitizing device rather than as a predictive engine. 2.5 Value propositions and how images carry them A value proposition is a claim about the benefit a customer will receive. Claims can be functional, economic, emotional, social, or ethical. Images carry these claim types with different efficiency. A functional claim about size or fit can be shown directly. An economic claim about durability can only be implied. A social claim about status depends entirely on the audience sharing the code. This asymmetry is the hinge of the argument developed later. Functional claims are the natural territory of objective #photorealism. Symbolic claims are the natural territory of scene, styling, and human presence. A single rendered image usually carries both at once, which is why the layered view is necessary. 2.6 The photographic assumption There is a habit of mind that this literature has not yet named clearly, so it will be named here as the photographic assumption. It is the belief, held automatically and usually without reflection, that an image which looks like a photograph must have been produced by a camera pointed at something real. The assumption has deep roots. For more than a century, courts, newspapers, insurers, and shoppers treated photographs as evidence. A photograph could be staged, cropped, or retouched, but something had to be in front of the lens. That residue of reality was the basis of photographic credibility. Rendering breaks the chain. A photorealistic render is a calculation, not a record. Nothing needs to exist. The product in the image may never have been manufactured. The room may never have been built. The light may come from a sun that does not shine anywhere. Consumers have not yet updated their instincts to match this. Most viewers, in most categories, still apply the photographic assumption to rendered marketing images, either because they do not know the image is synthetic or because they do not think about it at the moment of viewing. This is why photorealism persuades. It borrows credibility from a technology it has replaced. Two consequences follow. First, the persuasive advantage of #photorealism is partly borrowed and may erode as public awareness of synthetic media grows. Firms that build trust on accuracy rather than on realism alone will survive that erosion. Second, misuse of the assumption is a form of #deceptive_advertising even when no individual claim in the image is literally false, because the image as a whole implies a kind of evidence that it does not provide. 2.7 Gaps in the literature Four gaps stand out. First, the stimulus is under-specified. Papers frequently compare "AR" with "no AR" or "3D" with "2D" without describing the render quality, the lighting, or the scene. Two studies using the word "3D" may be studying different things. Second, culture is treated as a moderator variable and rarely as a design input. The typical question is whether the effect is stronger in country A than country B. The more useful question is which element of the image should change between A and B. Third, #objectivity is asserted rather than measured. Almost no marketing study reports whether the rendered product matched the physical product in colour, scale, or finish. Fourth, ethics arrives late. Discussion of #deceptive_advertising in rendering usually appears in the final paragraph of papers rather than in their design. 3. Theoretical Framework 3.1 Mental imagery and embodied simulation When people look at a product, they do not merely see it. They imagine using it. #Mental_imagery theory holds that consumers construct an internal simulation of ownership and use, and that this simulation, more than the raw information in the image, drives evaluation. Richer, more concrete stimuli generate stronger simulations. Interactive 3D supports simulation in a way that static photography does not, because the user's own action generates the next view. Turning a chair to see its back is a small act of agency, and agency makes the simulation feel personal. This is one reason why #interactivity often outperforms passive vividness in experimental work. 3.2 Presence and telepresence Presence is the sense of being somewhere, and telepresence is the sense of being somewhere other than one's physical location. Rauschnabel and colleagues (2022) argue that augmented reality primarily generates local presence, the sense that a virtual object is genuinely in the user's own space, while virtual reality generates telepresence, the sense of having travelled elsewhere. The distinction has direct managerial meaning. If a firm wants a customer to judge whether a sofa fits a living room, it needs local presence, and therefore AR. If a firm wants a customer to feel the atmosphere of a resort, it needs telepresence, and therefore VR or high-quality cinematic rendering. 3.3 Uncertainty reduction and perceived diagnosticity Buying online involves uncertainty. Will it fit. Will the colour match. Is it as large as it looks. Uncertainty is costly for the buyer, who may hesitate, and for the seller, who may receive a return. #Perceived_diagnosticity is the extent to which a presentation format helps a consumer evaluate a product. It is the most useful single construct in this literature, because it links a design feature directly to a business outcome. High diagnosticity should reduce #product_returns. This is a testable and commercially meaningful claim, and it is under-tested in cross-cultural settings. 3.4 Cognitive load More information is not always better. A high-detail render, a configurator with two hundred options, and a rotating scene with reflections may exceed the working memory of a shopper on a small screen on a slow connection. #Cognitive_load must therefore be treated as a boundary condition, not a footnote. The design goal is diagnostic richness at low load, which usually means fewer choices, cleaner backgrounds, and progressive disclosure. 3.5 Cultural context as a communication variable The distinction between high-context and low-context communication is useful here. In #high_context_cultures, meaning depends heavily on setting, relationship, and implication. In #low_context_cultures, meaning is expected to be explicit. Applied to imagery, this suggests that audiences differ in how much they expect the scene around a product to do communicative work. Dimensions such as #uncertainty_avoidance and #power_distance also have visual consequences. Audiences high in uncertainty avoidance may value explicit measurements, certification marks, and comparison scales inside the image. Audiences high in power distance may respond to imagery that signals expert endorsement or premium context. These are hypotheses, not findings, and they are stated as propositions below. 3.6 The Layered Visual Communication Model The framework proposed here decomposes a rendered marketing image into five layers. Layer 1: Geometry. The shape and dimensions of the product. Culture-general. A handle is a handle everywhere. Layer 2: Material and light. Surface finish, colour, reflectivity, and the light that reveals them. Mostly culture-general, with important exceptions around colour symbolism and skin-tone rendering. Layer 3: Scene and context. The environment in which the product sits. Strongly culture-loaded. Rooms, streets, weather, furniture, food, plug sockets, and architectural style all encode place. Layer 4: Human presence. Models, hands, avatars, virtual influencers. The most culture-loaded and the most ethically sensitive layer. Layer 5: Interaction. The interface through which the user manipulates the object. Culture-loaded in a quieter way, through language, reading direction, gesture conventions, and device norms. The central claim is that layers 1 and 2 should be standardized and treated as factual assertions subject to verification, while layers 3, 4, and 5 should be localized and treated as rhetorical choices subject to cultural judgment. 3.7 Propositions P1. Photorealistic rendering of geometry and material increases #perceived_diagnosticity, and this effect is stable across cultures. P2. The effect of Layer 1 and Layer 2 fidelity on trust is mediated by the perceived correspondence between the rendered product and the delivered product. P3. Localization of Layer 3 (scene) increases self-referencing and purchase intention, and the size of this gain is larger in high-context cultures. P4. Localization of Layer 4 (human presence) increases identification and perceived brand relevance, with the largest gains among consumers who are visually under-represented in the category. P5. Excessive realism in Layer 4 without behavioural realism produces eeriness and reduces trust, consistent with #uncanny_valley reasoning. P6. High-fidelity rendering increases purchase intention only up to the point where #cognitive_load rises sharply, after which the effect reverses. P7. Diagnostic 3D presentation reduces product return rates, and the reduction is largest in categories with high fit or scale uncertainty. P8. Discovering that a photorealistic image misrepresented the product produces a larger trust penalty than discovering the same misrepresentation in an obviously stylized image. 4. Methodology 4.1 Research design This study is an integrative conceptual review. It does not collect primary data. It synthesizes published research and industry practice in order to build a framework and generate propositions. This design is appropriate when a field is fragmented, when constructs are inconsistently defined, and when the practical questions have outrun the empirical evidence. All three conditions apply here. 4.2 Literature identification Sources were identified from peer-reviewed marketing, information systems, and computer graphics outlets, with priority given to work published between 2021 and 2025. Search terms combined technology terms (augmented reality, virtual reality, 3D visualization, photorealistic rendering, metaverse) with marketing terms (value proposition, purchase intention, brand trust, product presentation) and cultural terms (cross-cultural, cultural dimensions, localization, standardization). Priority was given to conceptual papers that define constructs, to reviews and meta-analytic work that summarize effect directions, and to empirical papers that manipulate visual features rather than merely observe technology adoption. 4.3 Analytical procedure The analysis proceeded in three passes. The first pass extracted the mechanisms proposed in each source. The second pass extracted the visual features that were manipulated or discussed. The third pass mapped features to layers and mechanisms to outcomes, which produced the model in Section 3. 4.4 Boundaries and validity Conceptual work is vulnerable to confirmation bias. The framework was therefore tested against disconfirming cases, in particular the observation that greater realism sometimes reduces liking, which the model must accommodate rather than ignore. The uncanny valley and cognitive load boundaries in P5 and P6 exist for that reason. The framework is offered as a structure for inquiry, not as a validated theory. Every proposition is stated in a form that can be falsified. 5. Analysis and Findings 5.1 Layer 1: Geometry as the culture-free core Geometry is where the case for #objectivity is strongest. A product has one shape. That shape does not vary by market. If a firm builds its model from engineering data rather than from an artist's interpretation, the resulting geometry is a factual assertion that can be checked with a ruler. Three practices follow. First, the source of truth should be the design file used in manufacturing, not a re-creation. Second, scale must be preserved absolutely, in real-world units, so that an AR placement in a customer's room is trustworthy. Third, the model should carry metadata describing its provenance and tolerance, so that anyone reusing it downstream knows what it is. Geometric objectivity has an underappreciated commercial payoff. It supports honest comparison. A shopper comparing three washing machines from three brands can only compare them fairly if each is modelled at true scale. Firms that adopt this discipline gain from it, because in most categories the product that is described accurately is the product that gets returned least. The link to #digital_twin thinking is direct. A digital twin is a virtual counterpart of a physical asset maintained across its lifecycle (Attaran and Celik, 2023). Marketing has usually treated the marketing asset and the engineering asset as separate objects with separate owners. The argument here is that they should converge. The marketing render should be a view of the engineering truth, not a parallel fiction. 5.2 Layer 2: Material, light, and the problem of colour Material and light are where realism is won or lost. A viewer will forgive an imperfect background. They will not forgive a metal that looks like plastic. Physically based material definitions describe surfaces in terms of measurable properties, which allows a fabric to behave like that fabric under any lighting. This is the technical foundation of cross-market consistency, because it means the same asset can be relit for a bright Middle Eastern showroom or a soft Northern European interior without the product changing identity. #Color_management is the hardest practical problem in this layer. Colour depends on the light, the display, the viewer's adaptation, and the colour space of the file. A rendered navy that appears correct on a calibrated studio monitor may appear black on a budget phone in daylight. Because colour is one of the most frequent causes of dissatisfaction and return in fashion and home categories, colour discipline is not a technical footnote. It is a trust mechanism. Colour also carries meaning, and meaning varies. White signals purity in some traditions and mourning in others. Red signals luck and celebration in parts of East Asia and danger or warning elsewhere. The important distinction is between the colour of the product, which is a fact, and the colour of the scene, which is a rhetorical choice. The first belongs to Layer 2 and should not move. The second belongs to Layer 3 and should. 5.3 Layer 3: Scene, context, and the return of culture The scene is everything around the product. It is also where most cross-cultural failure happens, because scenes leak information that designers do not consciously encode. Consider a rendered kitchen used to sell a blender. The kitchen contains a window, and through the window there is a street. The street has architecture, vegetation, and weather. There is a power socket on the wall. There is a bowl of fruit on the counter. Every one of these details tells the viewer where they are, and by implication, whether the product is for people like them. Because the 3D pipeline separates the product from the scene, this problem is unusually tractable. A firm can maintain one product asset and a library of scenes, and swap them per market. The marginal cost is a render, not a photoshoot. This is the single greatest operational advantage of 3D in #global_marketing, and many firms use only a fraction of it. Scene #localization should be evidence-led rather than stereotype-led. The failure mode is a caricature, which is worse than a neutral scene. A neutral studio background is often the safest global default, and a properly researched local scene is the best local option. A poorly researched local scene, full of borrowed clichés, is the worst of both. There is also a socioeconomic dimension that is often confused with culture. An aspirational scene shows a home that most viewers do not have. Whether that raises desire or produces alienation depends less on nationality than on income distribution, housing norms, and the category. Marketers should not label an income effect as a cultural effect. 5.4 Layer 4: Human presence, avatars, and virtual influencers Once a human enters the frame, everything changes. Viewers process faces differently from objects. They read status, age, ethnicity, mood, and intention within milliseconds. Rendered humans allow #avatar_diversity at low marginal cost. A brand can present the same garment on many body types and skin tones without a new shoot. This is a real opportunity for inclusive representation, and it is also an opportunity for cynical box-ticking. Consumers can usually tell the difference, and diversity that appears only in the render and never in the product line is quickly detected. The uncanny valley remains a live risk. Evidence on #virtual_influencers suggests that high anthropomorphism can raise uncanniness and reduce persuasion, though the relationship is not simple and is moderated by social cues (Gutuleac and colleagues, 2024; Franke, Groeppel-Klein, and Muller, 2023). The practical guidance is that form realism and behavioural realism must be matched. A face that looks entirely human but moves slightly wrongly is more unsettling than a face that is clearly stylized and moves consistently with its style. Disclosure is the other issue. If a human endorser is synthetic, the audience is entitled to know. Regulatory expectations are moving in this direction in several jurisdictions, and it is a poor strategy to wait until disclosure becomes compulsory. Skin-tone rendering also belongs here and is often mishandled. Rendering pipelines calibrated on light skin can misrepresent darker skin under certain lighting models. This is not merely an aesthetic failing. In cosmetics and apparel it is a factual failing, because the customer is trying to judge how a product will look on their own body. 5.5 Layer 5: Interaction and interface The interaction layer determines whether the visual asset can actually be used. It includes rotation controls, zoom, configurators, AR placement, and the text that explains them. Several factors vary internationally. Reading direction affects layout and the intuitive direction of sliders and carousels. Device capability varies sharply, and a heavy real-time asset that runs on a flagship phone in one market may fail entirely in another. Data cost is a genuine constraint in many countries, and a 40-megabyte model is a barrier disguised as a feature. This produces an uncomfortable conclusion. The most technically impressive experience is often the least globally available one. #Accessibility and progressive enhancement therefore belong in the marketing brief, not only in the engineering brief. A well-optimized asset that loads in two seconds on a mid-range phone will out-earn a beautiful asset that never finishes loading. 5.6 The objectivity paradox Here is the paradox at the centre of this article. The persuasive strength of photorealistic imagery comes from its apparent objectivity. But its commercial use is persuasive, and persuasion is not neutral. Every render involves choices: the lens, the light, the angle, the finish, the moment. Two positions are available, and both are inadequate on their own. The purist position says the render must show only what is true. This produces accurate, dull imagery that fails to build brands. The rhetorical position says the render is advertising and advertising has always idealized. This produces beautiful imagery that erodes trust the moment the parcel is opened. The workable position is a division of labour, which the layered model makes explicit. Layers 1 and 2 are held to a standard of factual accuracy. Layers 3, 4, and 5 are recognized as rhetoric and judged by cultural fit and honesty of impression. Under this rule, a brand may show its lamp in a beautiful room it does not own, but it may not make the lamp look larger, brighter, or better finished than it is. This is not just an ethical rule. It is an operational one, because it tells the render team which choices require sign-off from engineering and which require sign-off from the local market team. 5.7 The single-asset, many-markets pipeline The economic logic of 3D in global marketing can be summarized in one sentence. Model once, render many times. A mature pipeline has five stages. Capture or model the product from engineering data. Validate the asset against the physical product. Store it centrally with clear versioning. Compose market-specific scenes and human presence around it. Render and distribute at appropriate quality for each channel and device. #Asset_management is the stage most often neglected and the one that most often kills the value case. When there are six versions of a model in circulation and nobody knows which is current, the accuracy promise collapses. A governance rule is needed: one canonical asset per product, one owner, one validation record. Two further benefits deserve mention. First, speed. A product that does not exist physically can still be visualized, which compresses the launch calendar. Second, #sustainability. Fewer physical samples, fewer shoots, and fewer flights carry an environmental benefit, although honest accounting must also include the energy cost of rendering, which is not trivial. 5.8 Trust, returns, and the business case The business case for 3D usually rests on three claims: higher conversion, lower returns, and lower content cost. The third is straightforward. The first two depend on diagnosticity and honesty. Returns deserve particular attention because they are where accuracy pays. A return happens when the delivered product violates an expectation the image created. Every avoidable return is a failure of visual communication. This reframes the render team as a cost centre in reverse: they are one of the few groups in the company who can reduce reverse logistics by doing their job well. #Trust_calibration is the goal. The aim is not maximum trust but correct trust: the customer should expect exactly what will arrive. Over-promising raises short-term conversion and destroys long-term value. Under-promising is wasteful. Accurate imagery is the mechanism that aligns the two. 5.9 Category differences The value of 3D varies sharply by category, and any global rollout should sequence accordingly. Categories with high spatial uncertainty benefit most: furniture, appliances, vehicles, industrial equipment, and anything installed in a space. Here, scale and fit are the dominant questions, and AR placement answers them directly. Categories with high tactile uncertainty benefit moderately: apparel, textiles, and materials. Vision can suggest texture but cannot deliver it, which is why the need-for-touch question remains open. Categories with low uncertainty benefit least: commodities, consumables, and familiar repeat purchases. Rendering a bag of rice in photoreal 3D is a solution in search of a problem. 5.10 Measuring the effect A recurring weakness of the field is measurement. If students take one methodological lesson from this article, it should be to describe the stimulus properly. A research paper on 3D imagery should report the polygon count or equivalent fidelity measure, the lighting model, the render resolution, the device used, the loading time, and the degree of interactivity permitted. Without this, results cannot be compared or replicated. On the outcome side, the most informative variables are perceived diagnosticity, product-related uncertainty, mental imagery vividness, self-referencing, perceived realism, eeriness, purchase intention, willingness to pay, and where possible, actual return rates. Behavioural and field data are scarce in this literature and are the fastest route to a meaningful contribution. 5.11 A worked example The framework is easier to see through an example. The example below is illustrative and hypothetical. It is offered as a teaching device, not as a report of a real firm. A mid-sized European furniture brand sells a modular sofa. It plans to enter three new markets that differ in climate, housing, income distribution, and communication style. Its current visual assets are photographs taken in a Scandinavian apartment with pale wood floors, large windows, and low winter light. Under the old model, entering three markets meant three shoots, three crews, three logistics chains, and a set of images that would be out of date the moment the product range changed. Under a 3D pipeline, the brand builds one validated model of the sofa from its manufacturing data. Layer 1. The geometry comes from the engineering file. Seat height, depth, arm width, and module dimensions are exact. The brand publishes those dimensions alongside the render and offers AR placement at true scale. This layer does not change in any market. A customer measuring a wall in any of the three markets gets the same true answer. Layer 2. Fabric materials are scanned rather than approximated, so a linen weave scatters light like linen. Colour is managed end to end, and the brand tests each colourway on a range of common local devices rather than only on studio monitors. This layer also does not change, because the fabric is the fabric. What does change is the lighting used to reveal it, since a fabric shown under harsh equatorial daylight and under grey northern light is the same fabric, and the brand wants it to read as the same fabric in both. Layer 3. The scene changes completely. In one market the sofa is placed in a compact apartment with a balcony, in another in a multi-generational living room with a large low table, in the third in a house with tiled floors and shuttered windows. These scenes are built from field research: photographs of real customer homes, not from a stock image library. The sofa itself is untouched. Layer 4. Human presence is localized. Body types, clothing, ages, and group composition reflect local household patterns. In one market, the scene shows several generations together, because that is who uses the living room. In another, it shows a single occupant, because that is the target segment. Layer 5. The interface is rebuilt for the lowest common device in each market. The heavy configurator remains available on desktop, but the mobile version loads a compressed model in under two seconds and offers three material choices rather than thirty. The outcome the brand should track is not applause for the imagery. It is the return rate, the reasons customers give for returning, and the conversion difference between products with and without validated assets. If the sofa is returned because it was larger than expected, Layer 1 failed. If it is returned because the colour was wrong, Layer 2 failed. If nobody clicked in the first place, the failure was in Layers 3, 4, or 5. The layered model turns a vague complaint about imagery into a diagnosable fault. 6. Discussion 6.1 Theoretical implications The main theoretical implication is that "3D" and "AR" are not stimuli. They are containers for stimuli. Treating them as single independent variables has produced a literature with inconsistent effect sizes, because studies labelled the same are in fact manipulating different things. The layered model offers a vocabulary for describing what was actually manipulated. The second implication concerns culture. The prevailing approach treats culture as a moderator sitting outside the message. The layered model brings culture inside the message and locates it in specific layers. This converts an abstract moderation question into a concrete design question, and design questions are easier to test experimentally. The third implication concerns objectivity. Marketing theory has surprisingly little to say about factual accuracy in imagery, largely because photography was assumed to be accurate by default. Rendering removes that assumption. Accuracy must now be theorized as a constructed property with antecedents, costs, and consequences. 6.2 Managerial implications: a working playbook For practitioners, the framework produces a sequence of decisions. One. Build the canonical asset from engineering data. Validate it against the physical product for scale, colour, and finish. Record the validation. Two. Standardize Layers 1 and 2 globally. Do not permit local markets to alter product geometry, proportions, or finish for aesthetic reasons. Three. Localize Layer 3. Maintain a scene library. Base scene choices on research into actual homes, streets, and usage conditions in the market, not on assumptions. Four. Localize Layer 4 with care and with disclosure. Match form realism to behavioural realism. Ensure that represented diversity is reflected in the actual product range. Five. Engineer Layer 5 for the worst device in the target market, not the best. Measure load time and drop-off, and treat them as marketing metrics. Six. Instrument the outcome. Track conversion and returns for products with and without 3D presentation. Feed return reasons back to the render team. Seven. Govern the assets. One canonical model per product. One owner. Clear versioning. Retire obsolete assets aggressively. 6.3 Ethical implications Three ethical risks deserve explicit statement. The first is deceptive realism. Because rendered images inherit the credibility of photographs, misrepresentation in a render is more damaging than misrepresentation in an illustration. Firms should hold rendered product claims to at least the standard applied to written claims. The second is manufactured aspiration. Rendering makes it cheap to place any product in a flawless environment. At scale, this contributes to a visual culture in which ordinary life appears deficient. Marketers are not solely responsible for this, but they are participants in it, and the layered model at least makes the choice visible. The third is representational harm. Synthetic humans can be made to look like anyone. They can also be made to look like a narrow and unrealistic ideal, repeated at industrial scale, in every market. #Ethical_marketing here means treating human representation as a decision with consequences, not as an asset selection. 6.4 Implications for students For students, the practical message is that this field rewards people who can hold two literacies at once. Someone who understands lighting and materials but not consumer psychology will produce beautiful images that do not persuade. Someone who understands persuasion but not the pipeline will write briefs that cannot be executed. #Visual_literacy in the technical sense is now part of the marketing skill set. The research message is that the barrier to entry is low. A student with a phone camera, free rendering software, and a small online panel can run a well-designed experiment that manipulates one layer while holding the others constant. Very few such studies exist. The gap between what practitioners are doing and what researchers have measured is unusually wide, and that gap is where a thesis lives. 6.5 Generative imagery and the accuracy problem Any discussion of rendered marketing imagery written now must address generative image models, because they are already in the pipeline of many firms. Generative models produce images that look right. That is what they are optimized to do. They are not optimized to be dimensionally correct, and they have no access to the engineering truth of a product unless that truth is supplied to them. A generated image of a product can invent a seam, remove a button, or subtly change a proportion, and the result will still look convincing, which is precisely the danger. The distinction that matters is between generating the product and generating the world around the product. Generating the world, the scene, the light, the background, the atmosphere, is a legitimate and efficient use of the technology, and it sits squarely in Layer 3. Generating the product itself is a violation of the accuracy discipline argued for in this article, because the output is not derived from the object. The safe pattern is therefore a hybrid one. The product comes from the validated 3D asset. Everything around it may be generated, subject to cultural review. This preserves the factual core while capturing most of the cost saving. It also gives a clean audit rule: if a product pixel was generated rather than rendered from the canonical model, it should not ship. This is one of the most urgent open questions in the field, and it is a good one for students. There is very little published evidence on how consumers respond when they learn that a product image was generated rather than photographed or rendered from engineering data, or on whether that response varies across markets. 6.6 Skills and curriculum If the argument here is right, marketing education needs to change in a modest but specific way. Students do not need to become technical artists. They do need to understand enough about materials, lighting, scale, and file formats to write a brief, to judge whether an asset is honest, and to ask the right question when an image looks wrong. A short practical module can achieve most of this: model a simple object, light it two ways, place it in two scenes, and test the two versions on a small sample. The learning is not in the software. It is in seeing, with one's own hands, how much of the meaning of an image comes from choices that had nothing to do with the product. 7. Limitations and Future Research 7.1 Limitations This article is conceptual. Its propositions are argued, not tested. The literature it draws on is concentrated in a small number of countries and product categories, which means the cultural claims rest on a narrower base than one would like. Technology is also moving faster than peer review, and some technical claims will date quickly. Finally, the layered model is a simplification. Real images blur the layers, and a reflective product surface literally contains the scene in it, which is a nice reminder that the separation is analytical rather than physical. 7.2 Future research agenda Layer-isolating experiments. Manipulate one layer while holding others constant. Does scene localization add value once geometry is accurate. Does human presence add value once the scene is local. Cross-cultural replication with proper stimulus reporting. Run the same manipulation in several markets and report the stimulus in enough detail for others to reproduce it. Field studies on returns. Partner with a retailer. Compare return rates and return reasons for products presented with validated 3D assets versus photography. The accuracy-attractiveness frontier. Identify the point at which idealization begins to cost more in trust than it gains in appeal. Device and bandwidth inequality. Study how visual richness interacts with connection quality and device capability in emerging markets. This is where the largest untapped commercial value lies and where research is thinnest. Synthetic humans and disclosure. Test how disclosure of a synthetic model affects trust across cultures. Disclosure may be penalized in some markets and rewarded in others, and firms need to know which. Neural capture methods. Assess whether assets produced by radiance-field or splatting techniques are perceived as more or less credible than artist-built models, and whether their imperfections help or hurt. Longitudinal trust effects. Most studies measure intention at a single moment. The interesting question is what happens after delivery, when the render meets the product. 8. Conclusion 3D visualization gives global marketers something they have never had before: one representation of a product that is simultaneously factual, flexible, and infinitely reproducible. The factual part is what makes it valuable. The flexible part is what makes it global. The reproducible part is what makes it cheap. But the same technology that lets a firm show a product truthfully in sixty markets also lets it show a product untruthfully in sixty markets, at no extra cost. The technology is neutral. The pipeline is not. The argument of this article has been that the way out of this tension is structural, not moral. Split the image into layers. Hold the product layers to a standard of verifiable accuracy and standardize them everywhere. Recognize the scene, the people, and the interface as rhetorical and cultural, and adapt them thoughtfully, locally, and with evidence. Firms that adopt this discipline will find that accuracy is not the opposite of persuasion. In a market where every image is constructed and every buyer knows it, the brand whose images can be trusted has an advantage that no amount of rendering horsepower can imitate. For researchers and students entering this area, that is also where the most interesting questions are: not whether 3D works, but which part of the image is doing the work, for whom, and at what cost to the truth. Hashtags #3DVisualization #PhotorealisticImagery #GlobalMarketing #CrossCulturalMarketing #ValueProposition #ImmersiveTechnology #AugmentedRealityMarketing #VisualCommunication #ConsumerBehaviour #InternationalBusiness #MarketingStrategy #XRMarketing #EcommerceInnovation #DigitalMarketingResearch #STULIB References Attaran, M., and Celik, B. G. (2023). Digital twin: Benefits, use cases, challenges, and opportunities. Decision Analytics Journal, 6, 100165. doi:10.1016/j.dajour.2023.100165 Barrera, K. G., and Shah, D. (2023). Marketing in the metaverse: Conceptual understanding, framework, and research agenda. Journal of Business Research, 155, 113420. doi:10.1016/j.jbusres.2022.113420 Buhalis, D., Lin, M. S., and Leung, D. (2023). Metaverse as a driver for customer experience and value co-creation: Implications for hospitality and tourism management and marketing. International Journal of Contemporary Hospitality Management, 35(2), 701 to 716. doi:10.1108/IJCHM-05-2022-0631 Chiu, C. L., Ho, H. C., Yu, T., Liu, Y., and Mo, Y. (2021). Exploring information technology success of augmented reality retail applications in retail food chain. Journal of Retailing and Consumer Services, 61, 102561. doi:10.1016/j.jretconser.2021.102561 Connell, C., Marciniak, R., and Drylie Carey, L. (2023). The effect of cross-cultural dimensions on the manifestation of customer engagement behaviors. Journal of International Marketing, 31(2). doi:10.1177/1069031X221130690 Dwivedi, Y. K., Hughes, L., Baabdullah, A. M., Ribeiro-Navarrete, S., Giannakis, M., Al-Debei, M. M., and colleagues (2022). Metaverse beyond the hype: Multidisciplinary perspectives on emerging challenges, opportunities, and agenda for research, practice and policy. International Journal of Information Management, 66, 102542. doi:10.1016/j.ijinfomgt.2022.102542 Erensoy, A., Mathrani, A., Schnack, A., Elms, J., and Baghaei, N. (2024). Consumer behavior in immersive virtual reality retail environments: A systematic literature review using the stimuli-organisms-responses model. Journal of Consumer Behaviour, 23(6), 2781 to 2811. doi:10.1002/cb.2377 Franke, C., Groeppel-Klein, A., and Muller, K. (2023). Consumers responses to virtual influencers as advertising endorsers: Novel and effective or uncanny and deceiving. Journal of Advertising, 52(4), 523 to 539. doi:10.1080/00913367.2022.2154721 Gatter, S., Huttl-Maack, V., and Rauschnabel, P. A. (2022). Can augmented reality satisfy consumers need for touch. Psychology and Marketing, 39(3), 508 to 523. doi:10.1002/mar.21618 Gutuleac, R., and colleagues (2024). Will virtual influencers overcome the uncanny valley: The moderating role of social cues. Psychology and Marketing, 41, 1419 to 1431. doi:10.1002/mar.21989 Hennig-Thurau, T., Aliman, D. N., Herting, A. M., Cziehso, G. P., Linder, M., and Kubler, R. V. (2023). Social interactions in the metaverse: Framework, initial evidence, and research roadmap. Journal of the Academy of Marketing Science, 51, 889 to 913. doi:10.1007/s11747-022-00908-0 Hilken, T., Chylinski, M., Keeling, D. I., Heller, J., de Ruyter, K., and Mahr, D. (2022). How to strategically choose or combine augmented and virtual reality for improved online experiential retailing. Psychology and Marketing, 39(3), 495 to 507. doi:10.1002/mar.21600 Irshad, and colleagues (2025). Augmented reality in the retail outlet: Shaping the retail shopping experience, a cross-cultural study. Journal of Consumer Behaviour. doi:10.1002/cb.2483 Javornik, A., Marder, B., Barhorst, J. B., McLean, G., Rogers, Y., Marshall, P., and Warlop, L. (2022). What lies behind the filter: Uncovering the motivations for using augmented reality face filters on social media and their effect on well-being. Computers in Human Behavior, 128, 107126. doi:10.1016/j.chb.2021.107126 Kerbl, B., Kopanas, G., Leimkuhler, T., and Drettakis, G. (2023). 3D Gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics, 42(4), 1 to 14. doi:10.1145/3592433 Kotler, P., Kartajaya, H., and Setiawan, I. (2021). Marketing 5.0: Technology for humanity. Hoboken: Wiley. Kotler, P., Kartajaya, H., and Setiawan, I. (2023). Marketing 6.0: The future is immersive. Hoboken: Wiley. Kowalczuk, P., Siepmann, C., and Adler, J. (2021). Cognitive, affective, and behavioral consumer responses to augmented reality in e-commerce: A comparative study. Journal of Business Research, 124, 357 to 373. doi:10.1016/j.jbusres.2020.10.050 Kumar, H. (2022). Augmented reality in online retailing: A systematic review and research agenda. International Journal of Retail and Distribution Management, 50(4), 537 to 559. doi:10.1108/IJRDM-06-2021-0287 Mildenhall, B., Srinivasan, P. P., Tancik, M., Barron, J. T., Ramamoorthi, R., and Ng, R. (2021). NeRF: Representing scenes as neural radiance fields for view synthesis. Communications of the ACM, 65(1), 99 to 106. doi:10.1145/3503250 Minkov, M., and Kaasa, A. (2022). Do dimensions of culture exist objectively: A validation of the revised Minkov-Hofstede model of culture with World Values Survey items and scores for 102 countries. Journal of International Management, 28(4), 100971. doi:10.1016/j.intman.2022.100971 Nikhashemi, S. R., Knight, H. H., Nusair, K., and Liat, C. B. (2021). Augmented reality in smart retailing: A symmetric approach to continuous intention to use retail brands mobile AR apps. Journal of Retailing and Consumer Services, 60, 102464. doi:10.1016/j.jretconser.2021.102464 Pharr, M., Jakob, W., and Humphreys, G. (2023). Physically based rendering: From theory to implementation, fourth edition. Cambridge: MIT Press. Rauschnabel, P. A., Babin, B. J., tom Dieck, M. C., Krey, N., and Jung, T. (2022). What is augmented reality marketing: Its definition, complexity, and future. Journal of Business Research, 142, 1140 to 1150. doi:10.1016/j.jbusres.2021.12.084 Rauschnabel, P. A., Felix, R., Hinsch, C., Shahab, H., and Alt, F. (2022). What is XR: Towards a framework for augmented and virtual reality. Computers in Human Behavior, 133, 107289. doi:10.1016/j.chb.2022.107289 Tan, Y. C., Chandukala, S. R., and Reddy, S. K. (2022). Augmented reality in retail and its impact on sales. Journal of Marketing, 86(1), 48 to 66. doi:10.1177/0022242921995449

Executive education has moved from the margins of the university to the centre of how professional knowledge is renewed. At the same time, a new class of institutions has emerged in the global economy: international financial centres such as the Dubai International Financial Centre, the Abu Dhabi Global Market, the Astana International Financial Centre and the Qatar Financial Centre, each attached to specialised courts and tribunals that apply common law or hybrid procedures to cross-border commercial disputes. These centres employ thousands of regulators, bankers, compliance officers, arbitrators, registrars and judges whose learning needs are unusual: highly technical, fast moving, and difficult for any single university to satisfy alone. This article asks how universities and business schools can build durable #strategic_partnerships with these #international_financial_centers and their associated #judicial_centers, and how #collaborative_training programs can be designed, governed, delivered and evaluated so that they serve academic, professional and public interests at the same time. The study uses a qualitative, conceptual and document-based design. It synthesises recent scholarship on #executive_education, #university_industry_collaboration, #micro_credentials and the rise of international commercial courts, and it reads that literature against publicly reported practice in financial centre academies and judicial training institutes. From this synthesis the article develops an integrated framework, called the Alignment, Architecture and Accountability framework, that treats a partnership as three linked layers: the strategic reasons partners come together, the operating structures that make joint work possible, and the evidence systems that show whether learning changed anything. Each layer is broken into practical components, including partner selection, #memorandum_of_understanding design, #curriculum_co_design, faculty models, credentialing, quality assurance and impact measurement. The findings suggest that the most fragile point in such partnerships is not curriculum content but governance. Programs fail when roles are vague, when intellectual property is unsettled, when the academic partner is treated as a vendor rather than a co-author of knowledge, and when nobody is accountable for evidence of learning. The article also identifies risks specific to the judicial side of the relationship, especially the need to protect #judicial_independence and to keep education free of any appearance of client capture. The article closes with a practical agenda for students, program managers and researchers, including a call for longitudinal studies that follow participants over several years rather than surveying them on the last day of a course. Keywords: executive education; strategic partnerships; international financial centres; judicial training; collaborative programs; micro-credentials; capacity building; transnational adjudication 1. Introduction 1.1 The problem in plain terms A senior compliance officer in a global financial centre may need to understand digital asset regulation, sanctions screening, climate risk disclosure and cross-border insolvency in the same working week. A judge sitting in an international commercial court may need to interpret an English law contract on Monday, apply a foreign arbitration award on Tuesday, and manage electronic disclosure on Wednesday. Neither of these people has time for a three year degree. Both of them need something short, precise, current and credible. Universities are, in theory, well placed to supply that. They hold deep expertise, a research base, an assessment culture and the authority to award recognised qualifications. In practice, many struggle. Their timetables are slow. Their curricula are approved years in advance. Their faculty are rewarded for publication, not for practitioner impact. Meanwhile financial centres and judicial bodies have built their own academies, sometimes drawing on universities, sometimes bypassing them entirely. This article is about the space in between. It argues that neither side can do this work well alone, and that strategic partnerships between academic institutions and the professional bodies of global finance and justice are not a nice extra but a structural necessity. It also argues that partnership is a craft with its own rules, and that most of what goes wrong is predictable and preventable. 1.2 Why this matters now Three developments make the question urgent. First, the professional half-life of technical knowledge in finance and commercial law has shortened. Rules on digital assets, artificial intelligence in credit decisions, sustainability reporting and cross-border data transfer change every year. What a practitioner learned five years ago may now be wrong, not merely incomplete. This creates a permanent demand for #lifelong_learning that degree programs were never built to meet. Second, the institutional landscape has changed. Over the past two decades a set of specialised international commercial courts has appeared in Europe, the Middle East and Asia. Scholars have documented how these courts use English as the language of proceedings, appoint judges from other jurisdictions, and adopt procedures borrowed from arbitration, all in order to serve cross-border business. Antonopoulou has shown that many of these courts sit inside special economic zones and form part of wider strategies to attract offshore corporate and financial services, and that they actively market their procedural advantages to prospective litigants. Brekoulakis and Dimitropoulos, in the leading collection on the subject, present these courts as a possible paradigm for the future of transnational adjudication. Wherever such a court exists, a judicial centre ecosystem grows around it: registrars, court technologists, expert witnesses, mediators and a local bar that must be trained to appear before it. Third, the credentialing system itself is in motion. Research on micro-credentials has expanded quickly, and reviews now describe a global landscape in which short, stackable, work-related certificates sit alongside traditional degrees. Varadarajan, Koh and Daniel found that learners want short, practical and current courses, institutions want accreditation in order to build trust, employers want clarity about what competencies a credential actually certifies, and governments want employability. These four expectations rarely align on their own. Partnership is one of the few mechanisms that can align them. 1.3 Research questions The article addresses four questions: What strategic logics explain why universities, international financial centres and judicial institutions enter into collaborative training relationships, and what does each side actually want? What governance structures, contractual arrangements and operating models make such partnerships work, and which ones cause them to fail? How should collaborative training programs be designed, delivered, credentialed and assessed in this specific context, where the learners are experienced professionals and the subject matter is technical and contested? How can partners measure impact in a way that is honest, proportionate and useful, rather than ceremonial? 1.4 Contribution The article makes three contributions. Conceptually, it brings together three literatures that rarely speak to each other: the management literature on university and industry collaboration, the education literature on executive education and micro-credentials, and the legal literature on international commercial courts. Practically, it offers the Alignment, Architecture and Accountability framework, which program designers can use as a checklist. Pedagogically, it is written for students who may one day work inside these partnerships, either as program managers, as regulators, as court administrators or as academics, and it therefore explains its reasoning rather than assuming prior knowledge. 1.5 Structure Section 2 reviews the relevant literature. Section 3 sets out the theoretical framework. Section 4 explains the method. Sections 5 to 8 present the analysis: the strategic logic of partnership, the architecture of governance, the design of programs, and the accountability layer. Section 9 discusses risks and tensions. Section 10 draws out implications. Section 11 states limitations and a research agenda. Section 12 concludes. 2. Literature Review 2.1 Executive education as a field executive education refers to non-degree or short-form learning aimed at working managers, professionals and leaders. It has a long institutional history and a strong social function. Sorsa, Nevalainen, Korhonen and Kansikas, studying the Finnish Institute of Management across five decades, show that executive education operates not only as instruction but as a gateway into a business elite, working through four connected mechanisms: honouring, educating, networking and socialising. This is an important corrective to a purely technical reading of the field. When a financial centre sends its senior staff to a joint program with a leading law school, part of what it buys is content, and part of what it buys is #legitimacy, #reputation and access to a network. Any honest analysis of strategic partnerships must acknowledge both. The market side of the field has also shifted. Reporting by the Chartered Association of Business Schools on the impact of executive education, together with sector commentary, points consistently in one direction: away from open-enrolment short courses sold as products, and towards custom, co-created learning journeys tied to a client organisation's strategy. Custom programs are the fastest-growing segment. This shift is precisely what makes partnership, rather than transaction, the relevant unit of analysis. A co-created program cannot be bought off a shelf. It has to be built with someone. Two persistent weaknesses in the field deserve attention. The first is #evaluation. Much of what passes for evidence in executive education is participant satisfaction collected at the end of a course. The second is the relevance gap. Business schools are frequently criticised for producing research that practitioners do not read and teaching that does not survive contact with practice. Both weaknesses are directly relevant to the argument here, because a partnership with a financial or judicial institution exposes them immediately. A regulator will notice within one session if the faculty do not understand the applicable rulebook. 2.2 University and industry collaboration The management literature on university and industry collaboration is large and mostly concerned with research and technology transfer rather than teaching. Even so, its findings transfer well. Rossoni, de Vasconcellos and de Castilho Rossoni, in a systematic review of barriers and facilitators, describe collaboration as a relationship in which both parties benefit from the transfer of knowledge, expertise or technology, and in which research findings are applied to practical problems. The recurring facilitators they and others identify are #trust, communication, shared goals, the use of intermediaries, and prior experience of working together. The recurring barriers are differences in institutional culture, differences in time horizons, disagreements over #intellectual_property, and fear of #knowledge_transfer leaking value away. Two ideas from this literature are especially useful. The first is the evolutionary view: collaborations pass through phases, often described as embryonic, initiation, engagement and established, and different barriers dominate at different phases. Early on, the problem is lack of trust and fear of knowledge leakage. Later, the problem is coordination cost and the risk that the collaboration becomes routine and stops producing anything new. The second is #absorptive_capacity: an organisation can only use knowledge it is equipped to receive. A financial centre with no internal learning function will not absorb what a university offers, no matter how good the teaching is. Romero-Sanchez, Perdomo-Charry and Burbano-Vallejo, reviewing collaboration and spin-off creation, place absorptive capacity and social capital at the centre of their account, and the same logic applies to training. The literature also warns against a common mistake. Universities often treat collaboration as a revenue line. Firms often treat it as procurement. Both framings destroy the thing they are trying to buy, because the value of a genuine partnership lies in joint problem definition, and neither a revenue target nor a purchase order defines a problem. 2.3 Micro-credentials and the changing shape of professional learning The micro-credentials literature has grown rapidly and gives useful vocabulary for the design questions in this article. Ahsan, Akbar, Kam and Abdulrahman, in a systematic review of implementation in higher education, describe micro-credentials as an innovative and disruptive learning model and note that further work is needed to understand how technology and different stakeholder perspectives shape uptake. Brown, Nic Giolla Mhichil, Beirne and Mac Lochlainn map a global credential ecology for lifelong learning. Brown, McGreal and Peters set out the key strategic questions institutional leaders should answer before committing. Selvaratnam and Sankey review implementation in Australasia and draw out practical lessons. Tamoliune and colleagues examine the potential of micro-credentials in the post-pandemic recovery. Three findings from this body of work matter here. First, #quality_assurance is the weak point. Brown and Duart, mapping quality assurance practice globally, find real gaps. If a credential is to mean anything to an employer in a financial centre, someone must stand behind the assessment. That is where an accredited university partner adds value that a private training provider often cannot. Second, employers want clarity about competencies, not marketing language. Bruguera, Pages, Peters and Fito, studying employer perspectives on micro-credentials and soft skills in online education, find that employers value evidence of what a learner can now do. This pushes program design towards an explicit #competency_framework. Third, the regional evidence is relevant. Alasmari and Alzahrani examine micro-credential policy and practice in universities across the Middle East and North Africa in relation to employment, a region that contains several of the financial and judicial centres discussed in this article. Their work suggests that policy ambition often runs ahead of institutional capacity, which is another argument for partnership rather than solo effort. 2.4 International financial centres and their courts An international financial centre is a jurisdiction, or a zone within a jurisdiction, that concentrates financial services activity and supports it with tailored law, regulation and infrastructure. Several of the newer centres are built as special economic zones with their own legal frameworks and their own courts. The Dubai International Financial Centre is a leading example: it publicly reports a workforce in the tens of thousands, more than a thousand regulated firms, and a large and growing population of firms working in financial technology and innovation. It also runs its own academy, and reports that thousands of learners complete programs there each year, describing the cultivation of talent as a decisive indicator of economic development and competitiveness. This is not a marginal training operation. It is a #talent_pipeline strategy. The legal scholarship on the courts attached to such centres is now well developed. Brekoulakis and Dimitropoulos frame international commercial courts as institutions that attempt to bridge concerns about diminished sovereignty with the need to globalise #dispute_resolution. Antonopoulou describes a process she calls the arbitralization of courts, in which courts absorb popular features of #arbitration such as party autonomy over procedure, confidentiality and the appointment of specialist decision makers, in order to compete for cases. In later work she introduces the idea of forum marketing and argues that these courts do not merely make themselves attractive through good procedure but actively promote themselves, a practice she treats as normatively significant rather than neutral. Schultz and Bachmann, writing in the same collection, warn of possible problematic social externalities of treating dispute resolution as a product with market potential. Van Calster examines how competition between fora intensified in Europe after Brexit. Bantekas provides a comparative account of the courts themselves. Two consequences follow for education. The first is that these courts create genuinely new training needs. Judges may come from several jurisdictions and must be inducted into a hybrid procedural culture. Local lawyers must learn to litigate in English under common law rules that may differ from the surrounding civil law system. Court staff must run electronic case management. Arbitrators and mediators must understand where court supervision begins and ends. The second consequence is a caution. If courts are competing for business, then education can slide into promotion. A joint program that quietly functions as marketing for a court, or that trains lawyers to prefer one forum over another, has crossed a line. The scholarship on forum marketing gives us the language to name that risk, and any partnership in this space needs an explicit answer to it. 2.5 Judicial education as a distinct tradition #judicial_education is not simply corporate training applied to judges. It carries constitutional weight. Because judges must be independent, the design of their training is constrained in ways that management education is not. Who chooses the curriculum, who teaches, who evaluates, and who pays are all questions with implications for judicial independence. International practice reflects this. The International Organization for Judicial Training exists to support judicial education institutions worldwide and convenes regular conferences at which judicial educators exchange approaches to curriculum, faculty development and methodology, and it publishes a journal on judicial education and training. National institutions such as the United States Federal Judicial Center conduct research into court operations and work with judiciaries in other countries on judicial education and administration, usually in partnership with other organisations. Canada's National Judicial Institute frames judicial education as a companion to sustainable judicial reform and focuses its international work on building the capacity of judicial education institutions themselves rather than delivering one-off courses. Regional structures also exist, for example the East African Community's judicial education committee, in which national academies cooperate on training. In Europe, judicial training networks have published guidance on how to evaluate judicial training practice, including work that examines the use of established training evaluation models in judicial settings. Several design principles recur across these traditions and are directly relevant to partnership design. Judicial education should be led by judges. It should respect the experience of participants and treat them as knowledgeable professionals rather than students. It should use interactive methods, including discussion, role play and problem solving, rather than lecture alone. And it should be planned as a coherent curriculum rather than assembled from whatever speakers are available. A university partner that ignores these principles will be tolerated once and not invited back. 2.6 The gap Reading these literatures together reveals a gap. The executive education literature knows a great deal about corporate clients and almost nothing about courts. The judicial education literature knows a great deal about judges and treats universities mainly as suppliers of guest speakers. The university and industry collaboration literature is focused on patents and spin-offs, not on teaching. The micro-credential literature is focused on the credential, not on the institutional relationship that produces it. Meanwhile the legal literature on international commercial courts has described the institutions with care but has not asked how the people who staff them are trained. This article sits in that gap. It treats the financial centre, the court and the university as three partners in a single learning system, and it asks what it takes to make that system work. 3. Theoretical Framework 3.1 Why theory is needed here It is tempting to treat partnership design as a purely practical matter, a question of good contracts and good intentions. That is a mistake. Without theory, program designers cannot predict where a relationship will break, and they cannot explain to their own institutions why a partnership is worth the effort. Four theoretical lenses are used in this article. Each answers a different question. 3.2 Resource dependence and the resource-based view The #resource_based_view holds that an organisation gains advantage from resources that are valuable, rare, hard to imitate and hard to substitute. Applied here, it explains what each partner brings and what each partner lacks. A university brings research depth, faculty who understand how knowledge is produced and contested, an assessment infrastructure, and the authority to award credentials that carry weight beyond the local market. It usually lacks live regulatory data, access to practitioners, and the credibility to speak about how a rulebook is actually applied on a Tuesday morning. An international financial centres authority brings that live context. It has the rulebook, the supervisory experience, the enforcement history, the market participants and the problems. It usually lacks pedagogical capacity, independent assessment, and the ability to award a credential that means anything outside its own zone. A judicial centre brings authority over procedure, real cases, and access to the bench. It lacks, and often should lack, a commercial incentive to build a training business. Resource dependence theory adds a warning to this happy picture. Dependence is not symmetric. Whoever controls the scarcer resource controls the relationship. If a university is financially dependent on a single centre for a large share of its executive education revenue, its academic independence is at risk, and it will find it hard to say no to a curriculum request it disagrees with. Partnership design must therefore include deliberate limits on dependence. 3.3 Stakeholder theory #stakeholder_theory asks whose interests a program serves. In a two-party commercial training deal, the answer is simple: the buyer. In the setting of this article, it is not simple at all. The stakeholders include the sponsoring centre, the university, the individual learners, the employing firms, the regulator, the court, the wider legal profession, the domestic legal system in which the centre sits, and the public whose access to justice may be affected by how commercial dispute resolution develops. These interests do not automatically align. A centre may want training that promotes its jurisdiction. A regulator may want training that makes supervision easier. Learners may want a credential that helps them leave for a better job. The public may want a justice system that does not devote its best resources to the largest litigants. Stakeholder theory does not resolve these tensions. It requires that they be named, and it insists that a program which serves one stakeholder at the direct expense of another needs a justification. This is why the framework developed below includes an explicit public interest test. 3.4 Absorptive capacity and organisational learning absorptive capacity describes an organisation's ability to recognise the value of new external knowledge, take it in, and apply it. It is a strong predictor of whether collaboration produces anything. This concept reframes a common complaint. When a partner organisation says a program "did not work", the cause is frequently not the teaching. It is that the organisation had no mechanism for participants to apply what they learned. Nobody asked them to. No process changed. No manager followed up. The knowledge arrived and had nowhere to go. The practical implication is that partnership design must reach beyond the classroom into the partner's own systems. This means agreeing in advance on how learning will be applied, who will sponsor the application, and what will change as a result. It also means investing in the partner's internal learning function, not only in the course. 3.5 Institutional theory and legitimacy Institutional theory explains behaviour that a purely economic reading cannot. Organisations do things because those things are seen as proper, not only because they are profitable. Much of what happens in this space is legitimacy-seeking. A young financial centre partners with an old university to borrow its standing. A university partners with a prominent centre to demonstrate relevance and global reach to its own accreditors and rankings. A court runs a judicial exchange partly to signal that it belongs to the community of serious courts. None of this is illegitimate in itself. Sorsa and colleagues remind us that honouring and socialising have always been part of what executive education does. The danger comes when legitimacy work replaces learning work: when the photograph of the signing ceremony is the deliverable, and the program that follows is thin. The memorandum of understanding that is signed and never implemented is the characteristic failure mode of this sector, and institutional theory tells us exactly why it happens. 3.6 The integrated framework: Alignment, Architecture, Accountability Bringing these lenses together, the article proposes a three-layer framework. Layer one: Alignment. Why are we doing this, and do our reasons fit together? This layer covers strategic intent, stakeholder mapping, the public interest test, and an honest statement of what each partner wants. It draws on stakeholder theory and the resource-based view. Layer two: Architecture. How will we actually work together? This layer covers legal instruments, governance bodies, decision rights, faculty models, curriculum processes, intellectual property, funding, and the operational detail of delivery. It draws on the collaboration literature and on institutional design. Layer three: Accountability. How will we know if it worked, and who answers for it? This layer covers learning outcomes, assessment, quality assurance, impact measurement, transparency and review. It draws on absorptive capacity and on the evaluation literature. The layers are ordered deliberately. Architecture without alignment produces busy structures that serve no agreed purpose. Accountability without architecture produces measurement of things nobody is responsible for. Sections 5, 6, 7 and 8 develop each layer in turn. 4. Methodology 4.1 Research design This is a conceptual and qualitative study. It does not present new primary data. It synthesises three bodies of peer-reviewed literature with publicly available institutional documentation, and it uses that synthesis to build a framework. This design is appropriate when the aim is to connect fields that have developed separately and to produce a usable structure for practice, rather than to test a hypothesis. Readers should understand what this means for the strength of the claims. A conceptual synthesis can identify mechanisms, name risks and organise practice. It cannot establish causal effect. Where this article says that a design choice tends to produce a certain result, it is drawing an inference from converging evidence and reasoned argument, not reporting a measured effect. 4.2 Sources and selection Three source types were used. The first is peer-reviewed scholarship, drawn from management, education and law. Priority was given to work published within the last five years, and to systematic reviews, because reviews summarise a field more reliably than any single study. Searches covered executive education, university and industry collaboration, micro-credentials and lifelong learning, and international commercial courts. The second is institutional documentation in the public domain: published reports from financial centres about their own academies and workforce, published descriptions by judicial training institutions of their mandates and international cooperation, and published guidance from judicial training networks on curriculum and evaluation. The third is sector reporting and professional commentary on the executive education market, used only to establish broad direction of travel, such as the shift from open enrolment to custom co-created programs, and never as evidence of effect. 4.3 Analytical procedure The analysis proceeded in four steps. First, each literature was read separately and its main claims extracted. Second, claims were grouped thematically, which produced the alignment, architecture and accountability structure. Third, the institutional documentation was read against those themes to test whether the categories fitted observed practice and to surface issues the literature had missed, notably the tension between judicial independence and sponsor influence. Fourth, the framework was written up and stress-tested against known failure modes reported in the collaboration literature. 4.4 Limitations of the method Four limitations should be stated plainly. The reliance on public documentation favours partnerships that succeeded and were publicised. Failed partnerships are rarely written up, which introduces a survivorship bias that this article cannot fully correct. The evidence base is uneven across regions. There is more published material on courts in the Gulf, Singapore and Europe than on comparable institutions elsewhere, which risks a partial picture. The framework is normative in places. It says what partners should do. Those statements rest on reasoned argument and on the values embedded in judicial independence and academic freedom, not on measured outcomes. Finally, this is a fast-moving field. Regulation, technology and institutional design in both finance and dispute resolution change quickly, and any account of practice will date. The framework is intended to outlast the examples, but readers should verify current practice before relying on any specific description. 5. Alignment: The Strategic Logic of Partnership 5.1 What each partner actually wants Partnership begins with honesty about motive. The following account of interests is deliberately blunt, because polite versions of these statements are what produce vague agreements. What the university wants. Revenue that is less volatile than degree tuition. Evidence of relevance and #thought_leadership for accreditation bodies and rankings. Access to data and practitioners for research. A pipeline of executive learners who may later become degree students, donors or employers of graduates. Institutional prestige from association with a recognised centre. Some universities will also say, sincerely, that they want to contribute to the #rule_of_law and to sound financial regulation, and that motive should be taken seriously rather than dismissed. What the financial centre wants. A talent pipeline. #regulatory_literacy across the firms it supervises, because a supervisor's job is easier when the supervised understand the rules. Credible credentials that help it attract and retain #human_capital. Reputational association with academic quality. Support for national policy goals, which in several jurisdictions include #economic_diversification and the development of a domestic professional workforce. Financial centres are explicit that developing talent is a competitiveness indicator, not a welfare programme. What the judicial centre wants. Competence on the bench and in the registry. A local profession capable of appearing before it. Consistency of practice. Alignment with international standards. Professional development for judges that respects their standing. Judicial centres generally do not want revenue from training, and they are usually right not to want it. What the learners want. Something that helps them do their current job and improve their next one. Recognition that travels. Time efficiency. Peers. What the public interest requires. Competent regulators, honest courts, and a justice system that does not become the private property of the largest commercial users. 5.2 The alignment test Before any agreement is drafted, partners should be able to answer five questions in writing. This is the alignment test. First, what problem are we solving that neither of us can solve alone? If either partner could do it alone, the partnership is decorative. Second, what is each partner contributing that the other cannot buy on the open market? If the university is contributing only teaching hours, a commercial training provider will do it cheaper. If the centre is contributing only money, it is a client, not a partner. Third, whose interests could this program damage? This is the #public_private_partnership question that is most often skipped. A program that teaches lawyers to prefer one forum, or that trains supervisors in a manner shaped by the firms they supervise, damages someone. Fourth, what would make us walk away? Naming the exit conditions at the start is the single most effective protection against slow capture. Fifth, how long are we committing for, and what happens to the work if a key individual leaves? Many partnerships are in reality relationships between two enthusiastic people. When one of them moves on, the partnership dies. Institutionalising the relationship is the antidote. 5.3 Partner selection Not every well-known institution is a good partner. Selection criteria that matter more than brand include the following. Relevant faculty depth, not merely faculty count. One professor who understands cross-border insolvency is worth more than a large department that does not. Willingness to co-design. A university that insists on delivering an existing course with the client's logo on the cover is not a partner. Institutional patience. Partnerships in this space take years to reach maturity, and the collaboration literature is consistent that trust builds through phases. Capacity to assess. If the credential is to be worth anything, the university must be able and willing to fail people. This sounds obvious and is regularly the first thing that gets negotiated away. Alignment of values, especially on #ethics, judicial independence, academic freedom and the right to publish. These should be tested before signature, not after the first disagreement. 5.4 The public interest test Because this article deals with institutions that carry public functions, it proposes an explicit public interest test that should be applied to every proposed program. Does the program strengthen capability that the public benefits from, such as sound supervision, competent adjudication and honest professional practice? Does it avoid conferring an unfair advantage on any particular firm, forum or litigant? Would the partners be comfortable if the full curriculum, the funding arrangement and the faculty list were published? Is the academic partner free to teach conclusions the sponsor dislikes, and to publish research that reflects badly on the sponsor? Does the program avoid creating or appearing to create a relationship of obligation between a judge and any party who may appear before that judge? A program that fails any of these questions should be redesigned or abandoned. In practice, the last question is the one that most often needs work, and it is the subject of Section 9. 6. Architecture: Governance and Operating Models 6.1 The failure mode to avoid The characteristic failure of collaborative training is not a bad lecture. It is a signed memorandum of understanding that nobody implements. The document is celebrated, photographed, filed, and forgotten. Six months later each side believes the other was supposed to start. This happens because the agreement described intent rather than architecture. Intent is easy to agree on. Architecture requires decisions that someone will dislike. 6.2 Legal instruments Three instruments are usually needed, and they do different work. The framework agreement establishes the relationship. It sets out purpose, duration, governance bodies, principles including academic freedom and independence, dispute resolution, and termination. It should be short. Its most important clauses are the ones about what happens when things go wrong. The program schedule governs a specific program. It states learning outcomes, contact hours, assessment method, credential awarded, faculty, cohort size, fees, cost recovery, and delivery dates. There should be one schedule per program, and it should be signed by people who have the authority to deliver, not only by people with the authority to sign. The intellectual property and data annex deals with the two issues that most often destroy trust. Who owns the case studies written from the centre's material? Who may publish? What data may the university researcher see, and under what conditions of #confidentiality? What happens to teaching materials when the partnership ends? The collaboration literature identifies fear of knowledge leakage as a major early barrier, and the way to reduce that fear is to settle these questions in writing before anyone is invested. A reasonable default is a split. The centre owns its confidential material and any anonymised data set remains under its control. The university owns its pedagogy, its frameworks and its research outputs. Jointly created materials, such as a teaching case built from a real supervisory file, are jointly owned with a licence for each side to use them for non-competing purposes, and the case is published only with the centre's clearance on facts and the university's clearance on analysis. 6.3 Governance bodies Three bodies, with different memberships and different jobs, work better than one committee that does everything. A steering group at senior level meets two or three times a year. It owns the relationship, approves the program portfolio, resolves disputes and reviews performance. It is small. Its members must be senior enough to make commitments. A program board or joint academic committee meets more often and owns the content. Its membership must be genuinely mixed: academics, practitioners from the centre, and, where judicial training is involved, judges. Its job is curriculum co-design, faculty approval, assessment standards, and quality review. This is the body where the real work happens. An operations group handles logistics, admissions, scheduling, platforms and participant support. It is unglamorous and it is where most partnerships actually fail, because nobody senior pays attention to it. Decision rights should be written down. A simple table stating who proposes, who consults, who decides and who is informed for each category of decision will prevent more conflict than any amount of goodwill. The categories that most need clarity are curriculum content, faculty selection, assessment standards, admission of participants, use of the partners' names and marks, and publication. 6.4 Funding models Four models are common. Each has consequences. Fee for service. The centre pays the university a fee per program. Simple, and it makes the university a supplier. Risk: the university's academic independence weakens as the fee grows in importance. Mitigation: cap the share of the university's executive education revenue that any single partner may represent. Cost sharing. Each partner bears its own costs and shares the surplus, if any. This aligns incentives better and produces more genuine #cost_sharing, but it requires transparent accounting and is administratively harder. Sponsored capacity building. A third party, such as a development agency or a public authority, funds a program that neither partner would fund alone. Common in judicial education, where courts should not be selling training and universities cannot afford to give it away. Risk: the funder's priorities quietly become the curriculum. Mitigation: a written academic charter that the funder accepts. Participant-paid. Individuals or their firms pay. Works for open programs and for micro-credentials with clear labour-market value. Risk: content drifts towards what sells rather than what is needed, and the least profitable but most important subjects, such as ethics, get squeezed. Most mature partnerships blend these. What matters is that the model is explicit and that its distorting effects are consciously counterbalanced. 6.5 Faculty models Who teaches is the most visible signal of whether a partnership is real. A guest speaker model, in which practitioners give occasional talks inside a university course, is the weakest form. It adds colour, not capability. A co-teaching model, in which an academic and a practitioner design and deliver a module together, is much stronger. Each is present when the other teaches. Disagreements between them are surfaced in class rather than hidden, and that disagreement is itself instructive for experienced learners. A practitioner-faculty model appoints senior professionals as adjunct or professorial fellows with real academic obligations, including assessment and curriculum work. This builds durable capacity but requires the university to invest in teaching development for people who have never taught. A judicial faculty model is required for judicial programs. The consistent principle across judicial training traditions is that judges should lead the education of judges, with academics supporting. This is not professional protectiveness. It reflects the reality that a large part of judicial competence is judgecraft, which can only be transmitted by people who have exercised it, and it protects independence by keeping curricular control inside the judiciary. #faculty_exchange in both directions, where academics observe court practice and judges spend time in the university, deepens this over time. 6.6 Delivery models The delivery decision should follow the learning outcome, not the technology budget. Residential intensive blocks work well for #experiential_learning, negotiation, judgment and network formation. They are expensive and hard to schedule for senior people, but nothing else builds cohort trust as fast. Blended and hybrid designs, combining asynchronous preparation with live sessions and a short in-person capstone, are now the default for most professional programs. They preserve some of the intensity of the residential model while accommodating working schedules across #time_zones. Fully online delivery scales and reaches practitioners in smaller markets, which matters for regional #inclusion. It is weaker for judgment and networking unless deliberately designed with small-group live work. Work-based and action learning designs, in which participants bring a live problem from their own institution and work on it through the program, are the strongest option for absorptive capacity, because the application is built into the design rather than hoped for afterwards. A practical rule: use asynchronous #digital_delivery for knowledge that can be read, and reserve scarce live time for the things that only happen when people are in a room, which are argument, judgment, feedback and relationship. 6.7 Credentialing architecture Credentials should be designed as a system, not issued one at a time. A useful structure has three tiers. A short foundational tier certifies knowledge, is assessed by examination or structured task, and can be delivered at scale. A middle tier certifies applied competence, is assessed through a work-based project, and requires an assessor who knows the field. A senior tier, which may lead towards a formal postgraduate qualification, certifies the ability to lead and to reason under uncertainty, and is assessed through substantial written and oral work. Tiers should stack. A participant who completes several foundational credentials should be able to carry that work forward into the next tier. The micro-credential literature is clear that stackability is what turns a certificate into a pathway, and that #accreditation by a recognised body is what makes an employer trust it. Two design rules deserve emphasis. First, the credential must name the competency, not the course. "Applies risk-based customer due diligence to complex ownership structures" tells an employer something. "Attended the Advanced Compliance Programme" does not. Second, someone must be able to fail. A credential that everyone receives certifies attendance, and everyone in the market knows it. 7. Program Design: Content, Pedagogy and the Curriculum 7.1 Starting from needs, not from strengths The most common design error is that the university offers what it already teaches. A serious partnership starts with a #training_needs_analysis conducted jointly. A good needs analysis in this setting looks at four levels. At the level of the individual, what tasks do people struggle with, and what errors recur? At the level of the role, what will this role require in three years that it does not require now? At the level of the institution, what capability gap is holding back a strategic objective? At the level of the system, what weakness would matter to the public, such as inconsistent supervision or slow case management? The fourth level is routinely skipped and is the one that justifies the involvement of a university at all. 7.2 A curriculum map for financial centre programs The following is offered as a map, not a syllabus. Local law always governs. Core regulatory foundations. The architecture of #financial_regulation, the role of the supervisor, licensing, conduct and prudential rules, supervisory tools and enforcement. This is the base on which everything else rests, and it is often assumed rather than taught. Financial crime. #anti_money_laundering, counter-terrorist financing, beneficial ownership, sanctions screening and #sanctions_compliance. This area changes constantly and is where supervisory expectations most often outrun industry practice. Corporate and market conduct. #corporate_governance, directors' duties, disclosure, market abuse, and the treatment of clients. Technology and data. #fintech business models, digital assets and tokenisation, #artificial_intelligence in credit and surveillance, algorithmic accountability, cyber risk, #data_protection and cross-border data transfer. Sustainable and specialised finance. #green_finance, sustainability disclosure and #esg reporting; and, where relevant to the jurisdiction, #islamic_finance, including the interaction between contractual structures and regulatory treatment. Cross-border legal foundations. Choice of law, jurisdiction, recognition and enforcement, cross-border #insolvency, and the interaction of financial contracts with dispute resolution clauses. Professional judgment and ethics. #conflict_of_interest, whistleblowing, speaking up, and the specific ethical pressures of a regulated environment. 7.3 A curriculum map for judicial and dispute resolution programs Commercial law foundations for cross-border adjudication. Contract interpretation, agency, security interests, and the treatment of financial instruments. Procedure in international commercial courts. Case management, disclosure and evidence, expert evidence, interim relief, costs, and the use of English as the language of proceedings where relevant. The hybrid nature of these courts, documented in the scholarship on the arbitralization of courts, means that procedural training cannot simply be imported from either tradition. The arbitration interface. The supervisory role of courts, challenges to awards, enforcement, anti-suit relief, and the boundary between judicial and arbitral competence. #mediation and settlement facilitation belong here too. Court technology. Electronic filing, remote hearings, #digital_courts, case management systems, and the emerging and contested question of artificial intelligence in judicial decision support. This last topic requires unusual care, because the technology is advancing faster than the professional consensus about it. Judgment writing and reasoning. Often the most valued module, and one that only experienced judges can teach well. Judicial ethics and independence. Recusal, the appearance of bias, relations with the executive and with sponsors, and the specific pressures created when a court sits inside a commercially motivated zone. Court administration. Registry practice, listing, delay management, and user service. 7.4 Pedagogy for experienced professionals Senior professionals do not learn by being told. Four methods do most of the work. The #case_method, in which participants analyse a real situation with incomplete information and defend a decision, remains the strongest tool for developing judgment, provided the cases are local and current. Generic cases from another market signal that the partnership is not real. #simulation and #moot_court exercises let participants practise under pressure. A simulated enforcement panel, a mock case management conference, a mock supervisory interview, or a moot court hearing before retired judges will teach more in three hours than a week of lectures. #action_learning sets, in which small groups work on a real problem from a participant's own institution over the life of the program and report back, are the strongest lever for organisational impact. They also generate research material for the university, which is a legitimate and often overlooked benefit. Structured peer exchange is the method most often wasted. Participants in these programs are frequently the best available teachers of one another, and a well-facilitated session in which a supervisor from one authority explains to peers how a difficult case was handled is more valuable than an outside expert. #communities_of_practice that continue after the program formalise this, and #alumni_networks give them a home. #mentoring, pairing junior participants with senior practitioners, is inexpensive and consistently underused. 7.5 Designing for transfer Learning that is not applied is a cost, not an investment. Three design features improve transfer. A pre-program commitment, in which each participant and their line manager agree in writing what will be different after the program, converts vague attendance into a contract. A work-based project, assessed as part of the credential, forces application during the program rather than after it. A structured follow-up, usually a session ninety days later at which participants report what actually changed, does more for #impact_measurement than any end-of-course questionnaire, and it signals to participants that the organisation is serious. 7.6 Localisation and language Programs in this field frequently operate across #cultural_distance and across #language_barriers. Two practical points matter. Legal and regulatory content does not translate cleanly. A concept such as fiduciary duty or good faith carries different weight in different traditions, and a program that pretends otherwise will produce confident error. Where a course is taught in English inside a jurisdiction whose general law is not common law, the design should teach the difference explicitly rather than assume it away. Faculty from outside the jurisdiction must be briefed, not merely booked. A visiting professor who does not know the local rulebook will be politely destroyed by a room of supervisors, and the partnership will suffer for it. 8. Accountability: Quality, Evidence and Impact 8.1 The ceremonial evaluation problem Most executive education is evaluated by asking participants at the end whether they enjoyed it. The results are almost always positive, and they predict very little. This is ceremonial evaluation: it produces a number that can be reported, and it changes nothing. The reason it survives is that it is cheap and safe. Real evaluation is expensive and risky, because it can show that a program did not work. A partnership that is serious about #accountability must decide, at the start, that it is willing to learn bad news. 8.2 A layered evidence model A workable model asks five questions in sequence, each harder and more valuable than the last. Did they attend and engage? Completion rates, participation in discussion, submission of work. Cheap, necessary, and meaningless on its own. Did they learn? Assessed against explicit #learning_outcomes, using an assessment that a competent person could fail. This is the minimum standard for any credential. Can they do it? Demonstrated in a work-based task, a simulation, or an observed performance. This is where a competency framework earns its keep. Did behaviour change at work? Established through follow-up with participants and their managers some months later, and through evidence such as changed case handling, improved supervisory reports, or faster judgment turnaround. Did the institution change? The hardest and most important question. Evidence might include a reduced backlog, fewer supervisory findings of the same type, higher user satisfaction with a court, or improved consistency of decisions. Not every program needs all five. A one-day briefing does not warrant an impact study. But a multi-year partnership that never gets past the second question is not being accountable; it is being comfortable. 8.3 Return on investment, honestly Partners often want a figure for #return_on_investment. It can sometimes be estimated, but it should be done carefully. Where a program plausibly reduces a measurable cost, such as time to resolve a case, remediation following supervisory findings, or staff turnover in a hard-to-fill role, a reasoned estimate is possible and useful. Where it does not, an invented number is worse than no number. Judicial education in particular resists monetisation, and it is more honest to argue its value in terms of the quality and consistency of justice than to construct a spurious financial case. The safest practice is to state assumptions openly, present a range rather than a point estimate, and never claim causal attribution from correlation. Many things change in an institution in a year, and a training program is rarely the only one. 8.4 Quality assurance quality assurance in a joint program has to satisfy two systems at once: the university's academic regulations and the professional standards of the partner. These do not automatically fit. Four mechanisms make it work. Joint approval of learning outcomes before a program is announced, not after. Independent moderation of assessment, with an external examiner who is not employed by either partner. Periodic review of the program by the program board against evidence, with the power to close a program that is not working. And transparent publication of what the credential certifies, so that the labour market can judge it. The micro-credential literature identifies gaps in quality assurance as one of the field's central weaknesses. A university partner that brings rigorous assessment is offering something genuinely scarce, and it should not trade that away under commercial pressure. 8.5 Research as a partnership output A partnership that produces only teaching is under-using itself. The same relationship can produce research: on supervisory effectiveness, on court user experience, on how professionals actually make judgments under uncertainty, on whether a credential changes career outcomes. This requires the #research_collaboration to be built into the agreement from the start, with an agreed approach to data access, ethics approval, anonymisation and the right to publish. It also requires the sponsor to accept that findings may be uncomfortable. A sponsor that will only permit flattering research should not be offered a research relationship at all, and a university that accepts such a condition has stopped being a university. Done well, this is where #benchmarking becomes possible: comparing practice across centres, learning what works, and contributing evidence to a field that currently runs largely on assertion. 9. Risks, Tensions and Ethical Boundaries 9.1 The independence problem This is the central ethical issue of the article and it deserves to be stated without hedging. International commercial courts operate in a competitive environment. The scholarship on forum marketing shows that they promote themselves in order to attract cases, and that this promotion is not neutral: it disseminates a particular pro-business conception of civil justice. Meanwhile, the firms and law firms that appear before those courts are often the same organisations whose staff attend, and sometimes fund, executive programs run in partnership with the court's own jurisdiction. Three concrete dangers follow. The first is the appearance of obligation. If a law firm sponsors a judicial education program, and a judge who attended it later decides a case in which that firm appears, the appearance of bias is obvious, whatever the reality. The solution is structural: judicial education should not be funded by potential litigants or their advisers, and any funding should be pooled, disclosed and controlled by the judiciary or an independent body. The second is curricular capture. A curriculum written to please a sponsor will teach the sponsor's view of the law. In judicial training this is intolerable. The protection is that judges control the judicial curriculum, that academic partners retain the right to teach contested positions as contested, and that the curriculum is published. The third is promotional drift, in which the education function slowly becomes a marketing function. The test is simple: would this program still make sense if it did not attract a single additional case or firm to the jurisdiction? If not, it is marketing. 9.2 Academic freedom and commercial pressure The mirror image of the independence problem faces the university. A large, prestigious partnership generates revenue, visibility and internal political capital. Over time it becomes something the university cannot afford to lose. At that point the partner's preferences begin to shape what is taught and what is researched, usually without anyone deciding that they should. The defences are dull and effective: a cap on revenue concentration, a written charter of academic freedom, a right to publish that is not subject to sponsor veto on findings, and a governance structure in which the academic body, not the business development office, controls content. 9.3 Equity and access Executive education in global financial centres is expensive, and it is easy for these partnerships to serve only those who already have advantages. Sorsa and colleagues' account of executive education as a gateway into a business elite is a reminder that exclusion is not an accident of these systems but part of how they have historically worked. If a partnership claims a public interest justification, it should act on it. Practical measures include funded places for regulators from smaller jurisdictions, open-access foundational material, delivery formats that do not require travel, and deliberate attention to participation from the #global_south and from professionals outside the largest firms. This is not charity. A financial centre with an under-trained regional periphery imports risk from it. 9.4 Sustainability of the partnership Many partnerships depend on one or two individuals. When they leave, the relationship collapses. Institutionalisation is the remedy: written processes, more than one relationship-holder on each side, a documented curriculum that does not live in one person's head, and a governance calendar that continues regardless of who is in post. Financial #sustainability requires a realistic view of cost. Custom programs are expensive to build. A partnership that assumes the first program will be profitable will be disappointed and may be abandoned before it matures, which the collaboration literature suggests takes years. 9.5 Data, confidentiality and legal risk Programs in this field handle sensitive material: supervisory files, live disputes, personal data. #risk_management here is not optional. Clear rules are needed on what may be used in teaching, what must be anonymised, what must never leave the partner's systems, and what happens if a participant discloses something in class that the partner would rather not have heard. data protection law in the relevant jurisdictions must be complied with, and the compliance analysis should be done before, not after, the first case study is written. 9.6 Overclaiming Finally, a modest risk that does substantial damage. Partnerships in this space have a tendency to announce transformation. A short program does not transform a justice system. Overclaiming devalues the currency, invites cynicism among participants who know better, and makes honest evaluation politically impossible, because nobody can afford to report that the transformation did not occur. Modest, specific claims, backed by evidence, build more trust over time than ambitious ones. 10. Implications 10.1 For universities and business schools Treat these partnerships as academic projects with a commercial dimension, not commercial projects with an academic dimension. Invest in a small number of deep relationships rather than many shallow ones. Build faculty capacity deliberately, because the practitioner-facing skill set is not the same as the research skill set and cannot be assumed. Protect assessment standards, because they are the only thing the university uniquely has. And publish: a partnership that generates no research is a missed opportunity and a weaker claim to the word partnership. 10.2 For financial centres and regulators Recognise that training is #capacity_building and that capacity is a competitiveness asset, which the leading centres already report explicitly. Do not buy courses; build a learning system, with an internal function capable of absorbing what partners bring. Insist on assessment that can be failed. Fund the unglamorous subjects, especially ethics and supervisory judgment, which will never be commercially popular. And accept independent evaluation, including evaluation that may be unflattering. 10.3 For courts and judicial institutions Keep control of the curriculum. Let judges lead. Use universities for what they are good at, which is depth, method, assessment and research, and not as a source of prestige. Be rigorous about funding sources and disclose them. Insist that education is separated visibly from any promotion of the jurisdiction. And invest in the registry and in court administration, which are chronically under-trained relative to the bench and which determine how court users actually experience justice. 10.4 For students and early-career professionals For readers who are students, three practical points follow. First, this field is a career. Program design, learning strategy, regulatory training and judicial education are professional roles, and they are growing in exactly the places this article describes. Second, credentials are not equal. When you choose one, ask what competency it certifies, who assesses it, whether anyone fails, and whether the assessment is independently moderated. The literature on employer perceptions suggests employers ask the same questions, even if they phrase them differently. Third, the ethical questions in this article are not academic. If you work in a financial centre or a court, you will eventually be asked to design, attend or fund something that sits close to one of the lines drawn in Section 9. It is easier to recognise the line if you have thought about it beforehand. 11. Limitations and Future Research 11.1 Limitations This article is conceptual. It builds a framework from existing evidence; it does not test it. The framework's components are plausible and grounded, but their relative importance is unknown, and it is entirely possible that some elements matter far less in practice than the argument here implies. The evidence base is also uneven. Published material clusters around a small number of prominent centres and courts. There is much less published evidence about what has failed, and almost none about partnerships that were quietly abandoned. This means the article is likely to understate how often these arrangements do not work. Finally, the normative positions taken in Section 9, particularly on judicial independence, reflect a considered view of what public institutions owe the public. Reasonable people take different views on how strict these boundaries should be, and readers should engage with the argument rather than accept the conclusion. 11.2 A research agenda Six lines of work would materially advance this field. Longitudinal outcome studies. Follow cohorts for three to five years and ask what actually changed in their practice and their institutions. Almost nothing of this kind exists in either executive education or judicial training, and its absence is the single biggest gap. Comparative governance studies. Compare partnership structures across several centres and courts, and relate governance design to durability and to program quality. This would test the architecture layer of the framework directly. Credential value studies. Track whether the credentials produced by these partnerships carry weight in the labour market, and for whom. The micro-credential literature calls for exactly this and has very little of it. Independence and influence studies. Examine empirically whether and how sponsorship shapes curricula in judicial and regulatory training. This is uncomfortable research and is therefore rare, which is precisely why it is needed. Studies of the periphery. Investigate training capacity in smaller jurisdictions connected to the major centres, and whether partnership models reach them or bypass them. Pedagogical effectiveness research. Test, in this specific population of senior professionals, which methods actually change judgment. Much of what is believed about experiential learning in professional education rests on studies of much younger learners. 12. Conclusion The institutions at the centre of this article are new by the standards of legal and financial history. Specialised commercial courts attached to financial zones, applying hybrid procedure in English to cross-border disputes, barely existed a generation ago. They now form part of the everyday infrastructure of international business, and scholars have begun to argue that they may represent the future of transnational adjudication. The people who make them work, the supervisors, the compliance officers, the registrars, the arbitrators and the judges, need learning that does not fit the shape of a traditional degree. Universities can supply much of what is missing, but only if they stop behaving as vendors and start behaving as partners. Financial centres can supply the context, the problems and the urgency, but only if they build the internal capacity to absorb what they are given. Courts can supply the authority and the judgment, but only if they protect their independence while doing so. The argument of this article is that the difference between a strategic partnerships arrangement that works and one that decorates a press release lies in three things. It lies in #policy_alignment, meaning honesty at the outset about why each partner is present and whom the program serves. It lies in architecture, meaning governance, decision rights, funding and faculty models that are decided rather than assumed. And it lies in accountability, meaning a willingness to define success in advance, measure it afterwards, and act on the answer even when the answer is unwelcome. None of this is exotic. It is ordinary institutional discipline applied to an unusual setting. What makes the setting unusual is that the stakes are not only commercial. Badly designed training in a financial centre produces supervisors who cannot see risk. Badly designed training in a court produces judges who are competent but compromised, or independent but unequipped. Both outcomes are costly, and neither is visible until it is expensive. The opportunity is correspondingly large. A well-designed collaborative program can raise regulatory capability across a whole market, build a professional community that did not previously exist, generate research that improves practice everywhere, and give individual professionals credentials that genuinely represent what they can do. That is worth the effort of getting the architecture right, and it is worth the discomfort of measuring honestly whether it worked. Hashtags #Strategic_Partnerships_in_Executive_Education #Executive_Education_Partnerships #International_Financial_Centres #Judicial_Centres #Collaborative_Training_Programs #Financial_Center_Training #Judicial_Capacity_Building #Cross_Border_Legal_Education #Global_Finance_Education #Court_and_Campus_Collaboration #Professional_Learning #Higher_Education_Strategy #Workforce_Upskilling #Transnational_Adjudication #STULIB References Ahsan, K., Akbar, S., Kam, B., & Abdulrahman, M. D. A. (2023). Implementation of micro-credentials in higher education: A systematic literature review. Education and Information Technologies, 28(10), 13505-13540. https://doi.org/10.1007/s10639-023-11739-z Alasmari, T., & Alzahrani, A. (2024). Leveraging employment with micro-credentials: Policy and practice of the Middle East and North African universities. Educational Technology Research and Development, 72(3), 1869-1891. https://doi.org/10.1007/s11423-024-10367-9 Alsobhi, H. A., Alakhtar, R. A., Ubaid, A., Hussain, O. K., & Hussain, F. K. (2023). Blockchain-based micro-credentialing system in higher education institutions: Systematic literature review. Knowledge-Based Systems, 265, 110238. https://doi.org/10.1016/j.knosys.2022.110238 Antonopoulou, G. (2022). Procedure before international commercial courts and ordinary courts: A comparative perspective. In S. Brekoulakis & G. Dimitropoulos (Eds.), International commercial courts: The future of transnational adjudication (pp. 421-444). Cambridge University Press. https://doi.org/10.1017/9781009023122 Antonopoulou, G. (2023). The "arbitralization" of courts: The role of international commercial arbitration in the establishment and the procedural design of international commercial courts. Journal of International Dispute Settlement, 14(3), 328-349. https://doi.org/10.1093/jnlids/idad007 Antonopoulou, G. (2024). Forum marketing in international commercial courts? Oxford Journal of Legal Studies, 44(4), 860-888. https://doi.org/10.1093/ojls/gqae022 Bantekas, I. (2022). International commercial courts. In S. Brekoulakis & G. Dimitropoulos (Eds.), International commercial courts: The future of transnational adjudication. Cambridge University Press. https://doi.org/10.1017/9781009023122.007 Bayly-Castaneda, K., Ramirez-Montoya, M. S., & Morita-Alexander, A. (2024). Crafting personalized learning paths with AI for lifelong learning: A systematic literature review. Frontiers in Education, 9, 1424386. https://doi.org/10.3389/feduc.2024.1424386 Brekoulakis, S., & Dimitropoulos, G. (Eds.). (2022). International commercial courts: The future of transnational adjudication. Cambridge University Press. https://doi.org/10.1017/9781009023122 Brown, M., & Duart, J. M. (2024). Exploring gaps in the quality assurance of micro-credentials: A global mapping review of current practices. Journal of Open, Distance, and Digital Education, 1(1), 1-16. Brown, M., McGreal, R., & Peters, M. (2023). A strategic institutional response to micro-credentials: Key questions for educational leaders. Journal of Interactive Media in Education, 2023(1). https://doi.org/10.5334/jime.801 Brown, M., Nic Giolla Mhichil, M., Beirne, E., & Mac Lochlainn, C. (2021). The global micro-credential landscape: Charting a new credential ecology for lifelong learning. Journal of Learning for Development, 8(2), 228-254. https://doi.org/10.56059/jl4d.v8i2.525 Bruguera, C., Pages, C., Peters, M., & Fito, A. (2024). Micro-credentials and soft skills in online education: The employers' perspective. Distance Education. https://doi.org/10.1080/01587919.2024.2435645 Chartered Association of Business Schools. (2023). The impact of executive education. Chartered Association of Business Schools. Romero-Sanchez, A., Perdomo-Charry, G., & Burbano-Vallejo, E. L. (2024). Exploring the entrepreneurial landscape of university-industry collaboration on public university spin-off creation: A systematic literature review. Heliyon, 10(6), e27258. https://doi.org/10.1016/j.heliyon.2024.e27258 Rossoni, A. L., de Vasconcellos, E. P. G., & de Castilho Rossoni, R. L. (2024). Barriers and facilitators of university-industry collaboration for research, development and innovation: A systematic review. Management Review Quarterly, 74(3), 1841-1877. Schultz, T., & Bachmann, C. (2022). International commercial courts: Possible problematic social externalities of a dispute resolution product with good market potential. In S. Brekoulakis & G. Dimitropoulos (Eds.), International commercial courts: The future of transnational adjudication (pp. 52-70). Cambridge University Press. https://doi.org/10.1017/9781009023122 Selvaratnam, R. M., & Sankey, M. D. (2021). An integrative literature review of the implementation of micro-credentials in higher education: Implications for practice in Australasia. Journal of Teaching and Learning for Graduate Employability, 12(1), 1-17. Sorsa, V., Nevalainen, P., Korhonen, A., & Kansikas, J. (2026). Executive education tradition: Mechanisms of business elite formation. Academy of Management Learning and Education. https://doi.org/10.5465/amle.2024.0416 Tamoliune, G., Greenspon, R., Tereseviciene, M., Volungeviciene, A., Trepule, E., & Dauksiene, E. (2023). Exploring the potential of micro-credentials: A systematic literature review. Frontiers in Education, 7, 1006811. https://doi.org/10.3389/feduc.2022.1006811 Van Calster, G. (2022). Brexit and the competition of dispute resolution fora in Europe. In S. Brekoulakis & G. Dimitropoulos (Eds.), International commercial courts: The future of transnational adjudication. Cambridge University Press. https://doi.org/10.1017/9781009023122.024 Varadarajan, S., Koh, J. H. L., & Daniel, B. K. (2023). A systematic review of the opportunities and challenges of micro-credentials for multiple stakeholders: Learners, employers, higher education institutions and government. International Journal of Educational Technology in Higher Education, 20, Article 13. https://doi.org/10.1186/s41239-023-00381-x

Global supply chains have moved through a decade of repeated shocks. A pandemic, a canal blockage, port congestion, export controls on advanced chips, trade wars, sanctions, attacks on shipping lanes, and extreme weather have all tested networks that were designed mainly for cost efficiency. Traditional #risk_management tools, which rely on periodic reviews, static safety stock rules, and human judgement under time pressure, often react too slowly. This article examines whether #Agentic_AI, a class of artificial intelligence systems that can perceive, plan, use tools, act, and learn with limited human instruction, can meaningfully improve #supply_chain_resilience. The study uses an integrative review method, combining a structured reading of peer reviewed literature published between 2020 and 2026 with a conceptual synthesis grounded in the dynamic capabilities view, complex adaptive systems theory, and the viable supply chain model. The article develops a five layer reference architecture for agentic supply chain systems, covering perception, memory, reasoning and planning, tool use and action, and governance. It then maps this architecture onto two families of disruption: geopolitical shocks such as #sanctions, #export_controls, and #tariffs, and logistical shocks such as #port_congestion, carrier failure, and capacity collapse. Seven propositions are developed and eleven capability requirements are identified. The analysis argues that #Agentic_AI shifts resilience from a periodic planning exercise towards continuous #autonomous_adaptation, but only where data foundations, #interoperability, and #AI_governance are strong. The article also warns about new risks that agentic systems create, including #model_monoculture, correlated failure, agent driven amplification of the #ripple_effect, and accountability gaps. A maturity model and an implementation roadmap are offered for practitioners, along with a research agenda for scholars. The contribution is threefold: a clear definition of agentic autonomy in an operations context, a theory linked framework for autonomous disruption response, and a balanced assessment of benefits and systemic hazards. Keywords: agentic artificial intelligence; supply chain resilience; autonomous agents; geopolitical risk; logistics disruption; multi agent systems; digital supply chain twin 1. Introduction For most of the last thirty years, the guiding logic of global supply chain design was simple. Find the lowest landed cost, keep inventory thin, source from a small number of trusted suppliers, and rely on cheap and predictable ocean freight. That logic worked while the world was politically stable and transport capacity was abundant. It stopped working around 2018 and has not recovered since. The reasons are well documented. Trade tensions between major economies introduced tariffs that changed the economics of sourcing overnight. The COVID-19 pandemic closed factories, emptied container yards, and exposed how little most firms knew about their suppliers beyond the first tier. A single grounded vessel in the Suez Canal showed how one #chokepoints failure can propagate through thousands of shipments. Semiconductor shortages halted automotive assembly lines across three continents. Export controls on advanced computing hardware turned a technology market into an instrument of statecraft. Attacks on commercial shipping in the Red Sea forced carriers to reroute around southern Africa, adding weeks of transit time and reshaping freight rates. Each event was different in cause, but each produced the same pattern: a local shock that spread far beyond its origin. Researchers describe this spreading pattern as the #ripple_effect, where a disruption at one node cascades downstream and upstream through the network and degrades performance far from the point of impact. Dolgui and Ivanov (2021) argue that the ripple effect is now the central problem of disruption management, because modern networks are so tightly coupled that a supplier failure in one region can stop production in another within days. The traditional answer to this problem is to add buffers: more inventory, more suppliers, more capacity. Buffers work, but they are expensive, and Aldrighetti, Battini, Ivanov and Zennaro (2021) show that the cost of resilience is often underestimated in network design models, which explains why boards approve resilience projects slowly and cut them quickly. The alternative answer is speed. If a network cannot afford large buffers, it must instead detect problems earlier and reconfigure faster. This is where artificial intelligence enters the discussion. Over the past decade, firms have deployed #machine_learning for #demand_forecasting, anomaly detection, and #supplier_risk_scoring. These systems are useful, but they are fundamentally advisory. They produce a number, a score, or an alert, and a human then decides what to do. The decision loop still runs at human speed. Cohen and colleagues (2022) found a wide gap between resilience theory and resilience practice, and part of that gap is simply the time it takes for an insight to become an action. #Agentic_AI proposes a different arrangement. An agentic system does not stop at prediction. It is given a goal, it forms a plan, it calls external tools and systems to gather information and execute steps, it observes the result, and it revises the plan. Yao and colleagues (2023) demonstrated that interleaving reasoning and acting improves the reliability of language model agents on complex tasks. Schick and colleagues (2023) showed that models can learn to call external tools without hand written rules. Shinn and colleagues (2023) showed that agents can improve through structured self reflection on their own failures. Surveys by Wang and colleagues (2024) and Xi and colleagues (2025) trace how these building blocks combine into autonomous agents capable of long horizon tasks. Acharya, Kuppan and Divya (2025) describe agentic systems as goal directed rather than instruction directed, which is the key conceptual shift. Applied to operations, this means an agent could notice an unusual rise in congestion at a discharge port, check the exposure of open orders, simulate the delay impact on a production schedule, evaluate alternative routings, book capacity on a different lane, notify the affected plants, and log the decision, all within an hour, without waiting for a Monday planning meeting. Whether this is realistic, safe, and desirable is the question this article addresses. The research is guided by four questions. First, what exactly distinguishes agentic autonomy from the predictive and prescriptive analytics already used in supply chain management, and why does the distinction matter for resilience? Second, what architecture and capabilities does an agentic system need in order to act usefully during #geopolitical_disruption and #logistical_disruption? Third, through what mechanisms does autonomous adaptation improve the recognised phases of resilience, namely anticipation, resistance, recovery, and adaptation? Fourth, what new risks does agentic autonomy introduce at the level of the individual firm and at the level of the wider network, and how should these risks be governed? The article proceeds as follows. Section 2 reviews the literature on resilience, on AI in operations, and on agentic systems, and identifies the gap. Section 3 develops the theoretical framework and states seven propositions. Section 4 explains the method. Section 5 presents the findings, including the reference architecture, the disruption response mechanisms, and the capability requirements. Section 6 discusses implications, risks, and governance. Section 7 offers a maturity model and implementation roadmap. Section 8 sets out limitations and a research agenda. Section 9 concludes. 2. Literature Review 2.1 Supply chain resilience: from buffers to viability Supply chain resilience is usually defined as the ability of a network to prepare for unexpected events, respond to disruptions, and recover from them while maintaining continuity of operations at the desired level of connectedness and control. Katsaliaki, Galetsi and Kumar (2022) provide a broad review of disruption and resilience research and note that the field has grown quickly but remains fragmented, with many overlapping constructs. Hagele, Grosse and Ivanov (2023) go further and conduct a tertiary study, that is, a review of reviews, and confirm that resilience research is dominated by conceptual work and case studies rather than by tested interventions. Two capabilities appear repeatedly. The first is #redundancy, which means holding spare resources such as extra inventory, extra suppliers, or extra capacity. The second is #flexibility, which means the ability to switch between resources quickly. Kamalahmadi, Shekarian and Mellat Parast (2022) tested both empirically and found that flexibility generally outperforms redundancy in improving resilience, although the two are complements rather than substitutes. This finding matters for the present article, because flexibility is fundamentally a decision speed problem. A dual sourced supplier base only helps if the firm can actually switch, quickly, when the primary source fails. Ivanov and Dolgui (2020) extended resilience thinking with the concept of viability, arguing that in intertwined supply networks the goal is not only to bounce back from a shock but to survive long periods of instability and to keep serving society. Ivanov (2022) formalised this as the viable supply chain model, which integrates #agility, resilience, and sustainability. Ivanov (2023) later connected #viability to Industry 5.0, arguing that human centricity must sit alongside resilience and sustainability rather than being traded against them. Wieland and Durach (2021) offer a useful corrective by distinguishing an engineering perspective on resilience, which treats the network as a machine to be restored to equilibrium, from a social ecological perspective, which treats the network as a living system that must transform. Agentic systems, as this article argues, are attractive precisely because they promise to support transformation, not only restoration. A third strand of the literature concerns #visibility. Ivanov (2021) argued that #end_to_end_visibility built during the pandemic was one of the few durable gains of the period, because firms that could see their multi tier exposure were able to act while others were still gathering data. Kinra, Ivanov, Das and Dolgui (2020) showed how supplier risk exposure can be quantified through structural analysis of the network. Brintrup and colleagues (2020) demonstrated that supplier disruption can be predicted from historical transactional data in complex asset manufacturing. Kosasih and Brintrup (2022) went a step further and used #graph_neural_networks to infer hidden links in supply networks, which addresses the persistent problem that firms simply do not know who their sub tier suppliers are. Wichmann and colleagues (2020) extracted supply chain maps from news articles using deep neural networks, which is directly relevant to geopolitical monitoring. 2.2 From analytics to autonomy: artificial intelligence in supply chain management The application of AI to operations is not new, but its character has changed. Pournader, Ghaderi, Hassanzadegan and Fahimnia (2021) mapped AI applications across supply chain management and found heavy concentration in forecasting, classification, and optimisation. Sharma, Shishodia, Gunasekaran, Min and Munim (2022) mapped the same territory and identified a shift from descriptive towards #predictive_analytics and #prescriptive_analytics. Modgil, Gupta, Stekelorum and Laguir (2022) studied AI use during COVID-19 and found that firms with mature AI capability recovered faster, mainly because they could re-plan more often. Gupta, Modgil, Meissonier and Dwivedi (2021) reached a similar conclusion regarding information system resilience. Naz, Kumar, Majumdar and Agrawal (2022) reviewed whether AI is genuinely an enabler of resilience after COVID-19 and concluded that the evidence is promising but thin, with most studies reporting perceived rather than measured benefits. Zamani, Smyth, Gupta and Dennehy (2023) reviewed AI and big data analytics for resilience and identified an important limitation: most systems support human decisions but do not close the loop between insight and action. Dubey, Bryde, Dwivedi, Graham and Foropon (2022) showed, in a humanitarian setting, that an analytics culture, not just analytics technology, drives agility and resilience. This is a recurring finding. Technology alone does not create capability. A distinct and highly relevant strand concerns #reinforcement_learning. Oroojlooyjadid, Nazari, Snyder and Takac (2022) applied deep Q networks to the beer game and showed that learned policies can outperform classical heuristics under partial information. Gijsbrechts, Boute, Van Mieghem and Zhang (2022) tested deep reinforcement learning on lost sales, dual sourcing, and multi echelon problems and found competitive performance without problem specific tuning. Boute, Gijsbrechts, van Jaarsveld and Vanvuchelen (2022) offered a roadmap for deep reinforcement learning in inventory control and were careful to note where it does not help. Rolf and colleagues (2023) reviewed reinforcement learning across supply chain management and identified a persistent gap between simulation results and industrial deployment. These studies matter because reinforcement learning gives us a policy, that is, a mapping from state to action, which is the closest classical analogue to what an agent does. Digital twins bridge analytics and autonomy. Ivanov and Dolgui (2021) proposed the #digital_supply_chain_twin as a way to manage disruption risk in real time by mirroring the physical network in a computational model. Ivanov (2023) later developed the intelligent digital twin for #stress_testing, resilience, and viability, arguing that a twin is most valuable when it can be run continuously rather than as an occasional exercise. Spieske and Birkel (2021) reviewed Industry 4.0 technologies and resilience and found that #simulation, visibility, and connectivity technologies produce the clearest gains. Dolgui and Ivanov (2022) argued that 5G and the internet of everything provide the connectivity layer needed for real time twins. Fragapane, Ivanov, Peron, Sgarbossa and Strandhagen (2022) showed how autonomous mobile robots increase flexibility inside production networks, which is a physical form of the same idea. Generative AI is the most recent addition. Jackson, Ivanov, Dolgui and Namdar (2024) developed a capability based framework for generative artificial intelligence in supply chain and operations management, and explicitly raised the possibility of autonomous agents that plan and execute. Dwivedi and colleagues (2023) discussed the wider implications of generative conversational AI across research, practice, and policy. Sheffi (2023) argued that AI will change the nature of supply chain work rather than simply remove it, a point that connects directly to #workforce_transformation. 2.3 Agentic AI: definition, architecture, and capability The word agent has a long history in artificial intelligence. Russell and Norvig (2021) define an agent as anything that perceives its environment through sensors and acts upon it through actuators, and a rational agent as one that acts to maximise expected performance given what it knows. What has changed is not the definition but the capability of the underlying models. Bommasani and colleagues (2021) described the rise of foundation models, which are large models trained on broad data and adaptable to many tasks. When such a model is placed inside a loop that lets it reason, call tools, observe results, and try again, it becomes an agent in a practical rather than merely theoretical sense. Yao and colleagues (2023) introduced ReAct, which interleaves reasoning traces with actions and showed that this reduces hallucination and improves task success. Schick and colleagues (2023) introduced Toolformer, showing that models can teach themselves when and how to call APIs, calculators, and search engines. Shinn and colleagues (2023) introduced Reflexion, in which an agent verbally reflects on failure and stores the lesson for future attempts, producing a simple but effective form of learning without weight updates. Park and colleagues (2023) built generative agents with memory, reflection, and planning modules that produced believable long horizon behaviour. From these components, a consistent architecture emerges. Wang and colleagues (2024) survey large language model based autonomous agents and identify four modules: profile, memory, planning, and action. Xi and colleagues (2025) describe a brain, perception, and action structure. Acharya, Kuppan and Divya (2025) survey agentic systems more broadly and emphasise goal decomposition, tool orchestration, and multi agent collaboration. Across these treatments, five capabilities recur: Perception, meaning the ability to ingest heterogeneous signals from structured systems and unstructured sources. Memory, meaning both short term working context and long term storage of experience, often in vector databases or #knowledge_graphs. Reasoning and planning, meaning the decomposition of a goal into steps and the revision of those steps when the world changes. Tool use and action, meaning the ability to call external systems, including transactional systems of record, and to execute changes in them. Reflection and learning, meaning the ability to evaluate outcomes and update behaviour. #multi_agent_systems extend this further. Instead of one agent doing everything, specialised agents negotiate, delegate, and check one another. This is architecturally attractive in supply chains, which are themselves distributed systems of semi autonomous actors with partially conflicting objectives. The literature is also clear about risk. Chan and colleagues (2023) analyse harms from increasingly agentic algorithmic systems and identify a set of properties, including underspecification, directness of impact, goal directedness, and long term planning, that raise the stakes as autonomy grows. Shavit and colleagues (2023) propose practices for governing agentic systems, including agent identification, human oversight of high impact actions, and interruptibility. Gabriel and colleagues (2024) analyse the ethics of advanced AI assistants and stress alignment with user intent and the avoidance of manipulation. Hendrycks, Mazeika and Woodside (2023) survey catastrophic risks and note that competitive pressure can push organisations to deploy autonomy faster than they can supervise it, which is precisely the dynamic one would expect in a cost driven industry. 2.4 Geopolitics and the reconfiguration of global networks Geopolitical disruption is qualitatively different from a warehouse fire. It is often slow to build, fast to bite, legally complex, and irreversible. Roscoe and colleagues (2022) studied redesign under compounding geopolitical disruptions and found that firms apply different supply chain logics, and that these logics shape which responses they even consider. Gereffi, Pananond and Pedersen (2022) analysed resilience through the lens of global value chains and argued that state capability, firm capability, and chain structure interact, so that a purely firm level response is often insufficient. Handfield, Graham and Burns (2020) examined tariffs and trade wars and predicted an evolutionary redesign of networks, which is broadly what has occurred through #nearshoring and #friendshoring. Panwar, Pinkse and De Marchi (2022) considered the future of global supply chains after COVID-19 and questioned whether decoupling is economically viable at scale. Free and Hecimovic (2021) asked whether the pandemic marked the end of the road for a particular model of globalisation. The technology dimension is central. Miller (2022) documents how #semiconductors became the decisive strategic industry and how control over specific tools and materials creates leverage that is very hard to replicate. Similar dynamics apply to #critical_minerals, where processing capacity is geographically concentrated even where extraction is not. For a supply chain manager, this means the relevant risk is often not the supplier but the supplier of the supplier, and not the country of assembly but the country of a licence. Choi, Netland, Sanders, Sodhi and Wagner (2023) revisited just in time production in turbulent times and argued that the answer is not to abandon lean but to make it context aware. Sodhi and Tang (2021) called for research on supply chain management under extreme conditions. Craighead, Ketchen and Darby (2020) offered a theoretical toolbox for pandemic era research. Shih (2020) argued that firms must know their networks in depth and diversify deliberately. Sarkis (2021) connected the disruption debate to sustainability, warning that resilience measures can increase emissions if they simply add stock and airfreight. 2.5 The research gap Three observations follow from this review. First, resilience research knows what capabilities matter, particularly flexibility, visibility, and rapid re-planning, but it has said relatively little about who or what performs the re-planning when disruptions arrive faster than committees can meet. Second, AI in operations research has produced strong predictive tools and promising reinforcement learning policies, but Zamani and colleagues (2023) and Naz and colleagues (2022) both note the missing link between insight and action. #decision_latency remains a bottleneck. Third, the agentic AI literature is technically rich but domain thin. It demonstrates capability on benchmarks and simulated environments, not in regulated, capital intensive, multi party industrial settings with legal exposure. The gap, therefore, is an integrated account of how agentic autonomy could operate inside real supply networks under geopolitical and logistical stress, what it would require, what it would improve, and what it would endanger. This article addresses that gap. 3. Theoretical Framework and Propositions 3.1 Three theoretical lenses The framework rests on three complementary theories. The dynamic capabilities view holds that competitive advantage in changing environments comes from the capacity to sense opportunities and threats, seize them through timely decisions, and transform the resource base. Applied here, Agentic AI can be read as an instrument that industrialises sensing and compresses seizing. Sensing becomes continuous rather than periodic. Seizing becomes machine speed rather than meeting speed. Transformation, however, remains organisational and human, because it involves contracts, capital, and politics. This suggests that agentic systems will strengthen the first two capabilities far more than the third, at least in the near term. The complex adaptive systems lens treats a supply network as a population of interacting agents whose local behaviour produces global patterns that no single actor designed. The ripple effect is exactly such a pattern. This lens is valuable because it warns us that adding autonomous agents does not simply add speed to an existing system. It changes the system. If many firms deploy similar agents that respond to the same signals with similar logic, the aggregate behaviour may become more synchronised and therefore more volatile, not less. The viable supply chain model of Ivanov (2022) integrates agility, resilience, and sustainability and asks whether a network can survive long term instability while continuing to serve society. This lens keeps the analysis honest. An agentic system that protects one firm's service level by outbidding everyone else for scarce capacity has improved that firm's resilience and possibly reduced the viability of the network. 3.2 Propositions Proposition 1. Agentic systems improve the anticipation phase of resilience primarily by widening #perception_layer coverage, that is, by monitoring a far larger number of weak signals across structured and unstructured sources than human teams can monitor, thereby increasing the probability of #early_warning before a disruption becomes visible in transactional data. Proposition 2. Agentic systems improve the response phase primarily by reducing decision latency, that is, the elapsed time between signal detection and executed countermeasure, and this reduction produces disproportionate benefit in networks with low redundancy, because in such networks the value of a countermeasure decays rapidly with time. Proposition 3. The resilience benefit of agentic autonomy is conditional on the quality and coverage of the underlying data. Where multi tier mapping is absent and master data is inconsistent, agentic autonomy converts slow bad decisions into fast bad decisions, and #data_quality therefore moderates the relationship between agentic capability and resilience outcomes. Proposition 4. Agentic systems are more effective against logistical disruption than against geopolitical disruption in the short term, because logistical disruptions have richer real time telemetry, more standardised remedies, and lower legal ambiguity, whereas geopolitical responses involve licences, contracts, and sovereign decisions that lie outside the agent's action space. Proposition 5. multi agent systems that span organisational boundaries deliver greater network level resilience than single firm agents, but only where #interoperability standards and data sharing agreements exist, since without them agents optimise locally and can worsen the ripple effect. Proposition 6. Widespread adoption of similar agentic systems trained on similar data increases model monoculture and therefore raises #systemic_risk, because correlated agent behaviour synchronises demand for scarce alternative capacity and can transform a local shock into a market wide one. Proposition 7. Effective agentic deployment requires graduated autonomy governed by decision impact, with #human_in_the_loop control retained for irreversible, high value, or legally sensitive actions, and full autonomy reserved for reversible, bounded, and frequent actions. 4. Methodology This study uses an integrative conceptual review. This design is appropriate because the subject is emerging, the empirical base is thin, and the aim is to build a framework rather than test a hypothesis. Search strategy. Peer reviewed literature was identified from major operations, information systems, and computer science outlets. Search terms combined three clusters: a resilience cluster (supply chain resilience, disruption, ripple effect, viability, risk management), a technology cluster (artificial intelligence, agentic AI, autonomous agents, multi agent systems, reinforcement learning, large language models, digital twin), and a context cluster (geopolitical, sanctions, tariffs, export controls, logistics, ports, shipping). Inclusion criteria. Sources were included if they were peer reviewed articles, established scholarly books, or widely cited technical reports; if they were published between 2020 and 2026; and if they contributed either a theoretical construct, an empirical finding, or an architectural pattern relevant to the research questions. Practitioner marketing material was excluded. Sources older than the window were excluded except where they define a foundational construct that has no recent equivalent. Synthesis procedure. The synthesis followed four steps. First, resilience constructs were extracted and grouped into the four phases of anticipation, resistance, recovery, and adaptation. Second, agentic capabilities were extracted from the AI literature and grouped into the five layers described in Section 5.1. Third, the two sets were cross mapped in a capability to phase matrix, and gaps were recorded where a resilience phase had no corresponding agentic capability. Fourth, disruption archetypes were constructed for geopolitical and logistical shocks, and the mapped capabilities were traced through each archetype to identify preconditions and failure modes. Illustrative scenarios. Because deployed agentic systems in industry are rarely documented in the academic record, the article uses illustrative scenarios rather than case studies. These scenarios are constructed from documented disruption patterns reported in the reviewed literature and are used analytically, to test whether the proposed architecture could plausibly act, not as evidence that it has acted. Limitations of the method. The approach cannot establish causality and cannot estimate effect sizes. It is vulnerable to the optimism of a young technical literature, in which reported successes are more visible than failures. These limitations are addressed in Section 8. 5. Findings 5.1 A five layer reference architecture for agentic supply chain systems Synthesising the agent architectures of Wang and colleagues (2024), Xi and colleagues (2025), and Acharya and colleagues (2025) with the operational requirements identified by Ivanov (2023) and Jackson and colleagues (2024), the following five layer architecture emerges. Layer 1: Perception. The perception layer converts the world into machine readable state. It draws on three source families. The first is internal transactional data from enterprise resource planning, warehouse management, and transport management systems: open orders, inventory positions, lead times, bills of material, and shipment status. The second is partner and platform data: carrier schedules, vessel positions, port dwell times, customs status, and supplier confirmations. The third is external and largely unstructured data: news, regulatory notices, sanctions listings, weather and hazard feeds, and social signals. Wichmann and colleagues (2020) showed that supply network structure can be inferred from news text, which is precisely the capability needed here. Kosasih and Brintrup (2022) showed that missing links can be predicted with graph neural networks, which allows the perception layer to reason about tiers it cannot directly observe. The critical design point is that perception must be continuous and event driven, not batch. A weekly extract is not perception. It is a photograph. Layer 2: Memory and world model. Memory holds three things. Semantic memory holds the structure of the world: the supply network graph, product hierarchies, contractual terms, compliance rules, and site capabilities. This is best represented as a knowledge graph, because supply chains are relational and queries are inherently multi hop. Episodic memory holds what has happened: past disruptions, past decisions, and their outcomes. Procedural memory holds how things are done: playbooks, approved substitution rules, escalation paths. The world model is where the digital supply chain twin lives. Ivanov and Dolgui (2021) and Ivanov (2023) argue that a twin turns a data model into a decision instrument because it allows counterfactual questions. An agent without a twin can only pattern match on history. An agent with a twin can ask what would happen if. Layer 3: Reasoning and planning. The #planning_module decomposes a goal such as protect the launch schedule for product line X into subgoals, orders them, and revises them as the situation develops. Three planning styles are relevant. Deliberative planning searches over possible action sequences and is suited to slow, high stakes decisions such as network redesign. Reactive planning maps states to actions quickly and is suited to fast, bounded decisions such as re-routing a container. Learned policy planning, drawing on the reinforcement learning work of Boute and colleagues (2022) and Gijsbrechts and colleagues (2022), is suited to repeated decisions such as replenishment under uncertainty. In practice, these are combined. A large language model provides flexible reasoning over unstructured context and orchestrates the workflow, while classical solvers and learned policies provide reliable optimisation within well defined subproblems. This hybrid pattern is important: it uses the model where flexibility is needed and constrains it where correctness is needed. Layer 4: Tool use and action. The #action_layer is what separates an agent from an analyst. Through #tool_use, the agent invokes external capabilities: it queries the twin, runs an optimisation solver, calls a carrier booking interface, checks a denied party list, raises a purchase requisition, or opens a case in a workflow system. Schick and colleagues (2023) showed that tool invocation can be learned. In an industrial setting it should also be constrained, with each tool exposed through a permissioned interface and each action logged. Actions must be classified by reversibility and impact. Requesting a quote is reversible and low impact. Cancelling a purchase order is not. This classification is the foundation of the graduated autonomy model in Proposition 7. Layer 5: Governance, reflection, and learning. This layer contains everything that makes the system safe to run. It includes #explainable_AI mechanisms that render the agent's reasoning inspectable, following the argument of Ali and colleagues (2023) that explainability is a precondition of #trustworthy_AI rather than an optional extra. It includes audit logging of every observation, decision, and action. It includes guardrails: hard constraints the agent may not violate regardless of its objective, such as trade compliance rules. It includes escalation logic that routes decisions to humans when confidence is low or impact is high. And it includes reflection, in the sense of Shinn and colleagues (2023), so that outcomes feed back into episodic memory and improve future behaviour. The five layers can be summarised in a single sentence. Perception tells the agent what is happening, memory tells it what the world is like and what happened before, planning tells it what to do, tools let it do it, and governance decides whether it is allowed to. 5.2 Autonomous adaptation to logistical disruptions Logistical disruptions are the natural first domain for agentic autonomy because they are frequent, observable, and reversible. Archetype 1: Port congestion and berth delay. Congestion at a discharge port raises dwell times and delays inland delivery. The perception layer detects rising dwell times and vessel queuing. The world model maps affected vessels to affected containers to affected orders to affected production lines, which is the multi hop query that knowledge graphs make tractable. The planning module then evaluates options: wait, discharge at an alternative port and move inland, split the shipment, expedite a subset by #air_freight, or reschedule production around the delayed parts. The action layer executes the chosen option through booking and planning systems. A human approves anything above a cost threshold. The value here is not that the machine finds an option no human could find. It is that the machine finds it on day one rather than day six, when alternative capacity is still available and cheaper. This is Proposition 2 in concrete form. Archetype 2: Carrier or lane failure. A carrier withdraws capacity, an airline cancels a freighter rotation, or a rail corridor closes. The agent recomputes routing across the remaining network, respecting service commitments and cost constraints, and rebooks. #transport_routing is a well structured optimisation problem with mature solvers, so the agent's contribution is orchestration and speed rather than mathematical novelty. Archetype 3: Demand and supply shock in a single node. A supplier plant halts. The agent checks alternative qualified sources, checks inventory across the network including in transit stock, reallocates constrained supply across customers according to a policy the firm has approved, and triggers #dual_sourcing activation where a qualified second source exists. Here the constraint is rarely the algorithm. It is qualification. An agent cannot qualify a supplier, and this is a hard limit on autonomy. Archetype 4: Inventory drift under sustained uncertainty. Not all disruption is an event. Some is a slow degradation of lead time reliability. #autonomous_replenishment policies learned through reinforcement learning can adapt safety stock continuously rather than at quarterly reviews. Boute and colleagues (2022) caution that learned policies need careful validation and can fail silently when the environment shifts outside the training distribution, which argues for keeping a classical policy as a fallback and monitoring divergence. Across these archetypes, three enablers appear consistently. The first is end to end visibility, without which the agent cannot construct the exposure map. The second is interoperability, because the agent's usefulness is limited by which systems it can actually call. The third is a pre agreed decision policy, because an agent that must ask a human what to prioritise every time is not saving anyone time. 5.3 Autonomous adaptation to geopolitical disruptions Geopolitical shocks are harder, and the analysis here supports Proposition 4. Archetype 5: Sanctions and denied party changes. A new listing is published. Compliance exposure must be assessed immediately across suppliers, customers, banks, vessels, and beneficial owners. This is a text heavy, relational, time critical task, which is exactly where agentic systems are strong. The agent monitors regulatory publications, parses new listings, matches them against the entity graph including indirect ownership, flags exposed transactions, freezes what must be frozen, and produces a documented assessment for the compliance officer. Crucially, the agent should not decide the legal question. It should assemble the evidence and propose an action, and a qualified human should decide. This is #trade_compliance as machine assisted diligence, not machine judgement. Archetype 6: Export controls on a critical input. Controls are placed on a category of components or on the tools used to make them. The immediate operational task is to determine exposure across the bill of material, including at tiers the firm cannot see. graph neural networks based link prediction, following Kosasih and Brintrup (2022), can propose likely hidden dependencies for human verification. The agent then runs the twin to simulate the production impact under different licence outcomes and different substitution options. It cannot obtain a licence. It cannot redesign a product. But it can tell the organisation, within hours instead of weeks, which products stop and when. Archetype 7: Tariff changes and trade policy shifts. tariffs change landed cost and therefore change sourcing optimality. An agent can continuously recompute landed cost across the qualified supplier set, model the effect of rules of origin, and propose reallocation of volume between existing sources. What it cannot do is decide whether to build a plant in another country. Handfield and colleagues (2020) and Roscoe and colleagues (2022) both emphasise that such redesign is strategic, political, and slow. Archetype 8: Chokepoint closure and maritime security. When a strait or canal becomes unsafe, carriers reroute, transit times lengthen, and effective global capacity falls because the same ships are at sea for longer. The agent's role is anticipatory. It should detect the security signal early through #situational_awareness monitoring, quantify the exposure of goods in transit and goods planned, and reprioritise bookings before rates spike. Speed matters enormously here, because freight markets reprice quickly once everyone sees the same thing. Archetype 9: Concentration risk in critical inputs. critical minerals processing, advanced semiconductors, and certain #pharmaceutical_supply_chains inputs are concentrated in ways that no single firm can diversify away in the short run. Miller (2022) documents how deep and specific such dependencies can be. Here the honest conclusion is that agentic AI offers visibility and preparedness, not solution. It can identify the exposure, quantify it, and monitor it. Removing it requires capital, time, and often state policy, which is the point made by Gereffi and colleagues (2022). The pattern is clear. In geopolitical disruption, agentic systems excel at exposure discovery, monitoring, simulation, and documentation. They perform poorly, and should not be trusted, on legal interpretation, sovereign negotiation, and irreversible strategic commitment. The action space of the agent is bounded by what can be executed through a system interface, and the most consequential geopolitical responses cannot be. 5.4 Multi agent coordination across organisational boundaries The most interesting, and most difficult, possibility is coordination between agents belonging to different firms. In principle, a buyer's agent and a supplier's agent could negotiate allocation, exchange forecast updates, and jointly re-plan, achieving in minutes what currently takes weeks of email. Ivanov and Dolgui (2020) argue that viability is a property of intertwined networks, not individual firms, so network level coordination is where the largest gains should lie. Three obstacles stand in the way. The first is interoperability. Agents must share a common vocabulary for products, locations, events, and commitments. Without semantic standards, cross firm agent communication degrades into brittle integration projects. The second is trust and confidentiality. Firms do not want to reveal capacity, cost, or exposure to counterparties. Kshetri (2021) discusses how distributed ledger approaches can support traceability with controlled disclosure, and this line of work is relevant to the design of agent to agent trust, although blockchain is not a complete answer and adds its own cost. The third is incentive misalignment. An agent optimising its principal's service level will, under scarcity, compete for the same alternative capacity as everyone else's agent. If a hundred firms deploy agents that all detect a chokepoint closure at the same hour and all book the same alternative capacity at the same minute, the market clears violently. This is Proposition 6, and it deserves emphasis. Individually rational agentic behaviour can be collectively destabilising. This is not a hypothetical concern. It is the standard result in any market where reaction speed increases faster than supply elasticity. 5.5 Human and agent teaming Sheffi (2023) argues that AI will reshape supply chain work rather than eliminate it, moving people from execution towards exception handling, judgement, and relationship management. Fugener, Grahl, Gupta and Ketter (2022) study delegation between humans and AI and find that productive delegation requires people to understand where the machine is strong and where it is weak, which is harder than it sounds and does not happen automatically. Three design principles follow. Graduated autonomy. Actions should be tiered. Tier one actions are automatic: information gathering, simulation, alerting, reversible bookings below a threshold. Tier two actions are automatic with notification and a reversal window. Tier three actions require human approval: order cancellation, supplier switching, large financial commitments, anything with legal exposure. Tier four actions are advisory only: the agent may recommend but never execute, for example anything involving sanctions determination or contract termination. This tiering operationalises Proposition 7 and reflects the governance practices proposed by Shavit and colleagues (2023). Explainability by construction. explainable AI cannot be added afterwards. The agent should emit a reasoning trace, the evidence it used, the options it considered, the option it selected, and the expected consequence. Ali and colleagues (2023) argue that explainability is necessary for trust, and Glikson and Woolley (2020) show empirically that human trust in AI depends heavily on transparency and reliability signals. Deliberate friction. Some slowness is protective. A mandatory pause before irreversible action, a second agent that argues the opposite case, or a required human acknowledgement all reduce the chance of a fast, confident, wrong decision. In a system built for speed, friction must be designed in on purpose, because it will not appear by accident. 5.6 Capability requirements: what an organisation actually needs The analysis yields eleven requirements. They are ordered roughly by how often their absence blocks progress. One, a maintained multi tier supply network map, extending at minimum to tier three for critical parts. Two, clean and governed master data, particularly part, supplier, and site identifiers that reconcile across systems. Three, event driven data pipelines with latency measured in minutes, not days. Four, a digital supply chain twin capable of running counterfactual simulations on demand. Five, machine callable interfaces into planning, procurement, and transport systems, with permissioned write access. Six, an approved decision policy that encodes priorities, thresholds, and allocation rules in advance. Seven, a hard constraint library covering compliance, safety, and contractual limits that the agent may never override. Eight, complete audit logging of observations, reasoning, actions, and outcomes. Nine, an escalation and override mechanism that a human can trigger at any time, including a global stop. Ten, monitoring of the agent itself, tracking decision quality, drift, and the frequency of human overrides. Eleven, a workforce that understands the system well enough to supervise it, which is a #change_management task and not a technical one. Most organisations that attempt agentic deployment fail on requirements one through three. The intelligence layer is rarely the binding constraint. The data layer almost always is, which is the essence of Proposition 3. 6. Discussion 6.1 What genuinely changes The central change is the compression of the decision loop. Classical analytics improved the quality of a decision but left its timing to the organisation's calendar. Agentic systems attack the calendar itself. In a disruption, the value of a countermeasure decays with time, sometimes steeply, because alternatives are consumed by whoever moves first. A firm that reroutes on day one pays a premium. A firm that reroutes on day ten may find nothing to buy. Speed is therefore not a convenience. In a scarce market, speed is the resilience. The second change is the scope of attention. Human teams triage. They monitor what they have time to monitor, which means they monitor their largest suppliers and their loudest problems. An agentic system can monitor the long tail: the small supplier of a small component that stops an entire assembly line. This is a real and underappreciated benefit, because disruption history is full of trivial parts causing serious damage. The third change is institutional memory. Episodic memory, in the sense of Shinn and colleagues (2023), means that the lessons of a disruption survive the departure of the person who handled it. Most organisations lose their disruption learning through turnover. A well designed agent does not. 6.2 What does not change Agentic systems do not create capacity. They do not qualify suppliers. They do not obtain export licences, negotiate with governments, or build factories. They do not resolve the fundamental tension identified by Sarkis (2021) between resilience and #sustainability, in which faster and more flexible responses often mean more air freight and more emissions. Indeed, an agent optimising service level without an emissions constraint will systematically choose the higher carbon option, because that option is usually faster. This is an argument for writing sustainability constraints into the hard constraint library rather than hoping they emerge. Agentic systems also do not fix bad strategy. A network with a single point of failure in a politically exposed jurisdiction remains a network with a single point of failure. The agent will simply tell you about it more often and more precisely. 6.3 New risks introduced by autonomy Correlated behaviour and systemic risk. This is the most serious concern and follows directly from the complex adaptive systems lens. If agents across an industry share similar architectures, similar training data, similar signal sources, and similar objectives, their responses will correlate. Correlated responses concentrate demand for scarce alternatives, amplify price movements, and can convert a manageable regional problem into a market wide one. This is model monoculture, and it is a network level externality that no individual firm has an incentive to fix. Regulators and industry bodies may eventually need to address it, in the same way that financial regulators addressed correlated algorithmic trading. Amplification of the ripple effect. Dolgui and Ivanov (2021) describe how disruptions propagate. Autonomous agents propagate decisions as well as disruptions. If a buyer's agent cancels orders automatically, that cancellation becomes a demand shock for the supplier, whose agent may then cancel its own upstream orders. A cascade of automated cancellations can travel upstream faster than any human can intervene. Rate limits, cooling off periods, and cross tier notification requirements are plausible mitigations. Accountability gaps. When an agent makes a decision that causes loss, who is responsible? Chan and colleagues (2023) identify this as a core problem of agentic systems. The practical answer is that responsibility must remain with the deploying organisation, and this requires that every agent action be attributable, logged, and reviewable. #algorithmic_accountability is not achievable without infrastructure built for it in advance. Security and adversarial manipulation. An agent that reads external text and acts on it can be attacked through that text. A false report of a port closure, injected into a monitored feed, could trigger expensive and unnecessary rerouting. Prompt injection through supplier documents, invoices, or emails is a live threat surface. #cyber_security for agentic systems must include input provenance verification, cross source corroboration before consequential action, and strict separation between data the agent reads and instructions the agent follows. Silent degradation. Learned policies fail quietly when the world moves outside their training distribution. Boute and colleagues (2022) warn about this in inventory control. The organisational danger is that the agent keeps producing confident output that is quietly wrong, and because it has been reliable for months, nobody checks. Over trust and skill erosion. If humans stop practising disruption response because the machine does it, the human fallback degrades. When the machine fails, and eventually it will, the organisation discovers that the people who once knew how to handle a crisis have not done so in three years. 6.4 Governance Governance should be proportionate to impact, and it should exist before deployment rather than after the first incident. Four elements are essential. Autonomy tiering by reversibility and value. Described in Section 5.5. This is the single most important control. Constraint hardening. Compliance, safety, and contractual limits must be enforced outside the model, in the tool layer, not requested politely inside a prompt. An instruction can be argued with. A permission cannot. Continuous evaluation. Agent decisions should be sampled and reviewed, override rates tracked, and outcomes compared against a counterfactual baseline. If the override rate rises, something has drifted. Alignment with external frameworks. Public governance instruments, including the risk based approach of the European Union AI Act (2024) and the National Institute of Standards and Technology AI Risk Management Framework (2023), give organisations a defensible structure for classifying, documenting, and monitoring AI systems. Firms operating internationally will need to satisfy multiple regimes, and building for the strictest is usually cheaper than retrofitting. 6.5 Implications for theory For the dynamic capabilities view, the findings suggest that sensing and seizing can be substantially automated while transforming cannot, which implies that the competitive differentiator will migrate towards transformation capability once sensing and seizing become commodities available to everyone through similar tools. For complex adaptive systems theory, agentic deployment offers a rare natural experiment in what happens when the agents in a socio technical system become faster and more similar at the same time. The prediction from theory is increased synchronisation and reduced diversity of response, which means lower system level resilience even as firm level resilience improves. This is a testable and important claim. For the viable supply chain model, the findings suggest that viability requires network level rules of engagement for autonomous agents, not only firm level capability. Ivanov (2023) links viability to human centricity in Industry 5.0, and the graduated autonomy principle developed here is one concrete expression of #Industry_5 thinking in an operations context. 7. Implementation: A Maturity Model and Roadmap 7.1 Five levels of agentic maturity Level 0, Manual. Disruption response is entirely human. Data is gathered on demand. Response time is measured in days or weeks. Level 1, Assisted. Dashboards and alerts exist. Analytics predict risk. Humans decide and execute. Response time is measured in days. Level 2, Recommending. The system proposes specific actions with quantified consequences drawn from a twin. Humans approve and the system executes. Response time is measured in hours. Most advanced organisations that describe themselves as having a #control_tower are here. Level 3, Supervised autonomy. The system executes tier one and tier two actions independently within a defined envelope and escalates the rest. Humans supervise by exception. Response time is measured in minutes for bounded decisions. Level 4, Networked autonomy. Agents coordinate across organisational boundaries under shared protocols and interoperability standards. Response is coordinated rather than merely fast. Very few, if any, organisations are here today, and reaching this level requires collective action rather than individual investment. The important insight from the maturity model is that levels cannot be skipped. An organisation at Level 1 that attempts Level 3 will discover that its data foundation cannot support autonomous action, which returns us to Proposition 3. 7.2 A phased roadmap Phase one, foundations, roughly six to twelve months. Build the multi tier map for critical parts. Fix master data. Establish event driven pipelines. Stand up the twin. This phase produces no autonomy and is therefore politically difficult to fund, which is why so many programmes stall here. It is nonetheless the phase that determines whether everything after it works. Phase two, perception and simulation, roughly three to six months. Deploy agents in read only mode. Let them monitor, detect, simulate, and recommend. Measure their early warning performance against what the human team would have caught. This phase builds trust and produces evidence, and it carries almost no operational risk because the agent cannot act. Phase three, bounded action, roughly six to twelve months. Grant write access for a narrow, reversible, high frequency decision class, for example intra network stock transfers below a value threshold or expedite requests within an agreed budget. Instrument everything. Track override rates. Expand the envelope only when performance justifies it. Run #pilot_projects in a single region or product family rather than enterprise wide. Phase four, supervised autonomy, ongoing. Expand the action envelope by decision class, not by department. Formalise the escalation policy. Establish the review board. Begin measuring #return_on_investment against a counterfactual, which is difficult but necessary, since the natural comparison, namely what would have happened otherwise, is unobservable and must be estimated through simulation. Phase five, network extension, ongoing and collective. Engage suppliers, carriers, and industry bodies on shared protocols. This cannot be done alone. 7.3 Measurement Traditional metrics such as service level and inventory turns are necessary but insufficient, because they do not isolate the agent's contribution. Four additional measures are proposed. Time to detect, meaning the elapsed time between the real world onset of a disruption and its recognition by the system. Time to decide, meaning the elapsed time between recognition and a committed countermeasure. Exposure coverage, meaning the proportion of the bill of material for which multi tier exposure is actually known rather than assumed. Counterfactual loss avoided, meaning the estimated difference between realised outcome and simulated outcome under the previous decision process. This last measure is the hardest and the most valuable, and it requires the twin to be maintained precisely so that the counterfactual can be run. 8. Limitations and Future Research 8.1 Limitations This article is conceptual. It builds a framework from literature and reasons through scenarios. It does not test the framework against deployed systems, because such systems are, at present, either absent or undisclosed. The propositions are therefore plausible rather than established. The literature base is also asymmetric. The resilience literature is mature and empirical. The agentic AI literature is young and heavily weighted towards benchmarks, demonstrations, and surveys, with limited evidence from regulated industrial environments. There is a real risk that the capabilities of agentic systems are overstated in the sources available, since negative results are less likely to be published or publicised. Readers should treat the technical claims as capability ceilings rather than typical performance. Finally, the geopolitical analysis is necessarily general. Sanctions regimes, export controls, and tariff structures differ by jurisdiction and change frequently. The article deliberately describes mechanisms rather than specific measures, because specific measures would be out of date almost immediately. 8.2 A research agenda Empirical validation of decision latency effects. The core claim of this article, that reduced decision latency produces disproportionate resilience gains in low redundancy networks, is testable. Simulation studies using established ripple effect models could quantify the relationship between response time and total disruption cost across different network topologies and redundancy levels. Systemic effects of correlated agents. Agent based simulation should be used to test Proposition 6 directly. If many firms deploy similar agents, does aggregate volatility rise? Under what conditions? What diversity of policy is sufficient to prevent synchronisation? This is possibly the most important open question in the field, and it cannot be answered by studying single firms. Human agent teaming under stress. Fugener and colleagues (2022) studied delegation in controlled settings. Supply chain crises are not controlled settings. Field or experimental work is needed on how planners actually behave when an agent recommends an unpopular action under time pressure, and on when they override correctly versus incorrectly. Evaluation standards for operational agents. The field currently has no shared benchmark for agent performance on supply chain tasks. A public, disruption rich benchmark environment, analogous to those that accelerated progress in other AI domains, would allow claims to be compared rather than merely asserted. Explainability that planners actually use. Ali and colleagues (2023) survey explainability techniques, but few are designed for the specific question a planner asks, which is not how does the model work but why this option and not the one I would have chosen. Domain specific contrastive explanation is an open problem. Legal and accountability structures. Who is liable when an agent cancels an order that turns out to have been essential? Work at the intersection of operations, law, and AI governance is needed, and it should proceed faster than deployment, not behind it. Sustainability constrained autonomy. Given Sarkis (2021), research is needed on how emissions constraints change agentic behaviour under disruption, and on whether resilience and decarbonisation can be jointly optimised or must be traded. Small and medium enterprise access. Almost all of the capability requirements in Section 5.6 assume resources that large firms have and smaller ones do not. If agentic resilience is available only to the largest firms, network level fragility may increase, since the weakest nodes will become relatively weaker. This distributional question deserves serious attention from #policy_makers. 9. Conclusion Global supply chains now operate in an environment where disruption is normal rather than exceptional. Geopolitical fragmentation, technology controls, contested shipping lanes, and climate driven logistics failures are not passing turbulence. They are the operating conditions. Networks built for cost efficiency in a stable world must now perform in an unstable one, and they must do so without the option of unlimited buffers. This article has argued that Agentic AI represents a real change in what is possible, but a narrower one than the enthusiasm surrounding it suggests. The change is in the decision loop. Agentic systems can perceive more widely, remember more reliably, simulate more often, and act more quickly than human teams working through weekly planning cycles. Where a disruption is observable, where the remedy can be executed through a system interface, and where the action is reversible, autonomy delivers genuine value, and it does so most where redundancy is thinnest. That is the core of the argument. But the limits are equally real. Agents cannot qualify a supplier, obtain a licence, negotiate with a state, or build a factory. In the geopolitical domain, where the most consequential disruptions now originate, their contribution is exposure discovery, monitoring, simulation, and documentation, not resolution. Their performance depends entirely on a data foundation that most organisations do not yet have, which means the binding constraint on agentic resilience is not intelligence but plumbing. And autonomy creates its own hazards. Correlated agents can synchronise markets. Automated cancellations can cascade upstream faster than humans can intervene. Learned policies can fail silently. External text can be weaponised against systems that read it. Accountability can dissolve if it is not designed in from the start. These are not reasons to avoid agentic systems. They are reasons to govern them, through graduated autonomy, hard constraints enforced outside the model, complete auditability, continuous evaluation, and deliberate friction before irreversible action. The most useful way to think about Agentic AI in this context is therefore not as a replacement for human judgement but as an extension of organisational attention and reaction speed, bounded by rules that humans set and can revoke. The firms that benefit will be those that fix their data first, start in read only mode, expand autonomy by decision class rather than by ambition, and keep their people close enough to the system to know when it is wrong. The networks that benefit will be those that develop shared standards and shared rules of engagement, because #supply_chain_resilience, in the end, is a property of the network and not of any single firm within it. Hashtags #AgenticAI #Agentic_AI_in_Supply_Chains #Global_Supply_Chain_Resilience #Autonomous_Supply_Chains #Geopolitical_Risk #Logistics_Disruption #Self_Healing_Supply_Chains #AI_Agents #Supply_Chain_Research #Resilient_Operations #Supply_Chain_Digital_Twins #Autonomous_Decision_Making #Supply_Chain_Technology #Operations_Management #Future_of_Logistics References Acharya, D. B., Kuppan, K., and Divya, B. (2025). Agentic AI: Autonomous intelligence for complex goals, a comprehensive survey. IEEE Access, 13, 18912 to 18936. Aldrighetti, R., Battini, D., Ivanov, D., and Zennaro, I. (2021). Costs of resilience and disruptions in supply chain network design models: A review and future research directions. International Journal of Production Economics, 235, 108103. Ali, S., Abuhmed, T., El-Sappagh, S., Muhammad, K., Alonso-Moral, J. M., Confalonieri, R., Guidotti, R., Del Ser, J., Diaz-Rodriguez, N., and Herrera, F. (2023). Explainable Artificial Intelligence (XAI): What we know and what is left to attain trustworthy artificial intelligence. Information Fusion, 99, 101805. Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S., and others (2021). On the opportunities and risks of foundation models. Center for Research on Foundation Models, Stanford University, Technical Report. Boute, R. N., Gijsbrechts, J., van Jaarsveld, W., and Vanvuchelen, N. (2022). Deep reinforcement learning for inventory control: A roadmap. European Journal of Operational Research, 298(2), 401 to 412. Brintrup, A., Pak, J., Ratiney, D., Pearce, T., Wichmann, P., Woodall, P., and McFarlane, D. (2020). Supply chain data analytics for predicting supplier disruptions: A case study in complex asset manufacturing. International Journal of Production Research, 58(11), 3330 to 3341. Chan, A., Salganik, R., Markelius, A., Pang, C., Rajkumar, N., Krasheninnikov, D., and others (2023). Harms from increasingly agentic algorithmic systems. Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency (FAccT), 651 to 666. Choi, T. Y., Netland, T. H., Sanders, N., Sodhi, M. S., and Wagner, S. M. (2023). Just in time for supply chains in turbulent times. Production and Operations Management, 32(7), 2331 to 2340. Cohen, M. A., Cui, S., Doetsch, S., Ernst, R., Huchzermeier, A., Kouvelis, P., Lee, H. L., Matsuo, H., and Tsay, A. A. (2022). Bespoke supply chain resilience: The gap between theory and practice. Journal of Operations Management, 68(5), 515 to 531. Craighead, C. W., Ketchen, D. J., and Darby, J. L. (2020). Pandemics and supply chain management research: Toward a theoretical toolbox. Decision Sciences, 51(4), 838 to 866. Dolgui, A., and Ivanov, D. (2021). Ripple effect and supply chain disruption management: New trends and research directions. International Journal of Production Research, 59(1), 102 to 109. Dolgui, A., and Ivanov, D. (2022). 5G in digital supply chain and operations management: Fostering flexibility, end to end connectivity and real time visibility through internet of everything. International Journal of Production Research, 60(2), 442 to 451. Dubey, R., Bryde, D. J., Dwivedi, Y. K., Graham, G., and Foropon, C. (2022). Impact of artificial intelligence driven big data analytics culture on agility and resilience in humanitarian supply chain: A practice based view. International Journal of Production Economics, 250, 108618. Dwivedi, Y. K., Kshetri, N., Hughes, L., Slade, E. L., Jeyaraj, A., Kar, A. K., and others (2023). Opinion paper: So what if ChatGPT wrote it? Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, practice and policy. International Journal of Information Management, 71, 102642. European Union (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union. Fragapane, G., Ivanov, D., Peron, M., Sgarbossa, F., and Strandhagen, J. O. (2022). Increasing flexibility and productivity in Industry 4.0 production networks with autonomous mobile robots and smart intralogistics. Annals of Operations Research, 308, 125 to 143. Free, C., and Hecimovic, A. (2021). Global supply chains after COVID-19: The end of the road for neoliberal globalisation? Accounting, Auditing and Accountability Journal, 34(1), 58 to 84. Fugener, A., Grahl, J., Gupta, A., and Ketter, W. (2022). Cognitive challenges in human artificial intelligence collaboration: Investigating the path toward productive delegation. Information Systems Research, 33(2), 678 to 696. Gabriel, I., Manzini, A., Keeling, G., Hendricks, L. A., Rieser, V., Iqbal, H., and others (2024). The ethics of advanced AI assistants. Google DeepMind Technical Report. Gereffi, G., Pananond, P., and Pedersen, T. (2022). Resilience decoded: The role of firms, global value chains, and the state in COVID-19 capabilities. California Management Review, 64(2), 46 to 70. Gijsbrechts, J., Boute, R. N., Van Mieghem, J. A., and Zhang, D. J. (2022). Can deep reinforcement learning improve inventory management? Performance on lost sales, dual sourcing, and multi echelon problems. Manufacturing and Service Operations Management, 24(3), 1349 to 1368. Glikson, E., and Woolley, A. W. (2020). Human trust in artificial intelligence: Review of empirical research. Academy of Management Annals, 14(2), 627 to 660. Gupta, S., Modgil, S., Meissonier, R., and Dwivedi, Y. K. (2021). Artificial intelligence and information system resilience to cope with supply chain disruption. IEEE Transactions on Engineering Management, 71, 10496 to 10506. Hagele, S., Grosse, E. H., and Ivanov, D. (2023). Supply chain resilience: A tertiary study. International Journal of Systems Science: Operations and Logistics, 10(1), 2106936. Handfield, R. B., Graham, G., and Burns, L. (2020). Corona virus, tariffs, trade wars and supply chain evolutionary design. International Journal of Operations and Production Management, 40(10), 1649 to 1660. Hendrycks, D., Mazeika, M., and Woodside, T. (2023). An overview of catastrophic AI risks. Center for AI Safety, Technical Report. Ivanov, D. (2021). Supply chain viability and the COVID-19 pandemic: A conceptual and formal generalisation of four major adaptation strategies. International Journal of Production Research, 59(12), 3535 to 3552. Ivanov, D. (2022). Viable supply chain model: Integrating agility, resilience and sustainability perspectives, lessons from and thinking beyond the COVID-19 pandemic. Annals of Operations Research, 319, 1411 to 1431. Ivanov, D. (2023). Intelligent digital twin (iDT) for supply chain stress testing, resilience, and viability. International Journal of Production Economics, 263, 108938. Ivanov, D. (2023). The Industry 5.0 framework: Viability based integration of the resilience, sustainability, and human centricity perspectives. International Journal of Production Research, 61(5), 1683 to 1695. Ivanov, D., and Dolgui, A. (2020). Viability of intertwined supply networks: Extending the supply chain resilience angles towards survivability. A position paper motivated by COVID-19 outbreak. International Journal of Production Research, 58(10), 2904 to 2915. Ivanov, D., and Dolgui, A. (2021). A digital supply chain twin for managing the disruption risks and resilience in the era of Industry 4.0. Production Planning and Control, 32(9), 775 to 788. Jackson, I., Ivanov, D., Dolgui, A., and Namdar, J. (2024). Generative artificial intelligence in supply chain and operations management: A capability based framework for analysis and implementation. International Journal of Production Research, 62(17), 6120 to 6145. Kamalahmadi, M., Shekarian, M., and Mellat Parast, M. (2022). The impact of flexibility and redundancy on improving supply chain resilience to disruptions. International Journal of Production Research, 60(6), 1992 to 2020. Katsaliaki, K., Galetsi, P., and Kumar, S. (2022). Supply chain disruptions and resilience: A major review and future research agenda. Annals of Operations Research, 319, 965 to 1002. Kinra, A., Ivanov, D., Das, A., and Dolgui, A. (2020). Ripple effect quantification by supplier risk exposure assessment. International Journal of Production Research, 58(18), 5559 to 5578. Kosasih, E. E., and Brintrup, A. (2022). A machine learning approach for predicting hidden links in supply chain with graph neural networks. International Journal of Production Research, 60(17), 5380 to 5393. Kshetri, N. (2021). Blockchain and sustainable supply chain management in developing countries. International Journal of Information Management, 60, 102376. Miller, C. (2022). Chip War: The Fight for the World's Most Critical Technology. Scribner, New York. Modgil, S., Gupta, S., Stekelorum, R., and Laguir, I. (2022). AI technologies and their impact on supply chain resilience during COVID-19. International Journal of Physical Distribution and Logistics Management, 52(2), 130 to 149. National Institute of Standards and Technology (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). NIST, United States Department of Commerce. Naz, F., Kumar, A., Majumdar, A., and Agrawal, R. (2022). Is artificial intelligence an enabler of supply chain resiliency post COVID-19? An exploratory state of the art review for future research. Operations Management Research, 15, 378 to 398. Oroojlooyjadid, A., Nazari, M., Snyder, L. V., and Takac, M. (2022). A deep Q network for the beer game: Deep reinforcement learning for inventory optimization. Manufacturing and Service Operations Management, 24(1), 285 to 304. Panwar, R., Pinkse, J., and De Marchi, V. (2022). The future of global supply chains in a post COVID-19 world. California Management Review, 64(2), 5 to 23. Park, J. S., O'Brien, J., Cai, C. J., Morris, M. R., Liang, P., and Bernstein, M. S. (2023). Generative agents: Interactive simulacra of human behavior. Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology (UIST), 1 to 22. Pournader, M., Ghaderi, H., Hassanzadegan, A., and Fahimnia, B. (2021). Artificial intelligence applications in supply chain management. International Journal of Production Economics, 241, 108250. Rolf, B., Jackson, I., Muller, M., Lang, S., Reggelin, T., and Ivanov, D. (2023). A review on reinforcement learning algorithms and applications in supply chain management. International Journal of Production Research, 61(20), 7151 to 7179. Roscoe, S., Aktas, E., Petersen, K. J., Skipworth, H. D., Handfield, R. B., and Habib, F. (2022). Redesigning global supply chains during compounding geopolitical disruptions: The role of supply chain logics. International Journal of Operations and Production Management, 42(9), 1407 to 1434. Russell, S., and Norvig, P. (2021). Artificial Intelligence: A Modern Approach, 4th edition. Pearson, Harlow. Sarkis, J. (2021). Supply chain sustainability: Learning from the COVID-19 pandemic. International Journal of Operations and Production Management, 41(1), 63 to 73. Schick, T., Dwivedi-Yu, J., Dessi, R., Raileanu, R., Lomeli, M., Zettlemoyer, L., Cancedda, N., and Scialom, T. (2023). Toolformer: Language models can teach themselves to use tools. Advances in Neural Information Processing Systems, 36. Sharma, R., Shishodia, A., Gunasekaran, A., Min, H., and Munim, Z. H. (2022). The role of artificial intelligence in supply chain management: Mapping the territory. International Journal of Production Research, 60(24), 7527 to 7550. Shavit, Y., Agarwal, S., Brundage, M., Adler, S., O'Keefe, C., Campbell, R., and others (2023). Practices for governing agentic AI systems. OpenAI Technical Report. Sheffi, Y. (2023). The Magic Conveyor Belt: Supply Chains, A.I., and the Future of Work. MIT CTL Media, Cambridge, Massachusetts. Shih, W. C. (2020). Global supply chains in a post pandemic world. Harvard Business Review, 98(5), 82 to 89. Shinn, N., Cassano, F., Gopinath, A., Narasimhan, K., and Yao, S. (2023). Reflexion: Language agents with verbal reinforcement learning. Advances in Neural Information Processing Systems, 36. Sodhi, M. S., and Tang, C. S. (2021). Supply chain management for extreme conditions: Research opportunities. Journal of Supply Chain Management, 57(1), 7 to 16. Spieske, A., and Birkel, H. (2021). Improving supply chain resilience through Industry 4.0: A systematic literature review under the impressions of the COVID-19 pandemic. Computers and Industrial Engineering, 158, 107452. Wang, L., Ma, C., Feng, X., Zhang, Z., Yang, H., Zhang, J., and others (2024). A survey on large language model based autonomous agents. Frontiers of Computer Science, 18(6), 186345. Wichmann, P., Brintrup, A., Baker, S., Woodall, P., and McFarlane, D. (2020). Extracting supply chain maps from news articles using deep neural networks. International Journal of Production Research, 58(17), 5320 to 5336. Wieland, A., and Durach, C. F. (2021). Two perspectives on supply chain resilience. Journal of Business Logistics, 42(3), 315 to 322. Xi, Z., Chen, W., Guo, X., He, W., Ding, Y., Hong, B., and others (2025). The rise and potential of large language model based agents: A survey. Science China Information Sciences, 68(2), 121101. Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K., and Cao, Y. (2023). ReAct: Synergizing reasoning and acting in language models. Proceedings of the 11th International Conference on Learning Representations (ICLR). Zamani, E. D., Smyth, C., Gupta, S., and Dennehy, D. (2023). Artificial intelligence and big data analytics for supply chain resilience: A systematic literature review. Annals of Operations Research, 327, 605 to 632.

Managers who work across borders today are asked to do something that used to belong to embassies. They must read political signals, build relationships with regulators, calm worried communities, and still close commercial deals on time. This article studies how #diplomacy has moved from the state into the firm, and how it now sits at the centre of #international_management. It brings together three streams of scholarship that are usually kept apart: work on #business_diplomacy and #nonmarket_strategy, work on #cross_cultural_negotiation, and work on #regulatory_frameworks and institutions. The article uses an integrative literature review of research published mainly between 2021 and 2025, and it applies the findings to the setting of #global_hubs such as Dubai, Singapore, London, and Hong Kong, where firms from many legal traditions meet inside a single square kilometre. The article proposes a three pillar model of diplomatic capability made up of relational competence, cultural competence, and regulatory competence, joined by two connecting processes: sense making and #boundary_spanning. The discussion shows that cultural knowledge alone does not produce good outcomes at the table, that regulatory knowledge alone does not produce legitimacy, and that firms which separate these two capabilities into different departments tend to negotiate badly and comply expensively. The article closes with practical guidance for managers, a set of teaching implications for business students, and an agenda for future research. The central claim is simple. In a fragmented world economy, diplomatic skill is no longer a soft extra. It is a condition of survival for the #multinational_enterprises that operate across competing rule systems. Keywords: business diplomacy, cross-cultural negotiation, regulatory frameworks, global hubs, nonmarket strategy, cultural intelligence, international management, multinational enterprises 1. Introduction For most of the twentieth century, the word diplomacy belonged to governments. Ambassadors negotiated treaties, consuls protected citizens, and firms simply followed the trade rules that states agreed among themselves. That division of labour has broken down. A large firm today may employ more people in a host country than the embassy of its home state, may hold more useful data on local supply conditions than any ministry, and may be the target of political anger long before any government is. When a firm enters a new market, it does not only enter a market. It enters a political community, a legal order, a media environment, and a web of expectations about what a good corporate citizen looks like. This shift has produced a body of research on #business_diplomacy and #corporate_diplomacy, which studies how firms manage relationships with governments, regulators, civil society, media, and communities in order to protect and advance commercial interests. Doh, Dahan, and Casario (2022) position international business diplomacy inside the wider field of #nonmarket_strategy and argue that the practices a firm should use depend on two features of the governance regime it faces: the level at which rules are made, and how formal those rules are. Alammar and Pauleen (2022), working from interviews with diplomats and senior managers, describe business diplomacy as a set of integrated qualities used to manage and influence multiple stakeholders in a systematic way. Both studies point in the same direction. Diplomacy inside the firm is not public relations with a nicer name. It is a management function with its own logic, its own skills, and its own measurable outcomes. At the same time, the practical work of international managers has become harder. Global rule making has fragmented. Trade blocs are drifting apart. Technology policy has become a matter of national security, as Luo and Van Assche (2023) show in their study of the United States CHIPS and Science Act, which turned semiconductor supply into an instrument of statecraft. Cui, Vertinsky, Wang, and Zhou (2023) describe #decoupling as a new vulnerability of globalisation and set out the response strategies that firms adopt when the ground under a supply chain moves. Witt, Lewin, Li, and Gaur (2023) show that decoupling is not only rhetoric but a measurable pattern in trade and investment data. Devinney, Hartwell, Oetzel, and Vaaler (2023) argue that sociopolitical uncertainty is now a permanent feature of the environment rather than a passing storm. Managers cannot wait it out. Three consequences follow, and they organise this article. First, cross border negotiation has become more difficult, because negotiators now carry political baggage that has nothing to do with their firm. A Chinese engineer negotiating a data centre contract in Europe, or a European bank negotiating a licence in the Gulf, must handle suspicion that is aimed at a passport rather than a person. #cross_cultural_negotiation was always a matter of communication styles, time horizons, and hierarchy. It is now also a matter of geopolitics. Second, regulatory work has become strategic rather than administrative. #compliance used to be a back office cost. Today the choice of where to book a transaction, where to store data, and which court will hear a dispute can decide whether a business model works at all. #regulatory_frameworks are not a constraint that managers meet at the end of a deal. They are part of the deal. Third, #global_hubs have become the places where these pressures concentrate. Cities such as Dubai, Singapore, London, and Hong Kong host firms from dozens of legal traditions inside small districts with their own courts, their own regulators, and their own rulebooks. Dubai offers a clear example. The Dubai International Financial Centre was created in 2004 as a financial free zone under federal law, with an independent regulator, its own courts, and a legal framework based on principles of #common_law rather than on the civil law of the wider United Arab Emirates. A manager working there may negotiate in Arabic and English, apply an English style contract, sit before a court staffed by judges from several jurisdictions, and still answer to a group head office in Frankfurt or Mumbai. Hubs make the tensions of the world economy visible at street level. 1.1 Research questions The article addresses four questions. How has the concept of #diplomacy been transferred from the state to the firm, and what does the current literature say it consists of? How do cultural factors shape #negotiation_strategy in cross border settings, and where does the standard cultural dimensions approach fail? How do managers navigate overlapping and sometimes contradictory #regulatory_frameworks, and what capabilities does this require? What is distinctive about #global_hubs as a setting, and what can they teach us about diplomatic capability in general? 1.2 Contribution The contribution of this article is integrative rather than empirical. Research on business diplomacy sits mainly in international business journals. Research on negotiation sits mainly in organisational behaviour and psychology journals. Research on regulation sits mainly in law, political economy, and public policy. These fields rarely read one another. Practising managers, however, experience them as a single problem. This article joins them into one framework and shows why capability in one pillar without the others produces poor outcomes. It is written for students, and so it explains the concepts plainly, but it keeps the structure and the evidential standards of a research article. 2. Literature Review 2.1 From state diplomacy to firm diplomacy Diplomacy in its classical sense is the management of relations between states through negotiation, representation, and reporting. Jazbec (2022) traces the intersection between diplomatic practice and management practice and notes that both are, at heart, about achieving objectives through other people whose interests only partly overlap with your own. That definition travels easily into business. The transfer happened for structural reasons rather than fashionable ones. As firms internationalised, they came to depend on outcomes they could not buy: a licence, a permit, a favourable tax ruling, a community that would not block a plant, a media environment that would not turn a small incident into a crisis. Classic market strategy has little to say about these outcomes. nonmarket strategy emerged to fill that gap, and #business_diplomacy is the part of nonmarket strategy that deals with relationships rather than with rules alone. Doh et al. (2022) offer the most careful conceptualisation available. They treat international business diplomacy as a set of practices, not as a personality trait or a department. Which practices are appropriate depends on two dimensions of the institutional environment. The first is the level of governance, meaning whether the relevant rules are made locally, nationally, regionally, or globally. The second is the formality of the rules, meaning whether they are written and enforceable or informal and normative. A firm dealing with a formal national regulator will use technical submissions, licensing dossiers, and #lobbying through registered channels. A firm dealing with an informal local coalition of community leaders will use dialogue, presence, patience, and visible investment. Using the wrong practice in the wrong place is a common and expensive mistake. Alammar and Pauleen (2022) approach the same object from the inside, through grounded theory built on interviews with twenty one diplomats, chief executives, and business people. Their integrative framework treats business diplomacy as the professional and systematic management of multiple stakeholders in order to advance business interests. Two findings from that study deserve emphasis for students. The first is that respondents saw diplomacy as a learned skill rather than a natural charm. The second is that they saw it as an organisational property, not merely an individual one. A firm can be diplomatically capable even when individual managers are ordinary, provided that its systems capture political intelligence and act on it. Salvi and Ruel (2022) connect #corporate_diplomacy to sustainable internationalisation, arguing that diplomatic engagement with communities and regulators is one of the mechanisms through which firms build the #legitimacy they need for long horizon investments. Bucheli, Duran, and Kim (2024) take a historical route to the same conclusion. Using archival evidence, they show that multinationals have long conducted diplomatic activities and that these activities have both protective and predatory faces. Their work is a useful corrective to any romantic view of the subject. Diplomacy can be used to defend a firm against unfair treatment. It can also be used to capture regulators and to entrench advantage. Students should hold both possibilities in mind. Meyer (2023), studying elite financiers and senior officials, shows how the relationship works in practice at the top of the system: information moves in both directions, and the value of the relationship for the official is often the intelligence it brings rather than any favour granted. This is an important point. Effective #stakeholder_engagement is rarely transactional in a crude sense. It is informational. 2.2 What business diplomacy actually consists of Reading across these studies, six recurring elements appear. Representation. Speaking for the firm in a way that is credible to an external audience. This is harder than it sounds, because the audience often distrusts the messenger before a word is spoken. Intelligence gathering. Collecting and interpreting political, regulatory, and social information early enough to act on it. Most corporate crises are not surprises. They are signals that nobody was reading. Negotiation. Reaching agreements with parties who have different interests and who are not customers or suppliers. Regulators, ministries, unions, and local councils fall into this group. Coalition building. Working with industry associations, chambers of commerce, and even competitors to shape rules. Firms rarely have the standing to move a regulator alone. Reputation and communication. Managing what the firm means to people who will never buy from it. #reputation_management belongs to diplomacy because reputation is the currency in which regulatory goodwill is paid. Ethical judgement. Deciding where influence ends and capture begins. This is the boundary that separates #ethical_leadership from corruption, and it is often a thin line in practice. 2.3 Cross-cultural negotiation: the standard model and its limits The standard model taught in most business schools rests on national cultural dimensions. Managers learn that some cultures are high in #power_distance and some are low, that some use #high_context_communication in which meaning sits in the situation and the relationship, and some use #low_context_communication in which meaning sits in the explicit words. They learn that in some settings #face_saving is central and that direct contradiction destroys value, while in others directness is read as honesty. This vocabulary is useful. It gives students a way to notice differences they would otherwise dismiss as rudeness or incompetence. The problem is that the model is often applied badly. Three failures are common. The first is stereotyping. Dimensions describe averages across large populations. They do not describe the person sitting across the table, who may be a Lebanese engineer educated in France working for a Japanese firm in the Gulf. Treating a national average as an individual prediction is a basic statistical error, and it produces negotiators who perform culture at their counterparts instead of listening to them. The second is the assumption of a single culture per person. Khakhar, Rammal, and Pereira (2023), studying bicultural senior managers, show that #biculturals move between cultural frames rather than staying inside one. #cultural_frame_switching allows a negotiator to shift mindset, language, and register depending on the moment, and their study finds that this capacity is an asset in international negotiation rather than a source of confusion. The implication is significant. The most valuable negotiator in a hub city may be precisely the person who does not fit neatly into any national category. The third is the neglect of situation. Negotiation behaviour responds to power, deadlines, alternatives, and the structure of the deal at least as much as it responds to culture. Lewicki, Barry, and Saunders (2021) set out the standard distinction between #distributive_bargaining, in which parties divide a fixed pie, and #integrative_bargaining, in which parties search for trades that expand it. Whether a negotiator behaves distributively depends heavily on whether they believe a relationship will continue, on how strong their #BATNA is, and on what their principals will reward. A negotiator with no alternative and a hostile board will fight hard regardless of nationality. 2.4 Cultural intelligence as a bridge Because the dimensions model is blunt, researchers have turned to individual capability. #cultural_intelligence, usually shortened to CQ, describes the capability to function effectively in culturally diverse situations. Semenov and Randrianasolo (2024) offer an extended conceptualisation and examine the relationships between its dimensions, noting that some elements of CQ may themselves be culture specific rather than universal. Ng, Ang, and Rockstuhl (2022) place the construct in the longer history of research on intelligence and stress that CQ is a distinct interpersonal capability aimed at bridging cultural difference rather than at knowing facts about cultures. The practical value of CQ for negotiation is that it shifts attention from content to process. A manager with high CQ does not need to memorise a list of taboos. She notices that her counterpart has gone quiet, forms a hypothesis about why, tests it gently, and adjusts. That is a metacognitive skill, not an encyclopaedic one. Chande (2024), reviewing cross cultural negotiation strategies in international joint ventures, reports that adaptation, empathy, active listening, and flexibility improve the chance of durable partnerships, while ethnocentric tactics increase the risk of breakdown. The list is unsurprising, but the mechanism is worth stating: adaptation works because it generates information, and information is what turns a distributive fight into an integrative search. 2.5 Trust, relationship capital, and time Across the negotiation literature, #trust_building appears as the variable that converts cultural skill into commercial outcome. Trust is not a feeling in this context. It is a judgement about predictability and goodwill that reduces the cost of contracting. Where trust is high, parties disclose interests, accept ambiguity in contracts, and resolve problems informally. Where trust is low, every clause must be litigated in advance. Cultures differ in how trust is produced. In some settings, competence and delivery generate trust quickly and relationships follow. In others, relationships must be established before any serious business is discussed, and attempts to speed this up read as disrespect. The mistake made by managers from fast trust cultures is to treat relationship building as a delay. The mistake made by managers from slow trust cultures is to treat contractual precision as suspicion. Both are misreadings, and both destroy value. #relationship_capital accumulates slowly and depreciates quickly. This asymmetry has a practical consequence that is under taught: the time to build a regulatory relationship is before you need it. A firm that appears at a ministry for the first time on the day it has a problem is negotiating from the weakest possible position. 2.6 Virtual and digital negotiation The move of negotiation onto screens has changed the terrain. Cuhlova and Demel (2024) study intercultural negotiation in the digital age and find that although online platforms reduce cost, experienced negotiators still prefer face to face meetings when cultural nuance matters. Agamennone (2022) examines the rise of the video call negotiation and the cross cultural variations that appeared during the pandemic period. The underlying explanation is straightforward. #virtual_negotiation strips out the informal channels through which relationship and repair usually happen: the walk to the lift, the shared meal, the aside in the corridor. It also flattens status signals and makes silence ambiguous. In high context settings, where a great deal of meaning is carried by pause, posture, and the order in which people speak, this loss is severe. The practical response is not to refuse video calls but to rebuild the missing channels deliberately, through pre meetings, bilateral calls before plenary sessions, and explicit summaries after each session. 2.7 Regulatory frameworks and the institutional environment Regulation is where diplomacy becomes technical. Three bodies of work matter here. #institutional_theory tells us that firms need legitimacy as well as efficiency, and that legitimacy is granted by an environment with rules, both formal and informal. Where formal institutions are weak, firms encounter #institutional_voids, and they must either build substitutes or rely on relationships to fill the gap. Relationships that substitute for institutions are efficient in the short run and dangerous in the long run, because they are personal and therefore fragile. The literature on political connections shows that closeness to power is a double edged asset. Huang, Tsui-Auch, Dieleman, and Gomulya (2024), reviewing politically connected emerging market multinationals in a deglobalising world, show that connections that help at home can become a liability abroad, where they are read as evidence of state control. Shirodkar, Liedong, Rajwani, and Lawton (2024) argue for a values based approach to nonmarket strategy precisely because the complexity of the current environment makes purely instrumental approaches unstable. The literature on geopolitics tells us that the rules themselves have become instruments of competition. Luo and Van Assche (2023) describe techno-geopolitical uncertainty. Ren and Gao (2023) show how subsidiaries of Chinese state owned multinationals adapt their nonmarket strategies under selective de-globalisation. Fjellstrom, Bai, Oliveira, and Fang (2023) examine springboard internationalisation under geopolitical tension, showing that firms use third countries as staging posts when direct routes are blocked. This last mechanism is directly relevant to hubs, which frequently serve as the neutral ground on which otherwise blocked relationships are conducted. 2.8 Global hubs as a distinct setting Hub cities are not simply large markets. They have four features that make them analytically distinct. Legal layering. A hub often contains a special jurisdiction inside a national one. Dubai contains the Dubai International Financial Centre, a financial free zone established under federal law with its own regulator, its own courts, and a rulebook drawn from #common_law. The wider United Arab Emirates operates a civil law system. Both apply within a few kilometres of each other. Singapore, Hong Kong, and London each offer their own version of layered legal capacity, whether through specialised commercial courts, arbitration centres, or regulatory sandboxes. Concentration of intermediaries. Hubs concentrate law firms, arbitrators, auditors, consultants, and bankers. These intermediaries are not neutral. They shape how a deal is framed, which law is chosen, and which forum will hear a dispute. Density of counterparties. In a hub, a negotiator meets counterparts from many cultures within a single week. This makes the single culture model of negotiation almost useless and rewards the frame switching capacity described by Khakhar et al. (2023). Regulatory competition. Hubs compete with one another for firms. This produces a form of #regulatory_arbitrage in which firms can choose their regulator to a meaningful degree. Competition can raise standards, because a hub that becomes a haven loses its access to the global financial system. It can also lower them, because the pressure to attract business is constant. 3. Conceptual Framework: The Three Pillar Model of Diplomatic Capability The literature above suggests that diplomatic capability in international management rests on three pillars, joined by two connecting processes. The model is presented here as an organising device rather than as a tested causal claim. 3.1 Pillar one: relational competence Relational competence is the capacity to build and maintain productive relationships with parties whose interests differ from yours and who cannot be commanded. It includes the ability to identify who actually matters, which is often not the person with the most senior title, and the ability to sustain contact when there is nothing to ask for. Its components are stakeholder mapping, listening, credibility, patience, and reciprocity. Its failure mode is transactionalism, meaning the habit of appearing only when something is needed. Its measurable output is access, which is to say the ability to get a meeting and be heard before a decision is taken rather than after. 3.2 Pillar two: cultural competence Cultural competence is the capacity to interpret and adapt behaviour across cultural settings. Its components are knowledge, awareness, motivation, and behavioural flexibility, corresponding roughly to the dimensions of #cultural_intelligence in the literature. Its failure mode is the twin error described earlier: stereotyping on one side, and cultural blindness on the other. Its measurable output is accuracy, meaning the ability to predict how a counterpart will interpret a given move. 3.3 Pillar three: regulatory competence Regulatory competence is the capacity to understand, anticipate, and work within the rule systems that apply to the firm, including rules made in the home country, the host country, and by international bodies. It includes knowing which regulator has jurisdiction, what that regulator is worried about, and how it prefers to be approached. Its components are technical knowledge, forecasting, submission craft, and internal control. Its failure mode is the treatment of #compliance as a checklist exercise owned by lawyers alone. Its measurable output is predictability, meaning the ability to know in advance whether a proposed action will be permitted. 3.4 Connecting process one: sense making The three pillars do not add up on their own. Sense making is the process by which scattered signals from relationships, cultural interactions, and regulatory contacts are turned into an interpretation that leadership can act on. Most firms collect signals. Few interpret them, because interpretation requires someone to be paid to be wrong occasionally. 3.5 Connecting process two: boundary spanning #boundary_spanning is the work of translating between the firm and its environment, in both directions. The boundary spanner explains the regulator to the board and the board to the regulator. Biculturals and bilingual managers are natural candidates for this role, but the role must exist in the organisation chart or the capacity is wasted. 3.6 Propositions From the model, four propositions follow. They are offered as testable statements for future research. Proposition 1. Firms that possess cultural competence without regulatory competence will negotiate agreements that are warmly received and subsequently blocked. Proposition 2. Firms that possess regulatory competence without relational competence will achieve technical compliance while losing discretionary decisions, because discretion follows trust. Proposition 3. Firms that possess relational competence without cultural competence will build strong ties with a narrow group of counterparts who resemble them, and will be blind to the rest of the environment. Proposition 4. In #global_hubs, the returns to boundary spanning capacity are higher than in single culture markets, because the density and variety of counterparties increases the value of translation. 4. Methodology This article is a conceptual and integrative literature review. It does not report new primary data. The purpose of an integrative review is to synthesise findings across fields that do not normally speak to one another and to propose a framework that later empirical work can test. 4.1 Search and selection Literature was identified through structured searches of major management, international business, and negotiation journals, using combinations of the terms business diplomacy, corporate diplomacy, nonmarket strategy, cross cultural negotiation, cultural intelligence, regulatory frameworks, geopolitical risk, decoupling, and international financial centres. Priority was given to work published from 2021 onward in order to capture the post-pandemic and post-decoupling environment, on the reasoning that the institutional conditions facing managers changed materially in that period. Sources were included where they met three criteria: they appeared in peer reviewed outlets or in scholarly books, they addressed the interaction between firms and their political, cultural, or regulatory environment, and they contained either empirical evidence or explicit theory building. Purely descriptive practitioner commentary was excluded. 4.2 Analytical approach Selected works were read and coded thematically. Codes were grouped into the three pillars described in the previous section, and the connecting processes were derived from material that did not fit cleanly into any single pillar. This is an interpretive procedure, and a different reader could reasonably group the material differently. 4.3 Limitations of the method Three limitations should be stated openly. First, an integrative review inherits the biases of the literature it reviews, and the literature on international business is skewed toward large firms and toward English language publication. Second, the framework is proposed rather than tested. Third, the treatment of global hubs relies on institutional description rather than on comparative empirical measurement, because comparable data across hub jurisdictions is not readily available. 5. Findings and Discussion 5.1 Culture matters, but not in the way most training suggests The first finding is that cultural knowledge is necessary but weakly predictive on its own. Across the studies reviewed, culture consistently shapes the process of negotiation, meaning the pace, the sequencing, the tolerance for silence, the role of hierarchy, and the interpretation of directness. It shapes outcomes far less consistently, because outcomes are also driven by power, alternatives, and the mandate that each negotiator carries into the room. This distinction has been blurred in much corporate training, which teaches culture as though it were a code to be cracked. Once cracked, the promise goes, deals will follow. That is not what the evidence supports. Chande (2024) finds that adaptive strategies improve joint venture success, but adaptation is a behaviour, not a lookup table. Khakhar et al. (2023) find that #biculturals succeed because they can switch frames, which is the opposite of applying a fixed rule. The practical implication for #intercultural_communication training is that firms should spend less time on country briefings and more time on metacognitive practice: forming hypotheses about what a counterpart means, testing them, and revising. Country briefings are still useful as a starting hypothesis. They are dangerous as a conclusion. 5.2 The negotiation table is now a political site The second finding is that negotiations across borders now carry political content that the negotiators themselves did not create. A negotiator representing a firm from a country under sanctions, or from a country whose technology is viewed with suspicion, arrives with a handicap that no amount of cultural intelligence will remove. Luo and Van Assche (2023) show how industrial policy in the technology sector has reframed ordinary commercial decisions as questions of national security. Huang et al. (2024) show that political connections, which are an asset in the home market, can convert into a liability abroad. Three consequences follow for practice. First, the identity of the negotiator has become a strategic variable. Firms increasingly staff sensitive negotiations with managers whose passports and career histories reduce political noise. This is uncomfortable, and it raises fair questions about discrimination, but it is happening and students should understand why. Second, ownership and governance structures are now part of the negotiation. Counterparties and regulators ask who ultimately controls the firm, and a satisfactory answer must be prepared before the first meeting, not improvised during it. Third, the timeline has lengthened. Where a deal touches critical infrastructure, data, or advanced technology, regulatory review can take longer than the commercial negotiation itself. #policy_uncertainty must be priced into the deal, and the party that ignores this ends up paying for it through break fees and stranded costs. 5.3 Regulation is a negotiation, not a wall The third finding is that managers systematically misunderstand the nature of regulation. Regulation appears to them as a wall: fixed, external, and to be complied with or avoided. In practice, regulation is produced by institutions staffed by people who have objectives, resources, reputations, and fears. It is therefore negotiable, not in the corrupt sense of buying an outcome, but in the ordinary sense that rules are drafted, interpreted, applied with discretion, and revised. Doh et al. (2022) capture this with the distinction between formal and informal rules and between levels of governance. The firm that understands where a rule is made, who drafted it, what problem it was intended to solve, and what the regulator fears most is in a completely different position from the firm that reads only the final text. The legitimate channels are well known and underused: consultation responses, technical working groups, industry associations, pilot schemes, and regulatory sandboxes. The illegitimate channels are also well known, and this is where the ethical boundary of business diplomacy sits. Bucheli et al. (2024) remind us that corporate diplomatic activity has historically served both defensive and predatory ends. A firm that finds itself relying on personal relationships to obtain outcomes it could not obtain on the merits has crossed from diplomacy into capture, whatever it calls the activity internally. 5.4 Regulatory competence is now a source of competitive advantage The fourth finding follows from the third. Where rules are complex, unstable, and divergent across jurisdictions, the ability to read them well becomes a competitive asset rather than a cost centre. Consider the mundane example of data. A firm operating in Europe, the Gulf, India, and Southeast Asia faces distinct regimes governing where personal data may be stored, how it may be transferred, and who may access it. #data_localization requirements have expanded in the last decade. A firm that designs its architecture for a single regime and then patches it for others will pay repeatedly. A firm that designs for divergence from the beginning will move faster in every new market. The same logic applies to #ESG and #sustainability_reporting, where reporting obligations increasingly reach across borders and capture firms that are not domiciled in the regulating jurisdiction, and to #anti_money_laundering and #sanctions_compliance, where the cost of a mistake is exclusion from the dollar clearing system. Regulatory competence therefore has an architectural dimension. It is not only about answering the regulator. It is about designing the firm so that answering is cheap. 5.5 Hubs are laboratories of diplomatic capability The fifth finding is that global hubs concentrate every problem discussed so far and therefore reward every capability discussed so far. Take #Dubai. A firm establishing in the Dubai International Financial Centre operates under a common law framework and an independent financial regulator, inside a country whose wider legal system is civil law and whose commercial culture places substantial weight on personal relationships and long term presence. The manager must therefore hold two models at once: the technical, documentary, precedent driven logic of the zone, and the relational, patient, reputation driven logic of the region. Managers who master only the first produce excellent paperwork and no business. Managers who master only the second produce goodwill that cannot survive an audit. #Singapore offers a parallel case with a different flavour: a strong state, high regulatory quality, an explicit strategy of positioning as neutral ground between larger powers, and a dense concentration of arbitration and dispute resolution capacity. #London retains depth in law, insurance, and capital markets while operating in a post-Brexit regulatory environment that has diverged from the European Union in ways firms must actively track. #Hong_Kong illustrates the risk side of the hub model, where the value of a jurisdiction depends on external perceptions of its institutional autonomy as well as on its formal rules. Fjellstrom et al. (2023) show that firms use third countries as springboards when direct routes are politically blocked. Hubs are the natural springboards. This gives them commercial value and political exposure at the same time. A hub that becomes a route around sanctions attracts scrutiny. A hub that enforces every restriction perfectly loses the business that made it a hub. Managing that tension is itself an act of #economic_diplomacy, conducted by the hub authorities rather than by the firms, and firms operating in hubs are affected by how well their host performs it. 5.6 Dispute resolution is the shadow of every negotiation The sixth finding concerns what happens when agreement fails. #dispute_resolution is not merely a fallback. It is a live element of the negotiation, because every party negotiates in the shadow of what will happen if talks collapse. International commercial #arbitration has become the default mechanism for cross border commercial disputes, in large part because arbitral awards are more easily enforced across borders than national court judgments. Hubs compete to host arbitration precisely because it anchors legal and financial services around it. For the manager, the practical questions are concrete: which law governs the contract, which forum hears the dispute, in which language, and where can an award be enforced against real assets. #contract_enforcement is not an abstraction. It is the question of whether there is anything to seize. Students should note that the choice of forum is often the last item negotiated and the first item that matters when things go wrong. The party that treats it as boilerplate has effectively conceded it. 5.7 The compliance and diplomacy divide inside firms The seventh finding is organisational. In many multinationals, the people who understand regulation and the people who understand relationships sit in different functions, report to different executives, and meet only in crises. Legal and compliance own the rules. Government affairs and communications own the relationships. Commercial teams own the deal. Each function optimises its own metric. The result is predictable. Compliance says no late, after the commercial team has committed. Government affairs is asked to repair a relationship it was never allowed to build. The commercial team concludes that both functions are obstacles. Nobody is responsible for the whole. The three pillar model implies that diplomatic capability is a property of the firm rather than of any one function, and that the connecting processes of sense making and #boundary_spanning must be owned explicitly. Some firms address this with a standing cross functional forum that reviews the political and regulatory environment on a regular schedule rather than only when a problem arrives. The mechanism matters less than the principle: the pillars must be joined somewhere, by someone, before the crisis. 5.8 Trust, ethics, and the limits of influence The eighth finding is normative and unavoidable. Diplomacy is about influence, and influence has a boundary. The distinction that survives scrutiny is between influence exercised through argument, evidence, and legitimate coalition, and influence exercised through private benefit. The first strengthens the rule system. The second corrodes it. The corrosion is not only ethical. It is commercial, because a firm that obtains its position through private benefit holds an asset that evaporates the moment the official who granted it moves on. Shirodkar et al. (2024) argue for a values based approach to nonmarket strategy on grounds that are practical as much as ethical: in an environment of scrutiny, activist stakeholders, and rapid information flow, instrumental approaches that ignore values are unstable. #transparency and #due_diligence are therefore not merely defensive. They are what makes a diplomatic position durable. For students entering international careers, this is the single most useful thing to internalise. The pressure to cross the line will come, and it will come disguised as normal practice in the local market. The defence against it is not personal virtue alone. It is a firm that has decided in advance where its line sits, has written it down, and has made it survivable to say no. 5.9 Diplomacy as a response to fragmentation The ninth finding pulls the previous eight together. The reason diplomatic capability has risen up the management agenda is that the world economy is fragmenting into partially separate rule systems. Witt et al. (2023) document decoupling as an empirical pattern. Cui et al. (2023) analyse the vulnerability it creates and the strategies firms use in response, including diversification, localisation, and structural separation of operations across blocs. Devinney et al. (2023) frame sociopolitical uncertainty as a permanent condition for theory and policy alike. Ren and Gao (2023) show subsidiaries adapting their nonmarket strategies as the environment turns hostile. In a fully integrated world economy, a firm could run one global model and adapt at the margins. In a fragmented one, it must operate credibly inside multiple systems that view each other with suspicion, and it must do so without being accused of duplicity by any of them. That is a diplomatic problem before it is a strategic or an operational one. It is the reason #supply_chain_resilience, #risk_management, and #market_entry decisions now pass through political review in a way they did not fifteen years ago. 6. Implications for Managers The following guidance follows from the analysis. It is written directly, because students asked to apply this material generally need instructions rather than abstractions. 6.1 Map the environment before you need it Identify every party who can say no to you, slow you down, or damage your reputation. Include regulators, ministries, local authorities, community leaders, unions, media, competitors, and industry bodies. For each, record what they want, what they fear, who influences them, and what your last contact with them was. Update the map on a schedule. This is unglamorous work and it is the foundation of everything else. 6.2 Build relationships in peacetime Meet regulators when you have nothing to ask for. Respond to consultations even when the topic does not affect you directly. Join the industry association and do real work in it. #relationship_capital cannot be created at speed, and the attempt to create it at speed looks exactly like what it is. 6.3 Prepare culturally, but hold your preparation lightly Read about the counterpart's context, then treat what you have read as a hypothesis. Watch for the moment when the person in front of you contradicts the profile you built, and update immediately. The best negotiators are not the ones with the most cultural knowledge. They are the ones who notice fastest when their model is wrong. 6.4 Design the negotiation, not just the offer Decide in advance who attends, in what order people speak, what happens in the breaks, and what will be said informally. In high context settings, the meeting before the meeting is where the deal is actually made. In #virtual_negotiation, rebuild that space deliberately with bilateral calls and pre reads, because the screen will not create it for you. Know your #BATNA and estimate theirs. Almost every failure of nerve at the table comes from not knowing what happens if you walk away. 6.5 Bring regulation into the room early Involve legal and compliance at the design stage of a deal rather than at signature. The question is not whether a structure is permitted. It is whether it is permitted, defensible, and explainable to a regulator who is having a bad week and reading it in translation. 6.6 Choose your forum with care Negotiate governing law, dispute forum, language, and enforcement route as substantive terms. Ask where the counterparty holds assets that could actually satisfy an award. In hub jurisdictions, understand which court or tribunal will hear the matter and what body of law it will apply, because the answer may not be the one your head office assumes. 6.7 Decide your ethical line in advance Write down what the firm will not do, make it specific enough to be operational, and make it safe for a junior manager to invoke. A line that only a senior executive can enforce is not a line. 7. Implications for Students and Management Education This article is written for a student readership, and the teaching implications deserve their own treatment. 7.1 The skills gap Most #management_education still teaches negotiation and regulation separately, if it teaches regulation at all. Negotiation is taught in a behavioural course with role plays. Regulation appears in a business law elective. Political risk appears, if at all, in a strategy module. The student therefore graduates with three unconnected boxes and no way to assemble them. The evidence reviewed here suggests that the assembly is the skill. 7.2 What to learn, in order Learn to read an institution. Given any regulator, you should be able to say what its statutory objective is, what it is punished for failing at, who supervises it, and how it prefers to receive information. This is a research skill and it is learnable in an afternoon per institution. Learn to listen structurally. In a negotiation, distinguish between positions, which are what people say they want, and interests, which are why they want it. This distinction is old and it remains the most reliable value creating move available. Learn one legal system properly. It does not matter greatly which. Understanding one system in depth teaches you what to look for in another, and it inoculates you against the assumption that your home system is the natural order of things. Learn a second language to working level. Not for translation, which machines now do adequately, but because the effort of learning a language teaches you that concepts do not map one to one, which is the central insight of cross-cultural negotiation. Learn to write a submission. Regulators are persuaded by documents. A clear, honest, well evidenced submission is a diplomatic instrument, and few graduates can write one. 7.3 Careers The functions where this capability is exercised carry different names in different firms: government affairs, public policy, regulatory affairs, #public_affairs, corporate affairs, external relations, or simply strategy. In some firms it sits with the general counsel. In others it reports to the chief executive. Students should look at the reporting line rather than the title, because the reporting line tells you whether the function is listened to. 7.4 An exercise worth doing Take any recent cross border transaction that was blocked or delayed by a regulator. Reconstruct it from public sources. Identify what the firm assumed, what the regulator was actually worried about, and at what point the two diverged. Then write the submission the firm should have written. This exercise teaches more about diplomacy in #international_management than any amount of theory. 8. Limitations and Future Research 8.1 Limitations This article is conceptual and inherits the limitations set out in the methodology section. The three pillar model is an organising framework and has not been tested. The propositions offered in section three are plausible rather than demonstrated. The discussion of hubs draws on institutional description and on the general literature rather than on systematic comparative data. The literature reviewed is also unbalanced in ways worth naming. It over-represents large firms, which have the resources to maintain dedicated diplomatic functions, and under-represents the small and medium enterprises that make up most of international trade by number of firms. It over-represents Anglophone publication. And it tends to study the firms that survived, which introduces the usual survivorship problem: we read about the diplomacy that worked. 8.2 A research agenda Several questions follow directly from the framework. Measurement. Can diplomatic capability be measured at the firm level in a way that predicts outcomes such as licence approval rates, regulatory penalty exposure, or deal completion under political scrutiny? The construct currently lacks an accepted operational measure. Interaction effects. The propositions in section three imply interaction rather than addition. Does cultural competence without regulatory competence actually produce worse outcomes than either alone, as proposition one suggests? This is testable with deal level data. Small firms. How do smaller internationalising firms acquire diplomatic capability when they cannot staff a function? The likely answers involve intermediaries, industry associations, and home country trade promotion bodies, but the mechanisms are not well documented. Hubs as comparative cases. A systematic comparison of global hubs on dimensions of legal layering, regulatory competition, and neutrality would allow researchers to ask which hub features actually generate value for firms and which are marketing. The ethics boundary. Where, empirically, do managers believe the line between influence and capture sits, and does that belief vary by home country, by industry, and by seniority? Bucheli et al. (2024) show the historical duality of corporate diplomatic activity. Contemporary evidence on how practitioners themselves draw the line is thin. Technology. How does the use of artificial intelligence in translation, in regulatory monitoring, and in negotiation preparation change the distribution of diplomatic capability? It may democratise access to regulatory knowledge, which would reduce the advantage of large incumbents. It may equally concentrate advantage in firms that can afford the best systems. 9. Conclusion The argument of this article can be stated in a few sentences. Firms operating across borders face an environment in which rules are made in many places, applied unevenly, and increasingly used as instruments of national competition. In that environment, commercial skill is not enough. Firms need the ability to build relationships with parties they cannot command, to read cultural context accurately without collapsing into stereotype, and to understand the rule systems that govern them well enough to shape them legitimately. Those three capabilities, joined by sense making and boundary spanning, constitute diplomatic capability in #international_management. global hubs make the argument visible. In a city like Dubai or Singapore, a manager encounters layered legal systems, counterparties from a dozen cultures, competing regulators, and geopolitical pressure, all within a single working week. Hubs do not create these tensions. They concentrate them, and in doing so they show what the rest of the world economy increasingly looks like. For students, the practical message is that the boundary between commercial work and political work has dissolved. The negotiation you conduct is shaped by a regulator you have not met, in a legal system you did not choose, under political conditions you did not create. The managers who thrive are the ones who see all of it, who prepare for all of it, and who understand that a signature at the end of a deal is only the visible part of a much longer process of persuasion. That process is diplomacy, and it now belongs to management. Hashtags #diplomacy_in_international_management #cross_cultural_negotiations #global_business_hubs #regulatory_navigation #international_business_diplomacy #negotiating_across_cultures #global_regulatory_compliance #corporate_diplomacy_skills #business_government_relations #international_negotiation_skills #cross_border_deals #global_leadership #emerging_market_strategy #culture_and_business #global_trade_governance #international_management_research #strategic_diplomacy #global_careers #academic_writing #student_research References Abdal, A., and Ferreira, D. M. (2021). Deglobalization, globalization, and the pandemic: Current impasses of the capitalist world-economy. Journal of World-Systems Research, 27(1), 202 to 230. Agamennone, E. (2022). The rise and impact of the Zoom negotiation: Cross-cultural variations in virtual negotiations and lessons from the COVID-19 pandemic. In S. Lalani and S. G. Shapiro (Eds.), The Impact of Covid on International Disputes (pp. 5 to 27). Leiden: Brill Nijhoff. Alammar, F. M., and Pauleen, D. J. (2022). Business diplomacy in practice: A strategic response to global business challenges. Journal of General Management, 48(1), 14 to 31. https://doi.org/10.1177/03063070211059943 Bucheli, M., Duran, X., and Kim, M. (2024). My best frenemy: A history-to-theory approach to MNCs corporate diplomatic activities. Journal of International Business Studies, 55(3), 326 to 341. Chande, I. (2024). Effects of cross-cultural negotiation strategies on international joint venture success. International Journal of Business Strategies, 9(1), 49 to 62. https://doi.org/10.47672/ijbs.1822 Cui, V., Vertinsky, I., Wang, Y., and Zhou, D. (2023). Decoupling in international business: The new vulnerability of globalization and MNEs response strategies. Journal of International Business Studies, 54(8), 1562 to 1576. Cuhlova, R., and Demel, J. (2024). From diverse perspectives to unified agreements: Intercultural negotiation dynamics in the digital age. Business and Professional Communication Quarterly. https://doi.org/10.1177/23294906241253516 Devinney, T. M., Hartwell, C. A., Oetzel, J., and Vaaler, P. (2023). Managing, theorizing, and policymaking in an age of sociopolitical uncertainty: Introduction to the special issue. Journal of International Business Policy, 6(2), 133 to 140. Doh, J. P., Dahan, N. M., and Casario, M. (2022). MNEs and the practice of international business diplomacy. International Business Review, 31(1), 101926. https://doi.org/10.1016/j.ibusrev.2021.101926 Fjellstrom, D., Bai, W., Oliveira, L., and Fang, T. (2023). Springboard internationalisation in times of geopolitical tensions. International Business Review, 32(6), 102144. Huang, D., Tsui-Auch, L. S., Dieleman, M., and Gomulya, D. (2024). Politically connected EMNCs in a (de)globalising world: A review and future research directions. International Business Review, 33(3). Jazbec, M. (2022). Diplomacy and management: In a search of an intersection in the time of globalisation. International Journal of Diplomacy and Economy, 8(2), 209 to 221. Khakhar, P., Rammal, H. G., and Pereira, V. (2023). Biculturals in international business negotiations: Moving away from the single culture paradigm. Journal of Organizational Change Management. Lewicki, R. J., Barry, B., and Saunders, D. M. (2021). Essentials of Negotiation (7th ed.). New York: McGraw Hill. Luo, Y., and Van Assche, A. (2023). The rise of techno-geopolitical uncertainty: Implications of the United States CHIPS and Science Act. Journal of International Business Studies, 54(8), 1423 to 1440. Meyer, D. R. (2023). Working together: Elite global financiers and senior government officials. Economic Diplomacy, 1(1), 42 to 55. Ng, K. Y., Ang, S., and Rockstuhl, T. (2022). Cultural intelligence: From intelligence in context and across cultures to intercultural contexts. In R. J. Sternberg and D. D. Preiss (Eds.), Intelligence in Context: The Cultural and Historical Foundations of Human Intelligence (pp. 177 to 200). Cham: Palgrave Macmillan. Ren, M., and Gao, H. (2023). Chinese state-owned multinationals subsidiary nonmarket strategies in selective de-globalization: An integrated perspective of co-evolution theory and the yin-yang frame. Journal of International Management, 29(6), 101069. Salvi, R. M., and Ruel, H. J. M. (2022). Corporate diplomacy and sustainable internationalisation: A conceptual and empirical exploration. International Journal of Diplomacy and Economy, 8(1), 41 to 64. Semenov, A. V., and Randrianasolo, A. (2024). An extended conceptualization of cultural intelligence. International Journal of Cross Cultural Management. https://doi.org/10.1177/14705958241268289 Shirodkar, V., Liedong, T. A., Rajwani, T., and Lawton, T. C. (2024). MNE nonmarket strategy in a changing world: Complexities, varieties, and a values-based approach. International Business Review. Witt, M. A., Lewin, A. Y., Li, P. P., and Gaur, A. (2023). Decoupling in international business: Evidence, drivers, impact, and implications for IB research. Journal of World Business, 58(1).

Organizations that enter unfamiliar markets face a simple but stubborn problem. Buyers cannot see quality before they commit, and they have little reason to trust a name they do not know. This article examines how #quality_assurance systems and #international_accreditation can be used as strategic instruments that reduce this uncertainty and convert verified quality into durable #brand_equity. Drawing on an integrative review of recent peer reviewed research in quality management, higher education studies, international business, and services marketing, the article develops a conceptual framework that explains how a quality mark becomes a market asset. The framework has four stages: building genuine internal quality capability, obtaining credible external verification, communicating the verification in ways the local audience understands, and converting recognition into loyalty, price tolerance, and partnership access. The review identifies six recurring findings. First, accreditation works as a #signal only when the receiving audience already recognizes the accrediting body. Second, the strongest gains appear where #information_asymmetry is severe, such as new entrants in unfamiliar countries. Third, benefits are conditional on real internal change rather than paperwork compliance. Fourth, accreditation lowers perceived risk more than it raises perceived excellence. Fifth, multiple accreditations produce diminishing returns once a threshold of credibility is reached. Sixth, the advantage is temporary because rivals imitate the mark, which pushes organizations toward deeper and less imitable quality capabilities. The article closes with a practical roadmap for managers, deans, and quality directors, along with a research agenda for scholars. The contribution is a bridge between the quality management literature, which studies systems, and the branding literature, which studies perceptions, showing that neither alone explains why some accredited organizations gain #competitive_advantage while others simply gain a certificate. Keywords: quality assurance; international accreditation; brand equity; competitive advantage; signaling theory; market entry; higher education; certification 1. Introduction Every organization that crosses a border carries an invisible burden. Its reputation does not travel as fast as its ambition. A university that is well known at home is a stranger abroad. A manufacturer with forty years of reliable delivery in one region is an unproven supplier in another. A hospital group, a law firm, a testing laboratory, or a training provider all face the same wall. The people they want to serve cannot verify quality in advance, and they do not yet have a reason to believe a promise. This is the central problem of #market_entry, and it is not solved by advertising alone. Advertising is cheap to produce and easy to imitate, so it carries little information about the underlying truth of a claim. What buyers need is evidence that is costly to fake. #International_accreditation and formal certification provide exactly this kind of evidence. When an independent body with a known name inspects an organization against published standards and issues a public judgment, the organization gains something that money alone cannot buy quickly, which is third party #credibility. The idea that quality can be a source of advantage is old. What is new is the scale of the accreditation economy and the strategic weight now placed on it. Business schools pursue AACSB, EQUIS, and AMBA. Engineering programs pursue ABET and Washington Accord recognition. Hospitals pursue Joint Commission International. Laboratories pursue ISO 17025. Manufacturers and service firms pursue ISO 9001, and increasingly a portfolio of related standards. National agencies build their own frameworks and then seek recognition from regional networks so that their judgments carry weight abroad. In each case, the organization is buying more than a management system. It is buying a place in a #recognition network. The strategic question that follows is rarely asked with enough honesty. Does the certificate create value, or does it only record value that already existed? The empirical literature is unusually divided on this point. Reviews of #ISO_9001 research find that the effect of certification on performance is inconsistent across countries, sectors, and firm sizes, and that outcomes depend heavily on whether the organization adopted the standard for internal improvement or for external appearance. Studies of business school accreditation reach a similar conclusion from a different direction. Accreditation reshapes internal processes and raises research expectations, yet the external audiences that accreditation is supposed to impress often do not recognize the badges at all. A recent survey of business professionals in Australia found low awareness of major international accreditation bodies among the very stakeholders whose trust the accreditation is meant to secure (Attree, Neher, and Xie, 2025). A signal that no one reads is not a signal. It is an expense. This paradox is the starting point of the present article. #Quality_assurance is expensive. Preparation cycles run for years. Documentation consumes faculty and staff time. Fees, site visits, consultants, and continuous improvement reviews add cost. If the return is uncertain, then the practice deserves careful strategic analysis rather than institutional imitation. Yet the fact that many organizations do gain from accreditation, particularly when entering new markets, suggests that the return is not random. It is conditional. Understanding those conditions is the purpose of this study. The article makes three arguments. The first is that accreditation should be understood as a #brand_asset rather than as a compliance outcome. Its value lies in what it does to the beliefs of students, patients, buyers, regulators, employers, and partners. The second is that the conversion from certificate to brand equity is not automatic. It passes through a chain of steps, and the chain breaks easily. Recognition can fail, communication can be clumsy, the promise can be broken by weak delivery, or the market may already be saturated with the same mark. The third argument is that accreditation gives a #temporary_advantage. Because standards are public and available to all, rivals can copy the badge. Advantage therefore survives only when the badge sits on top of a quality culture that is difficult to imitate. The study addresses four research questions. How does international accreditation reduce uncertainty for buyers in unfamiliar markets? Through what mechanisms does it contribute to the dimensions of #brand_equity, meaning awareness, perceived quality, associations, and loyalty? What conditions strengthen or weaken this contribution? And how can organizations design an accreditation strategy that produces durable rather than temporary advantage? The contribution is integrative. Quality management research has studied certification for decades but has treated brand outcomes as a side effect. Branding research has studied #perceived_quality but has rarely examined the institutional machinery that produces credible quality claims. International business research has studied #legitimacy in foreign markets but has treated certification as one variable among many. This article brings the three streams into a single framework and states the conditions under which quality assurance becomes competitive advantage. The remainder of the article proceeds as follows. Section two reviews the relevant literature. Section three sets out the theoretical framework and develops propositions. Section four explains the review method. Section five presents the thematic findings. Section six presents the conceptual model. Section seven discusses the implications for theory. Section eight offers a practical roadmap. Section nine notes limitations and a research agenda. Section ten concludes. 2. Literature Review 2.1 The Accreditation and Certification Landscape Quality assurance refers to the planned and systematic activities that give confidence that an organization will meet defined requirements. In practice the term covers two related but distinct things. The first is the internal system, meaning the processes, measurements, reviews, and corrective actions that an organization uses to manage its own work. The second is the external judgment, meaning the inspection and verification carried out by an independent body against published standards. The internal system produces quality. The external judgment produces #credibility. Both matter, and confusing them is the source of much strategic error. External judgments come in several forms. Certification usually applies to a management system, as with ISO 9001, and it is granted by a certification body that is itself accredited by a national accreditation body. Accreditation usually applies to an institution or a program, as with a business school, an engineering degree, or a hospital, and it is granted by a recognized agency after a self study and a peer review visit. Recognition applies to agencies themselves, as when a national quality agency is admitted to a regional register and its decisions gain cross border weight. The three layers form a chain of trust, and the value of any single certificate depends on the strength of the chain above it. In higher education, this chain has become genuinely transnational. Research on cross border higher education finds that international accreditation, based on transnational guidelines and international agencies, has become a prominent feature of the field, and that this creates friction with national systems because international standards are written broadly to accommodate many institutions while national agencies are constrained by local requirements (Carvalho, Rosa, and Amaral, 2023). The result is a market in which institutions frequently hold both a national approval, which grants the legal right to operate, and an international accreditation, which grants #reputational_capital. The two do different work. One is a licence. The other is a #brand_signal. The scale of adoption is large. Business school accreditation has expanded steadily and now covers institutions across more than sixty countries and territories, with the accrediting bodies themselves publishing annual reports on outcomes, priorities, and improvement areas. Management system certification is larger still, with millions of certificates issued worldwide across almost every country. Growth of this magnitude has consequences. When a mark becomes common, it stops distinguishing and starts qualifying. It moves from being an advantage to being a #ticket_to_entry, a shift that has serious strategic implications discussed later in this article. 2.2 Evidence on the Performance Effects of Quality Assurance The empirical record on whether quality assurance improves performance is mixed, and honest reporting of that mixture is important. Systematic reviews of ISO 9001 research covering hundreds of studies conclude that the effect on organizational performance is contingent rather than universal. Some studies report improved operational efficiency, better productivity, higher sales, and improved market image. Others find no significant difference between certified and non certified firms. Others still find that effects appear only in some sectors, some cultures, or some firm sizes. The contingencies that recur most often are instructive. Leadership commitment matters. A study of Indonesian organizations found that the effectiveness of ISO practices in improving firm performance depends on the degree of leadership commitment and the prevailing work culture (Bakhtiar, Nugraha, Suliantoro, and Pujotomo, 2023). Motivation matters. Firms that adopt a standard because customers demand it, and who therefore treat it as paperwork, gain less than firms that adopt it to solve real internal problems. Evidence from Serbia indicates that certification improves both employee related and customer related performance, but that the incentives behind certification influence the strength of that relationship (Milovanovic, Paunovic, and Casadesus, 2023). Resources matter. Recent work on listed firms in emerging markets suggests that financial benefits from certification materialize mainly when the firm has enough financial slack to sustain investment in the quality system, which means that certification acts as a #performance_catalyst rather than a universal driver. Higher education research shows the same pattern in a different vocabulary. Accreditation is associated with institutional performance improvements, with clearer learning outcomes, and with stronger internal accountability. Research in Oman on private higher education institutions reports accreditation as a significant predictor of institutional performance. At the same time, faculty perspectives from private institutions describe accreditation as a process that can strengthen quality learning while also raising concerns about administrative load and academic freedom (Ashraf, Ratan, Amir, Joarder, and Osman, 2025). Work on quality assurance management systems argues that many institutions struggle to move from a compliance orientation to an effective, embedded system (Aldhobaid, 2024). The struggle is not a footnote. It is the main event. There is also a competitive dimension that is easy to overlook. Accreditation does not only improve institutions. It rearranges them. Research on business accreditation and positional competition shows that accreditation systems create hierarchies, and that schools use accreditation to defend or improve their position relative to rivals (Alajoutsijarvi, Kettunen, and Sohlo, 2018). German business schools, for example, responded to international accreditation as a strategic move in global competition rather than only as an improvement exercise (Sziegat, 2021). This is #positional_strategy, and it is closer to branding than to quality management. 2.3 Brand Equity and Its Dimensions Brand equity refers to the value that a name adds to an offering beyond the functional properties of the offering itself. In the customer based tradition, it is built from what sits in the mind of the customer. The standard components are #brand_awareness, meaning whether the name is known and recalled, #perceived_quality, meaning the belief that the offering is good, brand associations, meaning the ideas and feelings connected to the name, and #brand_loyalty, meaning the willingness to choose it again and to recommend it (Keller and Swaminathan, 2020). In services, and especially in education and healthcare, brand equity carries more weight than in goods markets. The offering is intangible, it is consumed over a long period, and the buyer cannot inspect it before purchase. Research on higher education brand equity confirms this. A large study of Indian students and alumni traced the strongest path to institutional brand equity through service quality, then brand image, then attachment, then word of mouth (Khanna, Singh, and Bedi, 2025). The path begins with quality and ends with advocacy, which is precisely the sequence this article is concerned with. Related work shows that brand equity and #service_quality shape university reputation and student intention to enrol (Rizard, Waluyo, and Jaswir, 2023), that social media engagement builds brand equity through social identity processes (Perera, Nguyen, and Nayak, 2023), and that a strong education brand can even influence how much students believe they are learning, a form of educational placebo effect in which brand shapes experience (Esteky and Kalati, 2024). That last finding deserves emphasis, because it explains why brand is not cosmetic. A credible quality signal changes expectations, expectations change effort and engagement, and engagement changes outcomes. #Brand_equity is therefore partly self fulfilling. The same logic operates in industrial markets, where a certified supplier is given more responsibility, more information, and more forgiveness during problem solving, all of which improve the actual relationship. International branding literature adds a further layer. Institutions seeking world class status pursue international visibility deliberately, and accreditation is one of the visible markers used in that pursuit (Gardiana, Rahmanto, and Satyawan, 2023). Yet visibility abroad is not the same as visibility at home, and a badge that impresses a ministry may mean nothing to a parent. 2.4 The Missing Bridge The two bodies of work described above rarely speak to each other. Quality management research measures systems, audits, and financial ratios. Branding research measures perceptions, attitudes, and intentions. The connective tissue, meaning the process by which a verified quality claim becomes a believed quality perception, is under theorized. Three gaps stand out. The first is the #recognition_gap. Almost all strategic arguments for accreditation assume that stakeholders know what the accreditation means, yet the direct evidence on stakeholder awareness is thin and what exists is discouraging (Attree, Neher, and Xie, 2025). The second is the #conversion_gap. Even where awareness exists, no clear model explains how a signal becomes loyalty, and which internal actions are needed at each step. The third is the #durability_gap. Accreditation is available to competitors, so any advantage should erode, yet the strategic literature has not carefully described what remains after erosion. This article addresses all three. 3. Theoretical Framework 3.1 Signaling Theory Signaling theory begins with a market in which one party knows more than the other. The seller knows the quality of the offering. The buyer does not. In such a market, cheap talk fails because a low quality seller can make the same claims as a high quality seller at the same cost. The solution is a signal that is expensive to produce and more expensive for a low quality actor than for a high quality one (Spence, 1973). Education itself was the original example. Accreditation is a modern one. Contemporary reviews of signaling in management research describe four elements of a functioning signal: a signaller who has private information, a signal that is observable and costly, a receiver who observes and interprets, and a feedback process that either confirms or destroys the signal over time (Bafera and Kleinert, 2023). Each element is a possible failure point. If the signal is cheap, it is ignored. If the receiver does not understand the code, the signal is noise. If the promise is not kept, the signal is destroyed and the signaller is worse off than before. Three implications follow for accreditation. First, #signal_cost is a feature and not a bug. The multi year preparation, the peer review visits, and the continuous improvement requirements are what make the mark believable. An accreditation that is easy to obtain, such as one purchased from an unrecognized body, is worthless precisely because it is easy. This is why #accreditation_mills damage the whole system. Second, #signal_fit matters. The signal must be legible to the specific receiver. Employers, students, parents, ministries, and partner institutions read different codes. Third, signals work hardest where uncertainty is greatest, which is exactly the condition of a #new_market entrant. Proposition 1. The brand value of an international accreditation increases with the perceived cost and rigour of obtaining it, and decreases when similar marks become easy to acquire. Proposition 2. The brand value of an international accreditation is conditional on receiver recognition, so that awareness among the target stakeholder group moderates the relationship between accreditation and brand equity. 3.2 Institutional Theory and Legitimacy Organizations do not only compete. They also seek to be seen as proper, appropriate, and safe. Institutional theory explains that firms adopt structures and practices because those practices are expected in their field, and that conformity produces #legitimacy which in turn produces access to resources, permission to operate, and the willingness of others to cooperate. For an entrant to a foreign market, this is critical. A foreign organization suffers a liability of foreignness, which includes unfamiliarity, discrimination, and the absence of local networks. Adopting a globally recognized standard is a way of borrowing legitimacy from an established institution. The organization says, in effect, that even if you do not know us, you know the body that has inspected us. Institutional theory also predicts a darker outcome, and the prediction is well supported. When adoption is driven by external pressure rather than internal conviction, organizations may separate the formal structure from the actual work. This is #decoupling, or ceremonial adoption. The manual exists, the audit is passed, and nothing changes. Ceremonial adoption explains a great deal of the inconsistent performance evidence reviewed earlier. It also explains why certification sometimes damages brands, because when the visible promise is contradicted by lived experience, trust falls further than it would have without the promise. Proposition 3. Accreditation increases legitimacy and resource access in a new market, but the effect on brand equity is negative when adoption is ceremonial and delivery contradicts the signal. 3.3 The Resource Based View The resource based view holds that sustained advantage comes from resources that are valuable, rare, difficult to imitate, and supported by the organization (Barney, 1991). Applied to quality assurance, the view produces an uncomfortable but useful conclusion. The certificate itself is valuable but not rare and not hard to imitate. Standards are published. Consultants are available. Any competitor with money and patience can obtain the same mark. Therefore the certificate alone cannot sustain advantage. What can sustain advantage is the underlying #quality_capability, meaning the routines, the data systems, the review habits, the trained people, and the culture of continuous improvement that the accreditation process may have helped to build. These are tacit, path dependent, and socially complex, which is to say they are hard to copy. Recent work on emerging markets similarly finds that strategic orientation and organizational capability, not the possession of a label, drive sustainable competitive advantage (Ed-Dafali, Al-Azad, Mohiuddin, and Reza, 2023). The strategic logic is therefore two stage. Accreditation opens the door. #Capability keeps the customer inside. Organizations that treat accreditation as the goal end up with a certificate and no advantage. Organizations that treat accreditation as a forcing device for building capability end up with both. Proposition 4. The competitive advantage from accreditation is temporary and imitable, while the advantage from the quality capability developed during accreditation is durable. 3.4 An Integrated View The three theories answer different questions. Signaling theory explains why an outsider believes the claim. Institutional theory explains why the organization is permitted to operate and to be taken seriously. The resource based view explains why the advantage lasts or does not. Together they describe a sequence rather than three alternatives, and that sequence is the basis of the conceptual model presented in section six. 4. Methodology 4.1 Design This study uses an integrative literature review with thematic synthesis. This design is appropriate when a field is fragmented across disciplines and when the aim is to build a conceptual framework rather than to estimate an effect size. Integrative review allows the inclusion of quantitative, qualitative, and conceptual work, which is necessary here because the relevant knowledge is distributed across quality management, higher education, international business, and marketing. 4.2 Selection Criteria Sources were selected against four criteria. They had to be peer reviewed journal articles or scholarly books. They had to address at least one of the four core constructs, meaning quality assurance systems, external accreditation or certification, brand equity or reputation, and internationalization or market entry. They had to be published recently, with a strong preference for work from the last five years, except for a small number of foundational theoretical works that remain necessary to explain the mechanisms. And they had to report evidence or theory rather than promotional material, which meant that accrediting body marketing content was excluded from the evidence base and used only as background context. 4.3 Analysis Sources were read and coded into themes using an iterative process. Initial codes were derived from the theoretical framework, meaning signal, legitimacy, and capability. Additional codes emerged from the material itself, notably recognition failure, cost burden, threshold effects, and imitation. Codes were then grouped into the six themes reported in section five. Where studies disagreed, the disagreement was retained and examined rather than averaged away, because the disagreement itself is informative and points to moderating conditions. 4.4 Limitations of the Method An integrative review is interpretive. Different reviewers reading the same literature might weight studies differently. The review is also limited by publication bias, since studies reporting no effect of certification are less likely to be published, which may make the average reported benefit look larger than the true one. Finally, the evidence base is uneven across sectors. Higher education and manufacturing are well studied. Healthcare, professional services, and vocational training are less so. These limitations are revisited in section nine. 5. Findings 5.1 Theme One: Accreditation Works Best Where Uncertainty Is Highest The clearest pattern across the literature is that quality signals matter most when buyers know least. In mature home markets, where an organization has decades of word of mouth, alumni, referrals, and local reputation, an added certificate contributes little. The market already has better information than any badge can provide. In a #new_market, the situation inverts. The entrant has no local history. Local buyers have no basis for judgment, and the cost of a wrong choice is high. A degree is four years of a young person's life. A hospital procedure is irreversible. A supply contract can shut a production line. Under these conditions, buyers actively look for external proof, and an internationally recognized mark is one of the few proofs available. This is why accreditation is best understood as a strategy of #risk_reduction rather than a strategy of excellence. Stakeholders in unfamiliar markets are not primarily asking which provider is best. They are asking which providers are safe to consider. Accreditation answers the second question, and answering the second question is what gets an entrant onto the shortlist. Everything that follows, including price, personal relationships, campus visits, and product trials, happens only after the shortlist is formed. The mark buys the right to compete. A practical implication follows immediately. The value of accreditation should be measured at the top of the funnel, not the bottom. If an institution measures only enrolment or sales conversion, it will underestimate the effect. The effect appears earlier, in inclusion, in inquiry, in willingness to consider, and in the shortening of the trust building period. 5.2 Theme Two: Recognition Is the Binding Constraint The second theme is the most uncomfortable one for organizations that have spent heavily on accreditation. A signal only works when the receiver can read it. Empirical evidence indicates that the audiences accreditation is supposed to impress often do not know the accrediting bodies exist. In the survey of Australian business professionals reported by Attree, Neher, and Xie (2025), awareness of professional bodies was considerably higher than awareness of international accreditation bodies, and a substantial share of respondents had not heard of the major accreditation names at all. Similar observations appear in studies of engineering accreditation, where industry members who were not directly involved in advisory boards reported that they were unaware of program accreditation because it never appeared in recruitment or in the way employers described their requirements. This finding does not mean accreditation is worthless. It means that the value is unevenly distributed across stakeholder groups. Ministries, regulators, partner universities, scholarship funders, ranking agencies, corporate procurement departments, and multinational employers usually do know the marks, because their work requires them to. Individual consumers, meaning students, parents, patients, and small buyers, frequently do not. The strategic consequence is that #accreditation_communication is not a supporting activity. It is half the investment. An organization that spends three years and a large budget earning a mark and then places the logo in a website footer has completed the expensive half of the work and skipped the cheap half. The logo carries no meaning to a receiver who has no code for it. Meaning has to be supplied. Supplying meaning requires translation into consequences that the audience already values. A parent does not care that a school met assurance of learning standards. A parent cares that the degree will be recognized by employers abroad, that credits will transfer, that the teaching is independently checked, and that the school cannot quietly lower its standards without being caught. Those are the same facts, expressed in the receiver's language. The literature on higher education branding supports this, since it consistently finds that brand equity is built through perceived #service_quality and image rather than through institutional jargon (Khanna, Singh, and Bedi, 2025). 5.3 Theme Three: Substance Determines Whether the Signal Survives The third theme concerns what happens after the promise is made. Signals invite scrutiny. Once an organization claims verified quality, every subsequent experience is evaluated against that claim. This raises the stakes of delivery. The certification literature is blunt about this. Where firms adopt standards for appearance, and where documentation replaces improvement, the measurable benefits disappear. Where leadership is committed and the system is used to solve real problems, benefits appear across operational and customer facing performance (Bakhtiar and colleagues, 2023; Milovanovic and colleagues, 2023). The higher education literature reaches the same conclusion through the language of #quality_culture, arguing that institutions must move beyond compliance driven quality assurance toward embedded systems that actually inform decisions (Aldhobaid, 2024). The branding consequence is severe and asymmetric. An unaccredited organization that disappoints a customer has broken an implied promise. An accredited organization that disappoints a customer has broken an explicit, externally witnessed promise. The second betrayal costs more, because it damages not only the relationship but also the buyer's confidence in the entire verification system. This is why #ceremonial_adoption is not merely a wasted expense. It is a liability. It converts a brand asset into a brand risk. There is a further internal effect that deserves attention. Accreditation processes, when taken seriously, force organizations to do things they have avoided. They require the collection of outcome data. They require closing the loop between measurement and change. They require documentation of decisions. They require faculty and staff to agree on what they are trying to achieve. Many organizations report that the preparation itself was more valuable than the certificate, because it revealed problems that internal politics had protected for years. This is the #capability_building effect, and it is the part of accreditation that competitors cannot copy by buying the same badge. 5.4 Theme Four: The Threshold Effect and the Limits of Stacking The fourth theme is that returns to accreditation are not linear. They follow a threshold pattern. The first credible international mark is transformative for an unknown entrant. It moves the organization from unverified to verified, which is a change of category rather than a change of degree. The second mark, if it addresses a different audience or a different domain, adds meaningful value. The third and fourth marks generally add much less, because the audience has already placed the organization in the trusted category and further badges do not change the category. Research on multiple certification is consistent with this. Studies of over certification note that organizations often hold several overlapping certificates whose requirements substantially duplicate one another, and that the profitability of such stacking remains unclear. The costs, however, are certain. Each mark carries fees, audits, documentation, and staff time. Each mark also carries a maintenance obligation, since accreditation is not permanent and continuous improvement reviews must be passed. There is also a #differentiation problem that grows with adoption. When a mark is rare, holding it distinguishes. When a mark becomes common in a market, not holding it stigmatizes, but holding it no longer distinguishes. The mark migrates from a source of advantage to a #hygiene_factor. This migration is predictable and should be planned for. An organization entering a market where the mark is still rare gains real positional benefit. An organization entering a market where every serious competitor already holds it gains only the right to be considered. The strategic guidance that follows is to sequence accreditations by audience rather than to collect them by prestige. The question is not which mark is most famous, but which mark is read by the specific stakeholder whose decision the organization needs to change in the next three years. 5.5 Theme Five: The Advantage Is Real but Temporary The fifth theme is that accreditation based advantage decays. Because standards are public, the pathway is open to every competitor. Consultancy markets exist precisely to help laggards catch up. Within a few years of a first mover gaining a mark in a given market, its main rivals will typically have it too. This does not make the first move pointless. Timing has value. The early holder shapes the local understanding of what the mark means, is often quoted in media and government discussion of quality, becomes the default partner for foreign institutions seeking a local ally, and accumulates a cohort of satisfied customers before rivals arrive. These accumulated assets, meaning alumni networks, published outcomes, partnership agreements, and word of mouth, outlive the exclusivity of the badge. The resource based view predicts exactly this outcome. What is imitable erodes. What is tacit and path dependent persists. The task for management is therefore to use the window of exclusivity to build things that cannot be bought, including a strong faculty or workforce, a data driven improvement routine, deep employer relationships, a distinctive teaching or service model, and a reputation for keeping promises. When the rivals finally arrive holding the same certificate, the organization must have something else to say. 5.6 Theme Six: Costs, Burdens, and Unintended Effects The sixth theme is the honest accounting of costs, which promotional literature about accreditation tends to omit. The direct costs are visible: application fees, annual membership fees, review fees, travel and hosting for peer review teams, and often external consultants. The indirect costs are larger and less visible. Senior academic and managerial time is diverted for months. Administrative staff spend long periods producing evidence. Faculty report increased documentation demands and questions about #academic_freedom when standardization pressure conflicts with pedagogical judgment (Ashraf and colleagues, 2025). Institutions may also change their behaviour in ways that satisfy the standard while narrowing their mission, for example by favouring research outputs that count toward accreditation metrics over community engagement that does not. There is a further systemic cost. As accreditation becomes more valuable, fraudulent providers appear. #Accreditation_mills sell certificates without meaningful review, and their existence degrades the value of legitimate marks by making buyers suspicious of all of them. This is a classic signal degradation problem. It also raises the burden on legitimate organizations, which must now explain not only that they are accredited but that their accreditor is real. Cross border quality assurance work at the regional and international level exists partly to address this problem by building registers of recognized agencies, and organizations entering new markets are wise to align themselves with agencies that appear on such registers. The purpose of listing these costs is not to argue against accreditation. It is to insist that accreditation is a strategic investment with a real price and an uncertain return, and that it should be evaluated with the same discipline as any other capital allocation. 6. A Conceptual Model: From Quality System to Brand Equity The findings can be organized into a four stage conversion model. Each stage has a necessary condition, and failure at any stage prevents the next. 6.1 Stage One: Build The first stage is the construction of a genuine internal quality system. This includes defining outcomes, measuring them, reviewing the results, acting on the results, and repeating. In education this means assurance of learning. In manufacturing it means process control and corrective action. In services it means service standards, complaint analysis, and recovery. The necessary condition at this stage is #leadership_commitment. Without it, the system becomes documentation. The evidence reviewed earlier consistently identifies leadership and culture as the difference between certification that improves performance and certification that does not. The output of stage one is capability, which is the durable asset. 6.2 Stage Two: Verify The second stage is external verification by a recognized body. The organization submits to inspection and receives a public judgment. The necessary condition here is #accreditor_credibility. The body must be recognized by the audiences that matter in the target market, and it must itself be recognized within the international quality architecture. A certificate from an unrecognized body is not merely useless. It is evidence of poor judgment and can damage the brand when discovered. The output of stage two is a signal. 6.3 Stage Three: Communicate The third stage translates the signal into meaning for specific audiences. This is where most organizations fail, and it is the cheapest stage to fix. The necessary condition is #audience_translation. The message must state consequences rather than compliance. For students and parents the consequence is recognition, mobility, employability, and protection. For employers the consequence is a predictable graduate profile and a reduced hiring risk. For partner institutions the consequence is eligibility for joint programs and credit transfer. For regulators the consequence is reduced supervisory risk. For corporate buyers the consequence is supply reliability and lower audit costs. Channels matter as much as content. Since consumer awareness of accreditation bodies is often low, the message must travel through sources the audience already trusts, including employers, alumni, agents, professional associations, media, and peer word of mouth. Research on higher education branding confirms that word of mouth and social brand engagement are powerful builders of brand equity (Perera, Nguyen, and Nayak, 2023; Khanna, Singh, and Bedi, 2025), which means that the most effective accreditation communication is often not owned advertising but the testimony of credible third parties. The output of stage three is #recognition. 6.4 Stage Four: Convert The fourth stage turns recognition into brand equity and then into competitive outcomes. Recognition raises perceived quality and lowers perceived risk. Lower perceived risk widens the consideration set and shortens the decision cycle. Positive experience, delivered by the capability built in stage one, confirms the signal. Confirmed signals produce satisfaction, satisfaction produces loyalty and advocacy, and advocacy produces the self reinforcing reputation that no longer depends on the certificate. The necessary condition is #delivery_consistency. If experience contradicts the signal, the conversion runs backwards and the brand is harmed. The output of stage four is #brand_equity, expressed as awareness, perceived quality, favourable associations, and loyalty, which in turn supports #price_premium, partnership access, talent attraction, and resilience during crises. 6.5 Moderators Four moderators govern the strength of the whole chain. Market maturity moderates stage four. In a market where accreditation is rare, the same signal produces more differentiation. Stakeholder awareness moderates stage three. Where awareness is low, communication investment must be higher, and where awareness is very low, the organization may need to borrow recognition from a partner brand instead. Cultural distance moderates stages three and four. In markets where trust is built through relationships rather than documents, a certificate supports but does not replace personal engagement. Category of buyer moderates everything. Institutional buyers read certificates. Individual consumers read stories, outcomes, and the experiences of people like themselves. 7. Discussion 7.1 Theoretical Contribution The model contributes to theory in three ways. First, it specifies the boundary conditions of the signaling argument in a way that the general literature does not. The usual claim is that certification signals quality. The refinement offered here is that certification signals quality only to audiences with the code to read it, and that in consumer facing markets this code is frequently absent. The binding constraint is not the quality of the signal but the literacy of the receiver. This reframes accreditation strategy from an audit problem into a #market_education problem. Second, it reconciles the contradictory empirical evidence on certification and performance. The contradiction dissolves once we distinguish between the two products of accreditation. The certificate is a signal, and its value depends on the market. The capability is a resource, and its value depends on the organization. Studies that measure certification without measuring implementation depth are measuring a mixture of the two, which is why their results scatter. This suggests that future empirical work should measure implementation depth and stakeholder awareness as separate variables rather than treating certification as a binary. Third, it connects quality management to brand equity through a causal chain rather than an assertion. The chain is build, verify, communicate, convert. It identifies the failure point at each link and names the moderating conditions. This gives both scholars and practitioners a diagnostic tool. When accreditation fails to deliver value, the question is not whether accreditation works, but which link broke. 7.2 The Deeper Strategic Logic Underneath the model lies a simple idea that is worth stating plainly. #Trust is the scarce resource in any new market, and trust can be built in only three ways. It can be earned over time, which is slow. It can be borrowed from someone the market already trusts, which is what accreditation and partnerships do. Or it can be claimed, which does not work. Accreditation is therefore a mechanism for buying time. It compresses the trust building period that would otherwise take a decade into something closer to a year or two. That compression is its true economic value, and it explains why accreditation matters most for entrants and least for incumbents. But borrowed trust must eventually be replaced by earned trust. A borrowed reputation is a loan, and the repayment schedule is performance. Organizations that use the borrowed period to build capability convert the loan into equity. Organizations that use it only to sell convert the loan into debt, and the debt is called in the moment that delivery fails. 7.3 Quality Assurance as a Strategic Choice Rather Than an Obligation The final discussion point is a reframing. In most organizations, quality assurance sits in a compliance function, reports to an administrative office, and is budgeted as a cost. If the arguments in this article are correct, that placement is a strategic mistake. Quality assurance produces two of the most valuable assets an organization can own in an unfamiliar market, meaning credibility with outsiders and capability inside. Both are directly connected to growth, pricing, partnerships, and survival. The function therefore belongs close to strategy, and its outputs belong in the marketing conversation, not in a drawer. The organizations that gain most from accreditation are those that treat the quality director and the marketing director as colleagues working on the same problem, which is the problem of being believed. 8. Practical Implications The following roadmap translates the model into managerial action. It is written for institutional leaders, quality directors, marketing teams, and entrepreneurs preparing to enter a market where they are unknown. 8.1 Before Applying: Decide What the Mark Must Do Begin with the decision the organization needs to change. Is the goal to win ministry approval, to attract international students, to be selected by a multinational employer, to secure a partnership with a foreign university, to enter a corporate supply chain, or to justify a higher price? Each goal implies a different audience, and each audience reads different marks. Choosing an accreditation because competitors have it is a weak reason. Choosing it because a specific audience requires it is a strong one. Map the audiences first, then map the marks, then choose. This is #strategic_sequencing. 8.2 Assess Readiness Honestly Accreditation reveals what is there. It does not create what is absent. An organization with weak data systems, unclear outcomes, and no habit of acting on evidence will find the process punishing and will likely fail or pass only ceremonially. A useful readiness test asks three questions. Can we state our intended outcomes clearly? Can we show evidence that we measure them? Can we show at least one example of changing something because the evidence told us to? An organization that cannot answer all three should build for a year before applying. The building is the point. 8.3 Budget the Full Cost Budget the visible costs, including fees, visits, and consultants. Then budget the invisible ones, including senior leadership time, staff time, systems, and the opportunity cost of projects deferred. Then add the maintenance cost, since accreditation must be renewed and continuous improvement must be demonstrated. A realistic multi year budget prevents the common failure in which an organization exhausts itself achieving the mark and then has nothing left for the communication stage that would have made the mark valuable. 8.4 Use the Process to Fix Real Problems Treat the self study as a genuine audit rather than a persuasion exercise. The peer review team is one of the few outside groups that will ever look carefully at the organization and say what it sees. Ask for the criticism. Record it. Act on it visibly. Assign ownership of each finding to a named person with a deadline. Report progress publicly inside the organization. This converts the accreditation cycle into an improvement engine, which is the #continuous_improvement outcome that produces durable advantage. 8.5 Plan Communication From Day One Communication should be designed before the mark is awarded, not after. Prepare three assets in advance. The first is a translation table that converts each accreditation claim into a stakeholder benefit in plain language. The second is a proof set, meaning the specific outcomes that the accreditation verifies, such as graduate employment, learning gains, defect rates, or service response times, since numbers carry the signal further than logos. The third is a #third_party_voice plan, meaning the employers, alumni, partners, and professional bodies who will say it on the organization's behalf, because a claim made by others is worth more than the same claim made by the organization. 8.6 Communicate Consequences, Not Compliance The logo is not the message. The consequence is the message. Weak communication says that the institution is accredited by a named body. Strong communication says that the qualification is recognized by employers in specific countries, that the curriculum is independently checked every five years, that students who transfer are accepted by named partner institutions, and that graduates are eligible for specific professional pathways. Same fact, different value. For consumer audiences, use stories and outcomes. For institutional audiences, use documents and data. For regulators, use compliance mapping. #Message_segmentation is not a luxury. 8.7 Protect the Promise Once the claim is public, delivery becomes a brand risk as well as an operational one. Put in place a monitoring system that detects service failures early, especially in the areas the accreditation is known for. Empower front line staff to fix problems immediately. Track complaints as a leading indicator of brand damage. The rule is simple. Never promise externally what the organization cannot deliver internally on its worst day. 8.8 Build the Second Layer of Advantage From the day the mark is awarded, the clock starts on imitation. Use the exclusivity window to build what rivals cannot copy quickly. Options include long term employer partnerships with joint curriculum design, a distinctive pedagogy or service model, proprietary data on outcomes, a strong alumni network in the new market, deep local community ties, and a workforce with unusual skills. These form the #second_layer, and they are what remains when every competitor holds the same certificate. 8.9 Measure the Right Things Measure the effect of accreditation at the top of the decision funnel, where it actually operates. Useful indicators include change in unaided awareness among the target group, change in the proportion of prospects who include the organization in their shortlist, change in the length of the sales or admissions cycle, change in the number of inbound partnership approaches, change in the quality of applicants and hires, and change in willingness to pay. Lagging indicators such as revenue will move eventually, but attributing revenue to accreditation is difficult and often unconvincing. Leading indicators tell management whether the signal is being received, which is the question that matters while there is still time to act. 8.10 Avoid the Common Traps Five traps recur. The first is the collector trap, meaning the accumulation of marks for prestige without a defined audience for each. The second is the silence trap, meaning heavy investment in earning the mark and negligible investment in explaining it. The third is the ceremony trap, meaning documentation without change, which converts a brand asset into a brand risk. The fourth is the imitation trap, meaning the assumption that the badge will keep working after competitors obtain it. The fifth is the mill trap, meaning association with an unrecognized accreditor, which is worse than having no accreditation at all because it signals either ignorance or dishonesty. 9. Limitations and Future Research 9.1 Limitations This study is a conceptual synthesis and does not test its propositions empirically. The model is therefore a hypothesis about how the world works rather than a demonstration that it does. The evidence base is also uneven. Business schools and manufacturing firms dominate the literature. Healthcare, engineering, vocational training, and professional services are less represented, and the model may need adjustment in those settings. Similarly, most of the accessible research is written in English and published in Western indexed journals, which may under represent the experience of institutions in regions where accreditation dynamics differ considerably. Publication bias is a further concern. Studies finding no effect of certification are less likely to be published, which means the true average effect of accreditation may be smaller than the literature suggests. Readers should treat the positive findings reported here with appropriate caution. Finally, the model is deliberately general. It applies across sectors, which is a strength for theory but a weakness for precision. A specific sector, a specific market, and a specific accreditation will each carry conditions this framework does not capture. 9.2 A Research Agenda Six questions deserve empirical attention. The first is measurement of #stakeholder_awareness across markets and stakeholder groups. Existing evidence suggests it is low among consumers, but the research is thin and geographically narrow. Comparative surveys across countries and sectors would establish how large the recognition gap really is and where it is smallest. The second is the separation of signal from substance. Studies should measure the depth of internal implementation independently of the possession of a certificate, and then test which one drives which outcome. This would resolve much of the current contradiction in the literature. The third is the threshold effect. At what point does an additional accreditation stop adding brand value? Research on multiple and overlapping certification suggests diminishing returns, but the shape of the curve is unknown. The fourth is decay. How long does the advantage from a first mover accreditation last in a given market before imitation erases it? Longitudinal studies tracking entry cohorts would answer this. The fifth is the negative case. What happens to brands when accredited organizations fail publicly? The theory predicts an amplified penalty. This has not been tested. The sixth concerns new forms of verification. #Digital_credentials, verified outcome data, real time performance dashboards, and independently audited results may eventually offer buyers direct evidence rather than institutional proxies. If buyers can see outcomes directly, the mediating role of the accrediting body weakens. Whether accreditation retains its signalling power in a world of transparent data is an open and important question. 10. Conclusion Organizations entering unfamiliar markets face a problem that no amount of promotion can solve. They must be believed before they have earned the right to be believed. #Quality_assurance and #international_accreditation offer a way through, because they allow an organization to borrow the credibility of an institution the market already trusts. This article has argued that the value of that borrowing is real but conditional. Accreditation reduces uncertainty, opens the consideration set, unlocks partnerships, and shortens the long slow work of building trust. Yet it does none of this automatically. It works only when the audience recognizes the accreditor, only when the message is translated into consequences the audience cares about, only when the organization actually delivers what the mark implies, and only for as long as competitors have not yet obtained the same mark. Three conclusions follow. First, accreditation should be managed as a #brand_asset, which means it belongs in the strategy conversation and not only in the compliance file. Second, communication is not the final step of an accreditation project but half of the investment, and the half that is most often neglected. Third, the certificate is the beginning of advantage and never the end of it, because the only part of quality assurance that competitors cannot copy is the capability that the organization built while earning it. The organizations that win in new markets are not the ones with the most badges. They are the ones that used the badge to buy time, and used the time to become genuinely worth choosing. The certificate says the promise is credible. Only the organization can make the promise true. #quality_assurance #international_accreditation #brand_equity #competitive_advantage #signaling_theory #market_entry #perceived_quality #institutional_legitimacy #global_standards #trust_building #higher_education_branding #ISO_certification #emerging_markets #continuous_improvement #reputation_management References Alajoutsijarvi, K., Kettunen, K., and Sohlo, S. (2018). Shaking the status quo: Business accreditation and positional competition. Academy of Management Learning and Education, 17(2), 203-225. DOI 10.5465/amle.2015.0199 Aldhobaid, M. (2024). Quality assurance struggle in higher education institutions: Moving towards an effective quality assurance management system. Higher Education, 88, 1547-1566. Ashraf, M. A., Ratan, S. R. A., Amir, T., Joarder, M. H. R., and Osman, A. R. (2025). Faculty perspectives on accreditation, academic freedom and quality learning in private higher education institutes: Empirical evidence. Quality Assurance in Education, 33(1), 93-108. Attree, K., Neher, A., and Xie, G. (2025). Are accreditation signals being recognised? Business professionals awareness and views on accredited university business programmes. Studies in Higher Education, 51(1), 66-87. DOI 10.1080/03075079.2025.2456571 Bafera, J., and Kleinert, S. (2023). Signaling theory in entrepreneurship research: A systematic review and research agenda. Entrepreneurship Theory and Practice, 47(6), 2419-2464. DOI 10.1177/10422587221138489 Bakhtiar, A., Nugraha, A., Suliantoro, H., and Pujotomo, D. (2023). The effect of the QMS (ISO 9001) on operational performance of various organizations in Indonesia. Cogent Business and Management, 10(2), 2203304. DOI 10.1080/23311975.2023.2203304 Barney, J. B. (1991). Firm resources and sustained competitive advantage. Journal of Management, 17(1), 99-120. Carvalho, N., Rosa, M. J., and Amaral, A. (2023). Cross-border higher education and quality assurance: Results from a systematic literature review. Journal of Studies in International Education. DOI 10.1177/10283153221076900 Ed-Dafali, S., Al-Azad, M. S., Mohiuddin, M., and Reza, M. N. H. (2023). Strategic orientations, organizational ambidexterity, and sustainable competitive advantage: Mediating role of Industry 4.0 readiness in emerging markets. Journal of Cleaner Production, 401, 136765. Esteky, S., and Kalati, S. H. (2024). The educational placebo effect: Branding higher education benefits student learning. Journal of Marketing for Higher Education, 34(1), 246-261. DOI 10.1080/08841241.2021.1975186 Gardiana, M., Rahmanto, A., and Satyawan, I. (2023). International branding of higher education institutions towards world-class universities: Literature study in 2017-2022. Journal of Social and Political Sciences, 6(1). DOI 10.31014/aior.1991.06.01.392 Keller, K. L., and Swaminathan, V. (2020). Strategic brand management: Building, measuring, and managing brand equity (5th ed.). Pearson. Khanna, K., Singh, J., and Bedi, S. S. (2025). Mechanism for developing brand equity of higher education institutes using intangible resources: A serial-multiple mediation analysis. Marketing Intelligence and Planning, 43(3), 541-573. DOI 10.1108/MIP-11-2023-0645 Milovanovic, V., Paunovic, M., and Casadesus, M. (2023). Measuring the impact of ISO 9001 on employee and customer related company performance. Quality Innovation Prosperity, 27(1). Perera, C. H., Nguyen, L. T. V., and Nayak, R. (2023). Brand engagement on social media and its impact on brand equity in higher education: Integrating the social identity perspective. International Journal of Educational Management, 37(6/7), 1335-1359. DOI 10.1108/IJEM-05-2023-0260 Rizard, S. R., Waluyo, B., and Jaswir, I. (2023). Impact of brand equity and service quality on the reputation of universities and students intention to choose them: The case of IIUM and UIN. F1000Research, 11, 1412. DOI 10.12688/f1000research.122386.3 Singh, S., Dutta, E., Soodan, V., Jassal, T., and Kumar, R. (2025). Comprehensive analysis of the accreditation status of Indian universities: Evaluating NAAC quality indicators for continuous improvement in higher education. Journal of Engineering Education Transformations, 140-150. Spence, M. (1973). Job market signaling. Quarterly Journal of Economics, 87(3), 355-374. Sziegat, H. (2021). The response of German business schools to international accreditation in global competition. Quality Assurance in Education, 29(2/3), 135-150.

Startups live and die by speed. They ship fast, learn fast, and often survive on the difference between a decision made today and the same decision made three weeks later. Since the arrival of accessible large language models, that instinct for speed has met a technology that is unusually easy to adopt and unusually hard to see. Employees and #founders now use generative tools that no one approved, no one logged, and no one reviewed. This behaviour has come to be called #Shadow_AI. This article examines Shadow AI in the specific context of the #startup_ecosystem, where the tension between #innovation_velocity and organisational control is sharper than anywhere else in the economy. Drawing on an integrative review of peer reviewed literature published mainly between 2021 and 2026, the paper builds on the older tradition of shadow IT research and extends it to account for what is different about #generative_AI: it produces content rather than merely storing or moving it, it is opaque to the people who use it, it can be wrong in confident and persuasive ways, and it sits outside the firm on infrastructure the firm does not own. The article develops a conceptual model, the #speed_control_paradox, which explains why the same conditions that make #startups innovative also make them structurally prone to ungoverned AI use. It then presents a typology of Shadow AI behaviours, a taxonomy of risks covering data exposure, intellectual property, model reliability, #security, and regulatory #compliance, and a set of arguments for the overlooked benefits of unsanctioned experimentation. The paper closes with a staged governance model designed for firms that cannot afford a #compliance department, and with implications for #founders, investors, regulators, and students preparing to enter technology ventures. The central claim is simple: banning Shadow AI in a startup does not remove it, it only removes the ability to see it. Keywords: Shadow AI, shadow IT, generative artificial intelligence, startup governance, #innovation velocity, AI #risk_management, large language models, #entrepreneurial technology adoption 1. Introduction There is a familiar scene in almost every small technology company today. A founder needs a pitch deck by tomorrow morning. A junior developer is stuck on a function that will not compile. A marketing hire has to produce twenty pieces of copy before a campaign launches. In each case, the fastest path forward is the same: open a chatbot in a browser tab, paste in whatever context is needed, and take what comes back. No purchase order. No #security review. No record that it happened. This is the everyday reality of #unsanctioned_AI use, and it has quietly become one of the most consequential governance problems of the current decade. The phenomenon is not new in kind. Information systems researchers have studied #shadow_IT for many years, describing how business units and individual employees deploy software and services outside the alignment of the formal IT organisation (Klotz et al., 2019). What is new is the scale, the speed, and the nature of the technology involved. A spreadsheet macro built without approval is a local problem. A #large_language_model that receives a customer database, a draft patent claim, or an unreleased source file is a different kind of problem entirely, because the data leaves the building and the reasoning behind the output cannot be inspected. Recent scholarship has begun to name this specific variant of the older phenomenon. Silic and Kind-Truller (2025) define Shadow AI as the unsanctioned use of artificial intelligence systems outside approved governance structures, and they argue that its generative, opaque, and increasingly autonomous character introduces risks that shadow IT research never had to address: hallucination, algorithmic #bias, and what they describe as governance drift. Their mixed methods study of professionals and executives shows that employees do not experience this behaviour as rule breaking. They experience it as getting the job done. The #startup context deserves separate treatment for three reasons that are not merely differences of degree. First, #startups are built to move quickly. The entire logic of early stage venture building rests on compressing learning cycles, testing assumptions cheaply, and reaching evidence of product market fit before capital runs out. Anything that adds friction to a decision is, in this context, expensive in a way it is not for an established firm. Rules that slow a large company by five percent can slow a startup out of existence. Second, #startups have almost no governance surface. A company of eleven people has no chief information #security officer, no #data_protection officer, no procurement committee, and frequently no written policy of any kind. There is no shadow to hide in, because there is no formal structure casting one. In such firms, almost all technology use is technically unsanctioned, since sanctioning is not a process anyone performs. Third, and most importantly, #startups hold assets that are disproportionately sensitive relative to their size. A young company's entire value may sit in a few thousand lines of code, an unpublished dataset, a customer list, or a strategy document. When these are pasted into a third party system with unclear retention terms, the firm is not risking a fraction of its value. It may be risking most of it. The result is a paradox. The organisations with the strongest incentives to adopt #generative_technology quickly are also the organisations least equipped to adopt it safely. And the standard response from the governance literature, which is to write policy, train staff, and audit #compliance, assumes an organisational apparatus that early stage ventures simply do not possess. This article addresses that gap. It asks four questions. Why does Shadow AI emerge in startups, and what mechanisms drive it? What forms does it take, and what risks does each form carry? Under what conditions, if any, does unsanctioned AI use create genuine value rather than only exposure? What kind of #governance is realistic for a firm with no compliance function and no time? The contribution is threefold. Theoretically, the paper extends #shadow_IT theory into the generative era by identifying the properties of AI that break the older models. Conceptually, it introduces the speed control paradox as an explanatory device for why the problem is structural rather than behavioural. Practically, it offers a staged, low overhead governance model calibrated to the resource reality of small ventures rather than to the compliance departments of large enterprises. The remainder of the paper is organised as follows. Section 2 reviews the literature across four streams. Section 3 develops the conceptual framework. Section 4 explains the method. Sections 5 and 6 present and discuss the analysis. Section 7 sets out the governance model. Section 8 draws implications, Section 9 states limitations, and Section 10 concludes. 2. Literature Review 2.1 From shadow IT to Shadow AI The intellectual ancestor of the current debate is the literature on shadow IT. Systematic reviews of that field identify a consistent pattern of causes: employees turn to unapproved tools when official systems are slow, missing, or badly matched to the work, and when the barriers to using alternatives are low or absent (Klotz et al., 2019). The consequences documented in that literature are also consistent. Shadow IT produces local efficiency and genuine #innovation on one side, and data fragmentation, security exposure, and integration debt on the other. Generative AI itself is defined in the information systems literature as a class of models that learn the structure of large bodies of content and then produce new content of the same kind, including text, code, images, and audio (Feuerriegel et al., 2024). The wider scholarly debate about what these systems mean for research, practice, and policy opened early and remains unresolved (Dwivedi et al., 2023; Storey et al., 2025), and a critical strand warns against accepting capability claims at face value (Narayanan and Kapoor, 2024). Three characteristics of #generative_AI break the assumptions embedded in that older work. The first is generativity. Traditional shadow IT stores, moves, or presents information. Generative systems produce new artefacts, including code that will run in production, text that will be sent to customers, and analysis that will inform decisions. The output enters the firm's work product directly, which means errors do not stay in a spreadsheet. They ship. The second is opacity. A shadow spreadsheet can be opened and read. Its logic is inspectable. A #large_language_model cannot be audited by the person using it. The user has no access to the training data, no view of the weights, and no reliable account of why a particular answer appeared. This makes #verification a matter of human judgement rather than technical review, and human judgement degrades under time pressure, which is the permanent condition of a startup. The third is externality. Shadow IT was frequently internal. Shadow AI is almost always external. The data crosses an organisational boundary into infrastructure operated by a third party, subject to that party's retention terms, jurisdiction, and security posture. This changes the risk from a control problem into a disclosure problem. Silic and Kind-Truller (2025) argue on this basis that Shadow AI should be understood not as a discipline failure by employees but as a socio-technical governance failure by organisations. Their empirical work shows that employees justify unsanctioned use in terms of #productivity and necessity, and that formal structures consistently fail to keep pace. A parallel line of theorising treats covert generative AI use as a form of #user_innovation rather than deviance, modelling the conditions under which employees choose to hide or reveal what they have built. This reframing matters, because a firm that treats Shadow AI purely as misconduct will design controls that suppress information, while a firm that treats it as unmanaged innovation will design controls that surface it. 2.2 The productivity evidence The reason Shadow AI is difficult to eliminate is that it works. The empirical record on generative AI and individual #productivity is unusually strong for such a young technology. In a preregistered experiment with college educated professionals performing realistic writing tasks, Noy and Zhang (2023) found that access to a chatbot cut the average time taken by roughly forty percent while raising output quality by roughly eighteen percent, with the largest gains going to the weakest performers. Brynjolfsson, Li and Raymond (2025), studying more than five thousand customer support agents in a real workplace over an extended period, found a fifteen percent increase in issues resolved per hour, again with the gains concentrated among less experienced workers. In software development, controlled work by Peng and colleagues (2023) reported that developers with an AI coding assistant completed a benchmark task substantially faster than those without. Two findings from this body of work are directly relevant to startups. The first is the #skill_levelling effect. Generative tools compress the performance distribution by helping the least experienced most. Startups are staffed disproportionately by generalists working outside their formal expertise: an engineer writing marketing copy, a designer negotiating a contract, a founder reading a term sheet. These are precisely the situations where the productivity gain is largest, which means the pull toward the tools is strongest exactly where the organisation is least able to check the output. The second is the jagged frontier. Dell'Acqua and colleagues (2023) found that access to a capable model improved performance on tasks inside its competence but degraded performance on tasks that fell just outside it, because users could not reliably tell the difference. In an environment without review, this is a serious concern. The failure mode is not the obviously wrong answer. It is the confidently wrong answer that a rushed team accepts because it looks right. 2.3 Startups, resources, and the logic of speed Research on artificial intelligence in small and medium enterprises consistently describes a democratisation story alongside a capability gap. Rajaram and Tinguely (2024) argue that generative AI gives smaller firms access to scale and creativity that previously required capital they did not have, while also warning that deployment depends on employee competence, leadership, culture, and relationships with technology partners, none of which small firms can assume. Survey evidence from the OECD (2025) covering thousands of small and medium enterprises across several countries found generative AI already in use in around a third of them, with adopters citing efficiency and innovation as the main benefits, and non adopters citing concerns about legal exposure and about what happens to the information they feed into the models. That last point is the crux. The barrier is not attitude. Very few small firms prohibit these tools. The barrier is that they do not know what the tools do with what they are given, and they do not have anyone whose job it is to find out. The entrepreneurship literature adds a further layer. Chalmers, MacKenzie and Carter (2021) note that artificial intelligence alters venture creation itself, changing which activities can be performed by a small team and which capabilities are needed to compete. Kanbach and colleagues (2024) analyse generative AI through a business model innovation lens and show that it affects value creation, delivery, and capture simultaneously, which is another way of saying that it does not stay in one department. Haefner and colleagues (2023) point out that implementing and scaling AI requires organisational routines that most firms build slowly, if at all, and Mariani and Dwivedi (2024) map how generative systems are reshaping the practice of innovation management itself. The picture that emerges is a mismatch of clocks. #Adoption happens in an afternoon. Capability building takes quarters. Governance, in most firms, takes years. 2.4 The risk literature The risks associated with unregulated generative technology are documented across several distinct research communities, and it is useful to keep them separate because they call for different responses. Model reliability. Large language models generate content that is plausible but not necessarily factual. The survey by Huang and colleagues (2025) provides a detailed taxonomy of #hallucination, distinguishing factuality failures from faithfulness failures and tracing their origins to data, training, and inference. Hallucination is not a bug awaiting a patch. It is a property of how these systems produce text, and while it can be reduced it has not been eliminated. For a startup, the exposure is direct: fabricated citations in an investor memo, an invented regulation in a compliance answer, a nonexistent library imported into a codebase. Security. The security literature identifies attack surfaces with no equivalent in conventional software. Greshake and colleagues (2023) demonstrated indirect #prompt_injection, in which instructions hidden inside content that a model reads can hijack the behaviour of an application built on top of it. Carlini and colleagues (2021) showed that training data can be extracted from language models, raising the question of what happens to text that a firm supplies. Broader surveys of language model security and privacy map a landscape that includes data poisoning, model theft, and jailbreaking (Yao et al., 2024). Kshetri (2023) documents how the same tools lower the cost of phishing and social engineering, which means Shadow AI is simultaneously a vulnerability and a weapon in the hands of attackers. Data and #privacy. When information is pasted into an external system, the firm loses control of it. Where retention is unclear, where the provider's terms permit use for training, or where the data crosses jurisdictions, the exposure may be legally significant regardless of whether any breach occurs. Personal data pasted into a consumer chatbot by an employee is still processed by the firm in the eyes of most #data_protection regimes. Broader risk taxonomies for language models cover discrimination, information hazards, misinformation, malicious use, and human computer interaction harms in a single structure (Weidinger et al., 2022), and a growing body of work catalogues the controversies attached to generative systems in business settings (Wach et al., 2023). Practitioner oriented scholarship on securing these systems is beginning to consolidate as well (Huang et al., 2024). Bias and fairness. Ferrara (2023) and others analyse how models reproduce and sometimes amplify patterns in their training data. A startup using a model to screen applicants or score leads may be importing discrimination it never designed and cannot detect. Intellectual property. Two directions of exposure exist at once. Proprietary material may leak outward into a provider's systems. Generated material may carry inward a claim from someone else's work, and its ownership status may be uncertain in a way that troubles acquirers and investors during #due_diligence. 2.5 AI governance frameworks A substantial governance literature now exists. Mantymaki and colleagues (2022) define organisational AI governance as the structures and practices that align AI use with the firm's obligations. Papagiannidis and colleagues (2023) identify practices, barriers, and outcomes empirically. Schneider and colleagues (2023) map governance mechanisms for business use. Berente and colleagues (2021) frame the management of AI as a problem of balancing autonomy, learning, and inscrutability. At the level of standards and law, three instruments dominate current practice. The NIST AI Risk Management Framework (2023) offers a voluntary structure organised around governing, mapping, measuring, and managing risk. ISO/IEC 42001 (2023) provides a certifiable management system for AI. The European Union's Artificial Intelligence Act (Regulation 2024/1689) introduces a risk tiered regulatory regime with obligations that scale with the potential for harm, and with specific provisions for general purpose models (Novelli et al., 2024). The problem for the present study is one of fit. These instruments were written with organisations in mind that have functions, roles, and audit trails. A seed stage company with fourteen employees cannot implement a management system. It can, at best, implement a habit. The literature has, so far, produced very little guidance that takes this constraint seriously. 2.6 The research gap Bringing these streams together, three gaps are visible. First, Shadow AI has been studied primarily in enterprises. The startup case, where the incentive to move fast is strongest and the capacity to govern is weakest, is largely absent from the peer reviewed record. Second, the literature is heavily weighted toward risk. The possibility that unsanctioned AI use functions as a distributed discovery mechanism, revealing which tools actually help before the firm commits resources, is acknowledged in a few recent contributions but not developed. Third, existing governance guidance does not scale down. There is no widely accepted model for a firm that must choose between writing an AI policy and shipping a feature, and will always choose the feature. This article addresses all three. 3. Conceptual Framework: The Speed Control Paradox 3.1 Statement of the paradox The framework proposed here rests on a single observation. The organisational properties that make a startup capable of rapid #innovation are the same properties that make it unable to regulate the technology that accelerates that innovation. Consider the four properties most consistently associated with #entrepreneurial speed. Low formalisation. Few written rules, few approval gates, decisions made in conversation rather than in process. Role fluidity. People work far outside their job titles, which means individual discretion is high. Resource scarcity. Every hour and every dollar is contested, so anything that does not directly advance the product is postponed. Existential urgency. The firm is running against a cash horizon, which converts delay into risk. Each of these is a documented enabler of speed. Each is also, precisely, a documented cause of shadow technology use. Low formalisation means there is no policy to violate. Role fluidity means employees face tasks they are not trained for, which is when they reach for assistance. Resource scarcity means governance work is never the priority. Existential urgency means that any control which costs a day will be bypassed by someone who does not have a day. The paradox, stated formally: in early stage ventures, #shadow_AI is not a deviation from the operating model. It is an expression of it. This has an important implication. Interventions that treat Shadow AI as an individual compliance failure will fail, because they target the person rather than the mechanism. A founder who sends an email banning chatbots has not changed a single one of the four properties above. They have only ensured that the next employee who uses one will not mention it. 3.2 The visibility function Two variables determine whether an instance of AI use is visible to the firm. The first is the cost of disclosure. How much time, friction, or reputational risk does an employee incur by telling someone what they used? In a firm with an approval process, disclosure means waiting. In a firm with a ban, disclosure means admitting a violation. In both cases, the rational move for a busy employee is silence. The second is the perceived benefit of the tool. The higher the productivity gain, the more the employee will pay in concealment effort to keep using it. The productivity evidence reviewed above suggests these gains are large, particularly for the inexperienced. Concealment is therefore well funded from the employee's point of view. Visibility, then, is a function of the gap between these two quantities. Governance regimes that raise the cost of disclosure without reducing the benefit of the tool do not reduce usage. They reduce reporting. This is the most important practical consequence of the framework, and it runs directly against the instinct of most #founders, whose first response to a discovered risk is to prohibit. The design principle that follows is: make the sanctioned path faster than the shadow path. Not safer. Not better documented. Faster. Nothing else will hold in an organisation whose defining constraint is time. 3.3 Three propositions From the framework, three propositions can be derived. They are offered here as testable statements for future empirical work rather than as findings of this study. Proposition 1. In ventures below a threshold of organisational formalisation, the prevalence of Shadow AI is independent of the existence of a written AI policy, and depends instead on the friction of the approved alternative. Proposition 2. The severity of harm from Shadow AI is not proportional to the frequency of use. It is proportional to the sensitivity of the data involved and the degree to which output is used without #verification. A firm with high usage and good data hygiene may be at less risk than a firm with low usage and none. Proposition 3. Firms that create a low cost disclosure channel will observe higher measured Shadow AI and lower realised harm than firms that prohibit it, because measurement and harm are inversely related in this domain. The third proposition is deliberately counterintuitive. It predicts that a company doing #governance well will appear, on paper, to have a bigger problem than a company doing it badly. Any investor or board member reading a Shadow AI report should understand this before drawing conclusions from the numbers. 4. Method 4.1 Design This study is a conceptual paper built on an integrative review of the literature. It does not report new survey or interview data, and no claims of empirical measurement are made. The purpose of an integrative review is to synthesise findings from separate research streams that have not previously been brought into contact, and to produce a framework that organises them. That is appropriate here, because the relevant knowledge is scattered across information systems, entrepreneurship, economics, computer security, and law, and no single field holds the whole picture. 4.2 Sources and selection Literature was drawn from peer reviewed journals, conference proceedings in computer security and machine learning, scholarly books, and official standards and regulatory documents. The search covered work published mainly between 2021 and 2026, with a small number of older items retained where they are foundational to the shadow IT tradition and have no recent equivalent. Items were selected against four criteria. A source was included if it (a) addressed unsanctioned or ungoverned technology use in organisations, (b) reported evidence on the effects of generative AI on individual or firm level performance, (c) documented a specific technical or legal risk of large language models, or (d) proposed an AI governance structure, standard, or regulation. Purely technical papers on model architecture were excluded, as were commercial white papers where a peer reviewed alternative existed on the same point. Where industry survey figures are referenced in the wider debate, this article treats them with caution and does not reproduce specific vendor statistics as findings. Numbers circulating in the practitioner press on the prevalence of Shadow AI vary widely and rest on different definitions and sampling frames. The argument here does not depend on them. 4.3 Analysis The material was coded thematically into four categories: drivers, forms, risks, and responses. From this coding, a typology and a risk taxonomy were constructed, and the conceptual framework in Section 3 was developed abductively by asking what mechanism would have to be true for the observed patterns across the literature to hold together. 4.4 Illustrative scenarios Sections 5 and 6 use short scenarios to make abstract points concrete. These are composite illustrations constructed by the author to represent patterns described in the literature. They are not case studies, they do not describe identifiable firms, and they should not be read as evidence. 5. Analysis 5.1 Drivers of Shadow AI in startups Six drivers emerge from the synthesis. They are presented in rough order of force. Time compression. The most powerful driver is the simplest. The tool is faster. When a task that took four hours takes forty minutes, the decision to use it is not really a decision. This is consistent with the productivity findings of Noy and Zhang (2023) and Brynjolfsson and colleagues (2025), and it explains why exhortation is ineffective: the employee is not being careless, they are being rational within the incentives the firm has set. Capability substitution. In small teams, people routinely perform work for which they were never trained. Generative models close that gap instantly. A backend engineer can produce acceptable ad copy. A first time founder can produce a term sheet summary. The #skill_levelling effect documented in the experimental literature is, in a startup, not a marginal improvement but the difference between a task being possible and impossible. Absence of a sanctioned alternative. Shadow IT research established that unapproved tools flourish where approved ones are missing or inadequate. Most early stage firms have no approved AI tool at all. There is nothing to compare the shadow option against, and therefore nothing to choose. Zero acquisition cost. Adoption requires no budget, no procurement, and no installation. A browser tab is not a purchase. This removes the single mechanism through which small firms usually notice new technology entering the business, which is the invoice. Cultural signalling. Founders and investors reward speed and resourcefulness. An employee who finds a way to do three days of work in one is behaving exactly as the culture asks. Punishing the method while praising the result produces concealment, not compliance. Peer diffusion. Practices spread horizontally through communities of founders, accelerator cohorts, and online networks far faster than any internal policy propagates vertically. By the time leadership considers a rule, the practice is already normal. Two observations follow. First, five of these six drivers are structural rather than attitudinal, which means training will not remove them. Second, none of them require any employee to hold a negative view of governance. Shadow AI in startups is overwhelmingly produced by conscientious people doing their best under pressure. 5.2 A typology of Shadow AI Not all unsanctioned use is equivalent, and treating it as one category is the most common analytical error in the practitioner debate. Four types can be distinguished, ordered by increasing depth of integration into the firm. Type 1: Assistive use. An individual uses a public chatbot to draft, summarise, translate, or brainstorm. Nothing is integrated into a product. The main risk is #data_leakage through what is pasted in, and quality risk through what is copied out. This is the most common form and the least severe, provided the input is not sensitive. Type 2: Embedded output. Generated artefacts enter the firm's work product. AI written code is committed. AI written text is published. AI generated analysis informs a decision. The distinguishing feature is that the output now has downstream consequences the author cannot recall or trace. Risks include #hallucination reaching customers, insecure code entering production, and #intellectual_property of uncertain provenance being shipped. Type 3: Unsanctioned integration. Someone connects a model to the firm's systems through an interface, a plugin, or a browser extension, so that data flows automatically rather than by copy and paste. This is a step change in exposure. The volume of data crossing the boundary is no longer limited by human typing, and the connection persists after the person who built it has forgotten about it or left the company. Type 4: Autonomous agents. A model is given the ability to take actions, such as sending messages, modifying records, executing code, or calling other services. Here the risk profile changes qualitatively. #Prompt_injection becomes a live attack surface, because content the agent reads can influence what the agent does (Greshake et al., 2023). Accountability becomes genuinely difficult, because the chain from human intent to system action is long and partly opaque. The types differ not only in severity but in detectability. Type 1 leaves almost no trace. Type 4 leaves a great deal of trace, but usually only after something has gone wrong. Governance effort should be weighted accordingly, and this weighting is the basis of the model presented in Section 7. 5.3 Risk taxonomy Data exposure. The most immediate risk. Source code, customer records, financial models, medical or biometric information, unreleased product plans, and legal documents are routinely pasted into consumer grade tools whose retention terms the user has not read. The harm is not always a breach in the technical sense. It may simply be that the firm can no longer state, truthfully, where its confidential information resides. Intellectual property erosion. For a startup, the code and the data often are the company. Two distinct problems arise. Outbound: proprietary material supplied to a third party may be retained, and even where a provider commits not to train on it, the firm's ability to prove that during an acquisition is limited. Inbound: generated code or content may reproduce protected material, and its copyright status may be unsettled. #Due_diligence in later funding rounds increasingly asks how AI generated components entered the codebase, and a firm that cannot answer is negotiating from weakness. Reliability and hallucination. Models produce fluent falsehood. Huang and colleagues (2025) show this is intrinsic to how the systems work rather than incidental. In a startup, the review layer that would catch it is thin or absent. The dangerous case is not the absurd error, which is caught. It is the plausible error, which is not. Security. Beyond data exposure, three specific vectors matter. Prompt injection allows untrusted content to steer model behaviour. Insecure generated code introduces vulnerabilities that pass code review precisely because the code looks idiomatic. And attackers use the same tools, lowering the cost of convincing #social_engineering against small teams that have no security function (Kshetri, 2023; Yao et al., 2024). Regulatory and contractual exposure. Data protection obligations do not depend on firm size. Neither do contractual confidentiality clauses with enterprise customers. A single employee action can breach both. Under the European Union's risk tiered regime, obligations attach to the use case rather than to the size of the user (Regulation 2024/1689; Novelli et al., 2024), and a startup building a hiring tool or a credit scoring tool sits in a regulated category whether it knows it or not. Bias and fairness. Where models are used in screening, ranking, or scoring, they may reproduce discriminatory patterns from training data (Ferrara, 2023). A firm with no monitoring will not detect this, and a firm with no documentation cannot defend against the accusation. Technical and #epistemic_debt. Two slower burning problems. #Technical_debt accumulates as generated code enters the codebase without any team member holding a full mental model of it. Epistemic debt accumulates as decisions are made on analysis that no one has independently checked, so that the firm's understanding of its own market becomes, over time, a chain of unverified assertions. Concentration and dependency. Building critical workflows on external models creates exposure to price changes, deprecation, policy changes, and outages that the firm cannot control and, in the shadow case, has not even inventoried. 5.4 The overlooked upside A literature dominated by risk tends to forget that the behaviour persists because it produces value. Four benefits deserve explicit statement, not because they cancel the risks but because a governance model that denies them will be rejected by the people it is meant to guide. Distributed discovery. Employees experimenting independently produce, in aggregate, a survey of which tools actually help with the firm's specific work. This is information the company would otherwise have to buy through a formal evaluation it cannot afford. Recent theorising on covert generative AI use as a form of #user_innovation makes exactly this point: the shadow user is often the firm's most effective scout. Capability building. Practical fluency with these systems is now a competitive skill. Firms whose staff have used them extensively, even informally, hold an advantage over firms whose staff have been forbidden from touching them. Survival economics. For a company with three months of #runway, a tool that halves the time to a deliverable is not a convenience. It may be the reason the company still exists at the next board meeting. Any risk calculation that ignores the risk of not shipping is incomplete. Signal about internal failure. Every instance of Shadow AI is a report from the front line about where the firm's official tooling is inadequate. Read this way, the phenomenon is diagnostic. A team that quietly uses an outside model to summarise support tickets is telling leadership that the support workflow is broken. The correct posture, then, is neither #prohibition nor permissiveness. It is #visibility. A firm that knows what is being used, on what data, for what purpose, can make proportionate decisions. A firm that has banned everything knows nothing. 6. Discussion 6.1 Governance must be staged, not uniform The dominant frameworks assume a stable organisation. Startups are not stable. They pass through phases with different risk profiles, and a governance model that ignores this will be either useless at the start or inadequate later. Pre-seed and seed. Two to fifteen people. No customer data of consequence, or very little. The realistic goal is not control but hygiene: a shared understanding of what must never be pasted into an external tool, and a single list of what people are actually using. Anything heavier will be ignored, and correctly so. Series A. Fifteen to sixty people. Real customers, real contracts, real data. This is the point at which the first enterprise customer's security questionnaire arrives, and the firm discovers that it cannot answer it. This is the moment when governance must be built, and it is almost always built late. Series B and beyond. The firm now has a security function, and the shadow problem becomes a conventional enterprise problem, addressable through the existing literature. The damage, however, was usually done earlier. Data that left the building at seed stage does not come back. The practical lesson is that the highest leverage governance intervention in a startup's life happens when the firm is too small to think it needs one. 6.2 The founder as the primary vector Discussions of Shadow AI in large firms focus on employees. In startups this is a misreading. The founder is frequently the heaviest user, and the founder handles the most sensitive material in the company: the cap table, the financial model, the fundraising narrative, the strategy, the legal documents, the personnel decisions. A founder who pastes a full investor update into a public model has performed an act with more potential consequence than anything a junior employee could do. And no one in the firm has the standing to question it. Governance in a startup is therefore a leadership behaviour before it is a policy document. Whatever the founder does at midnight before a board meeting becomes, in effect, the firm's real policy. 6.3 Why prohibition fails, and what it costs The failure of bans follows directly from the visibility function in Section 3.2. A ban raises the cost of disclosure without touching the benefit of the tool. Usage does not stop. It moves to personal devices and personal accounts, where the firm has no logging, no data controls, and no ability to respond to an incident. Prohibition also carries three costs that are rarely counted. It destroys the diagnostic signal described in Section 5.4. It selects against the most productive employees, who will either comply and become slower or leave for a firm that lets them work. And it teaches the organisation that honesty about tooling is punished, which is a lesson that will generalise to other subjects. This is not an argument for permitting everything. It is an argument that the choice is not between control and chaos. It is between #informed_risk and blind risk. 6.4 The regulatory horizon The regulatory environment is tightening, and small firms are not exempt. The European Union's Artificial Intelligence Act imposes obligations according to the risk of the application rather than the size of the provider, with the heaviest requirements falling on high risk uses such as employment, credit, and essential services (Regulation 2024/1689). Novelli and colleagues (2024) note that assessing where a system falls in this scheme is itself a non trivial exercise. Voluntary instruments including the NIST AI Risk Management Framework (2023) and ISO/IEC 42001 (2023) provide structure but presuppose organisational capacity. For a startup the strategic point is this: a company that cannot describe its own AI usage cannot demonstrate compliance with any of these regimes, and increasingly cannot pass the procurement review of the enterprise customer it is trying to win. #AI_governance is therefore becoming a commercial asset rather than only a legal cost. The first startup in a category that can hand a prospective customer a clear, honest account of how it uses AI, on what data, with what controls, will win deals its competitors cannot. 6.5 The investor dimension Investors occupy an unusual position. They demand growth, which drives the behaviour, and they demand clean diligence, which the behaviour compromises. Very few term sheets currently ask about AI usage. This will change, and it will change abruptly, most likely after the first well publicised failure of a funding round or acquisition on these grounds. The reasonable ask from an investor is not a policy document. It is three questions. What models do you use? What data touches them? Who checks the output before it reaches a customer? A founder who cannot answer these has an unquantified liability on the balance sheet, and a founder who can has a competitive story to tell. 7. A Proportionate Governance Model for Small Ventures The model below is designed against a single constraint: every element must be implementable by a company with no compliance function, in under a day, and must not slow anyone down. Anything that fails this test will not survive contact with a startup. It has five components, summarised as CLEAR: Classify, List, Enable, Assure, Revisit. 7.1 Classify the data, not the tool Most policies try to control which tools people use. This is the wrong axis, because tools change weekly and enforcement is impossible. Control the data instead, using three bands that a person can hold in their head. Red. Never leaves the firm. Customer personal data, credentials and keys, unreleased core source code, cap table and financials, legal documents under privilege, anything covered by a confidentiality clause. Amber. Permitted only in tools the firm has approved, and only where the provider commits not to train on the input. Internal drafts, aggregated or anonymised data, non public but non critical material. Green. Anything already public, or wholly synthetic, or entirely trivial. Use any tool. The value of this scheme is that an employee under pressure at eleven at night can apply it without reading a document. A rule that requires interpretation will not be applied. 7.2 List what is in use Maintain a single shared register with four columns: tool, owner, purpose, data band. It can be a spreadsheet. It should take ten minutes to add an entry, and adding an entry must never trigger an approval process, because the moment it does, people stop adding entries. The register is not a control. It is a map. Its only job is to convert the unknown into the known, which is the precondition for every other decision the firm will make. 7.3 Enable a fast sanctioned path This is the component most firms skip, and it is the one that determines whether the rest works. Provide a #sanctioned_tool that is at least as fast and at least as capable as the shadow alternative, on terms that protect the firm's data. If the approved option is slower, more restricted, or more annoying than the browser tab, the browser tab wins, and no amount of policy will change that. The design principle from Section 3.2 applies without exception: make the sanctioned path faster than the shadow path. 7.4 Assure the output where it matters Verification effort should be concentrated where the consequences are real. A short rule of thumb, applied to any AI generated artefact, is sufficient for most small firms. Does it go to a customer, a regulator, or an investor? Does it execute in production? Does it inform an irreversible decision? If the answer to any of these is yes, a named human reviews it and is accountable for it. If the answer to all three is no, use it and move on. The purpose is not to review everything. It is to stop the firm from pretending to review everything, which is what a blanket rule produces in practice. 7.5 Revisit on a fixed cadence Once a quarter, and immediately on any material change such as a first enterprise customer, an entry into a regulated market, or a security incident, the founders spend one hour on three questions. What changed in the register? What did we learn about which tools actually help? What data band assumptions no longer hold? One hour a quarter is a defensible cost. Anything larger will be cancelled the first time a release slips, and a governance practice that is cancelled under pressure was never a governance practice at all. 7.6 What the model deliberately omits The model contains no training programme, no approval workflow, no tooling committee, and no audit. These are appropriate for firms with the capacity to sustain them, and they will arrive naturally as the company grows. Introducing them at seed stage produces documents that describe an organisation that does not exist, which is worse than having nothing, because it creates a false record of control. 8. Implications 8.1 For theory The paper extends shadow IT theory by identifying three properties, generativity, opacity, and externality, that break its assumptions, and it proposes the #speed_control_paradox as an explanation for why the phenomenon is structural in entrepreneurial firms rather than behavioural. The visibility function offers a mechanism that can be tested, and the three propositions in Section 3.3 are stated in a form that permits falsification. The reframing of Shadow AI as partly a discovery process, rather than only a control failure, connects the governance literature to work on user innovation and suggests that the two have been talking past each other. 8.2 For founders and managers Four points carry the practical weight. First, assume it is already happening in your company, because the evidence says it almost certainly is. Second, stop trying to control tools and start controlling data. Third, your own behaviour sets the real policy, whatever the written one says. Fourth, the sanctioned path must be faster than the shadow path, or nothing else you do will matter. 8.3 For investors and boards Ask the three questions in Section 6.5 at the first meeting rather than at diligence. Recognise the counterintuitive prediction of Proposition 3: a portfolio company reporting extensive AI usage is more likely to be well governed than one reporting none, because the second is probably not measuring. Treat an inability to describe AI usage as a material unknown, in the same way an inability to describe revenue recognition would be. 8.4 For policymakers Regulation calibrated to large firms will be complied with on paper and ignored in practice by small ones, not out of defiance but out of incapacity. There is a strong case for lightweight, template driven compliance instruments aimed at firms under a certain headcount, and for public guidance that treats the small venture as a distinct regulatory subject rather than a scaled down enterprise. Nothing in the current instruments prevents this. It simply has not been done. 8.5 For students and educators For students preparing to work in or found technology ventures, three competences follow from this analysis and none of them are currently taught systematically. The first is #data_classification as an instinct: knowing, without looking anything up, what may and may not leave the organisation. The second is verification literacy: the habit of checking generated output against a source, especially when it is fluent and especially when it is convenient. The third is the ability to hold both truths at once, that these tools create real value and real exposure, and that a professional judgement is required rather than a slogan. Students should also understand that #AI_literacy is no longer only a technical skill. It is a governance skill, and the graduate who can write a one page AI usage note that a customer's security team will accept is more valuable to an early stage firm than the graduate who cannot. 9. Limitations and Future Research This is a conceptual paper, and it inherits the limitations of that form. No primary data was collected, and the framework has not been empirically tested. The three propositions are arguments, not findings. The literature base is uneven. Peer reviewed work specifically on Shadow AI is very recent and thin, and the practitioner material that fills the gap varies enormously in quality and definition. The productivity evidence, though strong, comes largely from experiments and large firm deployments rather than from startups, and its transfer to the entrepreneurial context is an inference rather than a demonstration. The analysis is also geographically and sectorally general. Regulatory exposure differs sharply between jurisdictions, and a health technology venture faces a different risk profile from a consumer application. The governance model in Section 7 is a floor, not a ceiling, and firms in regulated sectors will need more. Five directions for future research follow. Prevalence and form in startups. Direct survey and interview work with early stage firms, using a definition consistent enough to allow comparison, would replace the current reliance on enterprise data and vendor claims. Testing the visibility function. Do firms with low cost disclosure channels genuinely observe more usage and less harm, as Proposition 3 predicts? This is measurable. Longitudinal work on the governance transition. What happens at the moment a startup is forced to formalise, typically at the first enterprise contract? What is recoverable and what is not? The diligence effect. As investors begin to ask about AI usage, what happens to firm behaviour, and does the change improve outcomes or only improve documentation? The agentic shift. As Type 4 usage grows, the #accountability questions become far harder. Almost no scholarly work addresses ungoverned #autonomous_agents inside small firms, and this is likely to be the most consequential gap within a very short time. 10. Conclusion Shadow AI in startups is not a story about careless employees. It is a story about an organisational form that was built for speed meeting a technology that rewards speed, in an environment where the mechanisms that would normally provide oversight do not yet exist. The argument of this paper can be reduced to four claims. Unsanctioned generative technology use in early stage ventures is structural, not accidental, and follows directly from the properties that make those ventures innovative. It carries risks that are qualitatively different from those of earlier shadow technology, because the systems involved generate content, cannot be inspected, and sit outside the firm. It also carries benefits that the risk focused literature has consistently understated, including distributed discovery of what actually works and a diagnostic signal about where the firm's own tooling is failing. And it cannot be managed by prohibition, because prohibition destroys visibility while leaving usage intact. What remains is an unglamorous but achievable programme. Classify the data. List what is in use. Make the approved path the fastest one. Verify what actually matters. Revisit it on a schedule. None of this requires a compliance department. All of it requires a decision by the founders that the firm would rather know than not know. The startups that get this right will not be the ones that use these tools least. They will be the ones that can say, precisely and without embarrassment, what they use, on what, and why. In a market where the next enterprise customer, the next regulator, and the next acquirer will all ask that question, being able to answer it is not a constraint on speed. It is what makes speed sustainable. Hashtags #ShadowAI #ShadowAI_in_Startups #StartupEcosystem #GenerativeAI_Risk #UnregulatedAI #AIGovernance_for_Startups #RapidInnovation #ResponsibleInnovation #DataSecurity_in_Startups #LLM_Risk #AI_Compliance #TechEntrepreneurship #AI_Policy #InnovationVsControl #FutureOfWork_AI References Berente, N., Gu, B., Recker, J., and Santhanam, R. (2021). Managing artificial intelligence. MIS Quarterly, 45(3), 1433-1450. Brynjolfsson, E., Li, D., and Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889-942. https://doi.org/10.1093/qje/qjae044 Carlini, N., Tramer, F., Wallace, E., Jagielski, M., Herbert-Voss, A., Lee, K., Roberts, A., Brown, T., Song, D., Erlingsson, U., Oprea, A., and Raffel, C. (2021). Extracting training data from large language models. In Proceedings of the 30th USENIX Security Symposium (pp. 2633-2650). USENIX Association. Chalmers, D., MacKenzie, N. G., and Carter, S. (2021). Artificial intelligence and entrepreneurship: Implications for venture creation in the fourth industrial revolution. Entrepreneurship Theory and Practice, 45(5), 1028-1053. Dell'Acqua, F., McFowland, E., Mollick, E., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F., and Lakhani, K. R. (2023). Navigating the jagged technological frontier: Field experimental evidence of the effects of AI on knowledge worker productivity and quality. Harvard Business School Working Paper 24-013. Harvard Business School. Dwivedi, Y. K., Kshetri, N., Hughes, L., Slade, E. L., Jeyaraj, A., Kar, A. K., et al. (2023). Opinion paper: So what if ChatGPT wrote it? Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, practice and policy. International Journal of Information Management, 71, 102642. European Parliament and Council of the European Union. (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union. Ferrara, E. (2023). Should ChatGPT be biased? Challenges and risks of bias in large language models. First Monday, 28(11). Feuerriegel, S., Hartmann, J., Janiesch, C., and Zschech, P. (2024). Generative AI. Business and Information Systems Engineering, 66(1), 111-126. Greshake, K., Abdelnabi, S., Mishra, S., Endres, C., Holz, T., and Fritz, M. (2023). Not what you've signed up for: Compromising real-world LLM-integrated applications with indirect prompt injection. In Proceedings of the 16th ACM Workshop on Artificial Intelligence and Security (pp. 79-90). Association for Computing Machinery. Haefner, N., Parida, V., Gassmann, O., and Wincent, J. (2023). Implementing and scaling artificial intelligence: A review, framework, and research agenda. Technological Forecasting and Social Change, 197, 122878. Huang, K., Wang, Y., Goertzel, B., Li, Y., Wright, S., and Ponnapalli, J. (Eds.). (2024). Generative AI security: Theories and practices. Springer. Huang, L., Yu, W., Ma, W., Zhong, W., Feng, Z., Wang, H., Chen, Q., Peng, W., Feng, X., Qin, B., and Liu, T. (2025). A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions. ACM Transactions on Information Systems, 43(2), Article 42. https://doi.org/10.1145/3703155 International Organization for Standardization and International Electrotechnical Commission. (2023). ISO/IEC 42001:2023 Information technology, artificial intelligence, management system. ISO. Kanbach, D. K., Heiduk, L., Blueher, G., Schreiter, M., and Lahmann, A. (2024). The GenAI is out of the bottle: Generative artificial intelligence from a business model innovation perspective. Review of Managerial Science, 18(4), 1189-1220. Klotz, S., Kopper, A., Westner, M., and Strahringer, S. (2019). Causing factors, outcomes, and governance of shadow IT and business-managed IT: A systematic literature review. International Journal of Information Systems and Project Management, 7(1), 15-43. https://doi.org/10.12821/ijispm070102 Kshetri, N. (2023). Cybercrime and privacy threats of large language models. IT Professional, 25(3), 9-13. Mantymaki, M., Minkkinen, M., Birkstedt, T., and Viljanen, M. (2022). Defining organizational AI governance. AI and Ethics, 2, 603-609. Mariani, M., and Dwivedi, Y. K. (2024). Generative artificial intelligence in innovation management: A preview of future research developments. Journal of Business Research, 175, 114542. Narayanan, A., and Kapoor, S. (2024). AI snake oil: What artificial intelligence can do, what it cannot, and how to tell the difference. Princeton University Press. National Institute of Standards and Technology. (2023). Artificial intelligence risk management framework (AI RMF 1.0), NIST AI 100-1. U.S. Department of Commerce. Novelli, C., Casolari, F., Rotolo, A., Taddeo, M., and Floridi, L. (2024). Taking AI risks seriously: A new assessment model for the AI Act. AI and Society, 39, 2493-2497. Noy, S., and Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381(6654), 187-192. https://doi.org/10.1126/science.adh2586 Organisation for Economic Co-operation and Development. (2025). Generative AI and the SME workforce. OECD Publishing. Papagiannidis, E., Enholm, I. M., Dremel, C., Mikalef, P., and Krogstie, J. (2023). Toward AI governance: Identifying best practices and potential barriers and outcomes. Information Systems Frontiers, 25, 123-141. Peng, S., Kalliamvakou, E., Cihon, P., and Demirer, M. (2023). The impact of AI on developer productivity: Evidence from GitHub Copilot. arXiv preprint arXiv:2302.06590. Rajaram, K., and Tinguely, P. N. (2024). Generative artificial intelligence in small and medium enterprises: Navigating its promises and challenges. Business Horizons, 67(5), 629-648. https://doi.org/10.1016/j.bushor.2024.05.008 Schneider, J., Abraham, R., Meske, C., and vom Brocke, J. (2023). Artificial intelligence governance for businesses. Information Systems Management, 40(3), 229-249. Silic, M., and Kind-Truller, K. (2025). From shadow IT to shadow AI: Threats, risks and opportunities for organizations. Strategic Change. https://doi.org/10.1002/jsc.2682 Storey, V. C., Yue, W. T., Zhao, J. L., and Lukyanenko, R. (2025). Generative artificial intelligence: Evolving technology, growing societal impact, and opportunities for information systems research. Information Systems Frontiers. Wach, K., Duong, C. D., Ejdys, J., Kazlauskaite, R., Korzynski, P., Mazurek, G., Paliszkiewicz, J., and Ziemba, E. (2023). The dark side of generative artificial intelligence: A critical analysis of controversies and risks of ChatGPT. Entrepreneurial Business and Economics Review, 11(2), 7-30. Weidinger, L., Uesato, J., Rauh, M., Griffin, C., Huang, P. S., Mellor, J., et al. (2022). Taxonomy of risks posed by language models. In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency (pp. 214-229). Association for Computing Machinery. Yao, Y., Duan, J., Xu, K., Cai, Y., Sun, Z., and Zhang, Y. (2024). A survey on large language model (LLM) security and privacy: The good, the bad, and the ugly. High-Confidence Computing, 4(2), 100211.

This article asks a simple question with complicated consequences: who, or what, actually makes strategy in a modern multinational firm? Standard answers point to boards, chief executives, and strategy teams. This paper argues that such answers are incomplete. Using #actor_network_theory (ANT) as its lens, the study treats strategy as the outcome of a working alliance between people and things, where contracts, sensors, dashboards, container ships, customs codes, port software, supplier scorecards, and #algorithmic_governance systems all shape what executives believe is possible. The paper develops an integrative conceptual review of three literatures that rarely speak to one another: international management and the theory of the #multinational_enterprise; #global_supply_chains and global value chain governance; and the fast-growing scholarship on #digital_platforms and platform-based internationalisation. From these, the paper builds a framework it calls the Strategy Translation Cycle, which adapts the classical ANT moments of #problematisation, #interessement, #enrolment, and #mobilisation to the setting of the globally distributed firm. Four propositions are developed. First, that the visible field of executive choice is narrowed and widened by #inscription devices long before deliberation begins. Second, that digital platforms increasingly occupy the position of the #obligatory_passage_point in cross-border exchange, converting commercial dependence into strategic dependence. Third, that when supply networks are stable, executives experience them as single objects rather than as networks, a condition ANT calls #punctualisation, and that #geopolitical_risk works precisely by breaking that condition. Fourth, that strategic failure in international business is often not a failure of analysis but a failure of enrolment, in which some actor, human or not, refuses the role assigned to it. The paper closes with methodological guidance for students who wish to conduct ANT-informed research, a discussion of the theory's well-known limits, and an agenda for future work on #digital_sovereignty, #artificial_intelligence in procurement, and the reconfiguration of supply networks under industrial policy. The contribution is to move international management away from a picture of the strategist as a sovereign mind facing an environment, and toward a picture of the strategist as a builder and repairer of heterogeneous networks whose stability can never be assumed. Keywords: actor-network theory; international management; global supply chains; digital platforms; executive strategy; sociomateriality; translation; global value chains 1. Introduction 1.1 The puzzle In most textbooks on international business, the strategist appears as a lonely and rather heroic figure. She scans the environment, weighs the options, calculates the trade-offs, and then commits the firm to a course of action. The environment is out there. The decision is in here. Between them sits the analytical apparatus of the field: entry mode models, ownership advantages, location advantages, internalisation logic, and a growing family of tools for measuring risk. Anyone who has spent time inside a large multinational knows that this picture is at best a useful simplification and at worst a serious distortion. Consider a plausible scene. A chief operating officer at a European electronics firm is deciding whether to move final assembly out of one country and into another. She does not begin with a blank page. She begins with a #dashboards view produced by an enterprise system that already classifies suppliers into tiers. Those tiers were defined years ago by a consultant. The system pulls landed cost figures that depend on customs classifications, and those classifications were set by a trade lawyer who has since left the firm. The system cannot show her tier-three suppliers at all, because no one built the data fields for them. Her risk map is colour-coded by an external ratings provider whose methodology she has never read. Her finance team has a #key_performance_indicators pack that rewards inventory turns, which quietly penalises the buffer stock that a more resilient design would require. By the time she thinks she is deciding, an enormous amount of deciding has already happened. It has happened in the software, in the categories, in the metrics, in the contract templates, in the physical geometry of a shipping container. The question this paper takes seriously is whether our theory should keep treating all of that as background noise, or whether it should be brought into the foreground and treated as part of the strategy itself. actor-network theory offers a way to do the second thing. Developed within science and technology studies, and associated above all with the work of Bruno Latour, Michel Callon, and John Law, ANT is less a theory in the predictive sense than a method of description and a set of instructions for what to notice. Its central and most provocative move is the principle of #symmetry: do not decide in advance that people act and objects merely sit there. Instead, follow the actors, whoever and whatever they turn out to be, and see what they do to each other. 1.2 Why now There are three reasons why this move matters more today than it did twenty years ago. The first is that the material substrate of #internationalisation has become far denser and far more visible. The pandemic years taught executives, journalists, and finance ministers a vocabulary they had never needed before: container rates, chip lead times, port dwell times, semiconductor lithography, critical mineral refining capacity. Objects that had been safely invisible became loudly present. A single vessel wedged sideways in a canal reorganised the working lives of thousands of managers who had never thought about that canal. This is exactly the condition ANT is built to describe: the moment when a stable network breaks and reveals itself to have been a network all along. The second reason is the rise of digital platforms as the connective tissue of cross-border commerce. Platform-based multinationals now sit among the largest firms on earth by market value, and their infrastructures increasingly mediate the transactions of everybody else. A mid-size exporter in an emerging economy may reach global customers through a marketplace, be discovered through a search algorithm, be paid through a payments rail, be reviewed by a rating system, and be logistically served by a fulfilment network, all of which belong to other firms. Its strategy is formally its own. Practically, it is written inside somebody else's architecture. Research on platform-based internationalisation has documented this pattern in detail, showing that platforms function simultaneously as enablers and as constraints, and that the strategies of platform-based firms diverge in important ways from those of traditional multinationals. The third reason is geopolitical. The steady, if uneven, integration that characterised the late twentieth century has given way to fragmentation, #techno_nationalism, and an aggressive return of #industrial_policy. Export controls, screening regimes, subsidy programmes, and outright bans have made the political character of supply networks impossible to ignore. Scholars of international business have responded with a growing body of work on #decoupling, on bifurcated governance, and on the reconfiguration of #global_value_chains under stress. What has been less developed is a theoretical vocabulary for describing how these forces enter the executive mind in the first place. ANT provides one, because it insists that political pressure does not arrive as an abstraction. It arrives as a licence application, a customs hold, a compliance module, a piece of software that will no longer update. 1.3 Purpose and contribution The purpose of this article is fourfold. First, it introduces ANT to students of #international_management in accessible language, without diluting the parts of the theory that make it demanding. Second, it reviews and connects three literatures that have grown up largely apart: the theory of the multinational enterprise, the study of global supply chains and value chains, and the emerging scholarship on digital platforms in international business. Third, it develops an original conceptual framework, the Strategy Translation Cycle, which shows how executive strategy is assembled, stabilised, disrupted, and reassembled through the four ANT moments of #translation. Fourth, it derives implications for practice and a research agenda, with practical guidance for students designing ANT-informed empirical projects. The contribution is theoretical rather than empirical. The paper does not claim that ANT should replace existing perspectives such as #transaction_cost_economics, the #resource_based_view, #institutional_theory, or #dynamic_capabilities. It claims something narrower and, we hope, more defensible: that these perspectives share an assumption, namely that agency belongs to firms and managers and that technology is a resource or a constraint acted upon by them, and that this assumption systematically hides an important part of what happens in international strategy. ANT relaxes the assumption. What becomes visible when it is relaxed is the subject of this paper. 1.4 Structure Section 2 sets out the theoretical foundations of ANT. Section 3 reviews the three relevant literatures. Section 4 presents the conceptual framework and four propositions. Section 5 describes the methodological approach of this review and offers guidance for ANT-informed empirical research. Section 6 develops the analysis in detail across supply chains, platforms, and executive practice. Section 7 discusses the implications for theory. Section 8 turns to practice. Section 9 addresses limitations and future research. Section 10 concludes. 2. Theoretical Foundations: What Actor-Network Theory Actually Claims 2.1 Origins ANT emerged in the 1980s at the intersection of sociology, anthropology, and the study of science. Its founders were interested in a question that seems innocent and turns out not to be: how does a scientific fact become a fact? Their answer was that facts are built, not found, and that building them requires assembling a network of laboratories, instruments, funders, texts, colleagues, animals, samples, and machines. Remove any element and the fact wobbles. A fact, in this account, is not a statement that happens to correspond to reality. A fact is a statement that has enough allies to survive challenge. That answer proved portable. If a scientific fact is a network, so is a market, a technology, a policy, and, as this paper argues, a corporate strategy. 2.2 Three principles Scholars who have brought ANT into management research, most systematically in the field of supply chain management, typically summarise the theory through three linked principles. #relationality. Nothing has properties on its own. Entities acquire their characteristics through their relations with other entities. A supplier is not intrinsically strategic or peripheral. It becomes strategic through the relations that make it so: a sole-source contract, a qualification certificate, a tooling investment, a component that fits nothing else. Change the relations and you change the thing. This is a strong claim and it cuts against the grain of most management theory, which likes to speak of firms possessing capabilities and resources as though these were personal belongings. heterogeneity. Networks are made of unlike things. There is no separate social world of people and technical world of objects that occasionally interfere with each other. There is one world, in which a procurement manager, a purchase order, a quality standard, a shipping container, and a customs officer are all participants in the same chain of action. ANT's technical term for a participant is the #actants, chosen deliberately to avoid the human overtones of "actor." symmetry. Human and non-human participants should be described in the same terms, at least at the start of the analysis. This is the most misunderstood principle. It does not claim that a spreadsheet has intentions, consciousness, or moral responsibility. It claims something more modest and more useful: that if we want to explain why a course of action happened, we are not allowed to give explanatory priority to human intentions before we have looked. A spreadsheet with a hard-coded formula can defeat a chief executive's stated intention. That is an empirical fact about organisations. symmetry simply asks us not to define it away. 2.3 The vocabulary of translation ANT's central process concept is translation, understood as the work by which one actor speaks for others and enrols them into a shared programme of action. Callon's classic account divides translation into four moments, and this paper will use them repeatedly. problematisation. An actor defines a problem and, in doing so, defines the other actors, their interests, and their identities. Crucially, the actor also positions itself as indispensable to the solution. "The real issue facing this business is fragmented supplier data" is a problematisation. It creates a problem, casts the suppliers as sources of data, casts the analytics team as the solution, and quietly demotes every other possible framing. interessement. The actor works to lock the others into the roles assigned to them, and to cut off competing definitions. Devices are used to do this: pilot projects, memoranda, targets, incentive schemes, and above all technologies that make alternative behaviours costly or impossible. enrolment. The roles are accepted, negotiated, or resisted. Enrolment is never a simple imposition. It is a bargain, and the terms are usually revised in the doing. mobilisation. A few actors come to speak for the many. A supplier scorecard speaks for two thousand suppliers. A country risk index speaks for a nation. A dashboard tile speaks for a continent. If the representation holds, the network can be moved as a bloc. When translation succeeds completely, the network becomes what ANT calls a #black_box: it is used without being questioned, and its internal complexity disappears from view. The related idea of punctualisation describes the moment when an entire network is treated as a single point, a single actor, a single line item. "Our Asian supply base" is a punctualised network. So is "the cloud." So, for that matter, is "the strategy." 2.4 Inscription, delegation, and control at a distance Two further concepts are essential for the argument that follows. inscription refers to the way in which assumptions, interests, and prescriptions get built into objects and documents. A form that has no field for a subcontractor's subcontractor inscribes the assumption that tier-three suppliers do not matter. A carbon accounting template that measures only scope one and scope two emissions inscribes a boundary around responsibility. Inscriptions travel. They can be sent from headquarters to a subsidiary and enforce a way of seeing without anyone from headquarters being present. #control_at_a_distance names the effect. The classic ANT insight, developed in the accounting literature and now widely used in supply chain research, is that headquarters governs distant operations not by watching them but by receiving stable, comparable, mobile representations of them. #calculative_devices, ratios, indices, and reports are what make a global firm governable at all. They are not neutral mirrors. They are actors, and they have effects on what they claim merely to describe. This last point deserves emphasis, because it is where ANT earns its keep in the study of #executive_strategy. The devices that allow an executive to see the world also decide what the world will look like. Change the device and you change the strategy, without changing a single fact on the ground. 2.5 Where ANT sits among the theories It is fair to ask what ANT adds that other perspectives do not. transaction cost economics explains why a firm internalises some activities and outsources others. It is powerful, but it treats the technology of transacting as a parameter rather than as a participant. ANT asks how the transaction cost became calculable in the first place, and who built the instrument that made it so. The resource-based view and dynamic capabilities explain competitive difference through possession and reconfiguration of resources. ANT is uncomfortable with possession. It would ask how a "capability" is held together, what documents and machines carry it, and what would happen if a key engineer left or a key vendor withdrew support. institutional theory explains conformity and legitimacy. It is strong on norms and weak on wires. ANT would want to know how an institutional pressure becomes concrete enough to bite: which audit, which certificate, which software update. #stakeholder_theory identifies who has a claim on the firm. ANT would note, uncomfortably, that some of the most consequential participants in a supply network have no voice, no claim, and no seat, and yet exert real force. A river does. A mineral deposit does. A legacy IT system certainly does. None of this makes the other theories wrong. It makes them partial in a specific and correctable way. ANT is best understood as a complement, a way of reopening black boxes that other theories have prematurely sealed. Recent scholarship in supply chain management has argued exactly this: that adopting ANT allows a radically different conceptualisation of the supply chain and of its management, and yields insight into the fragility of chains, the real conduct of managerial work, the role of non-human participants, and the emergence of power. 3. Literature Review 3.1 Stream one: the multinational enterprise and its recent troubles International business theory was built to explain a world of expanding integration. Its foundational questions concerned why firms cross borders, how they choose entry modes, and how they manage the tension between global efficiency and local responsiveness. That world has changed. A substantial recent literature documents the fracturing of the global economy and its consequences for corporate behaviour. Scholars have examined how firms respond when the economic order they were built for begins to come apart, how divestment cascades through politically exposed markets, and how bifurcated governance regimes create genuinely separate technological and regulatory blocs. Work on techno-geopolitical uncertainty has shown that legislation aimed at securing strategic technologies reshapes the calculus of firms far removed from the industries directly targeted. Studies of decoupling have shown that it is uneven, partial, and sector-specific rather than total, and that firms often reallocate within existing networks rather than exiting them. Two observations are worth drawing from this literature. First, most of it is written from the outside in. It describes environmental shifts and firm-level responses. The mechanism connecting the two, the actual process by which a geopolitical event becomes a strategic decision inside a specific company, is usually left as a black box. That black box is exactly what ANT proposes to open. Second, the empirical material is saturated with objects. Export controls operate on specific machines. Sanctions operate on specific payment rails. Screening regimes operate on specific corporate structures. Critical minerals policy operates on specific chemistry. It is difficult to read this literature carefully without concluding that the material world has re-entered international business whether or not the theory has made room for it. 3.2 Stream two: global supply chains and value chain governance The global value chain tradition has always been more materially minded than mainstream international management. It attends to lead firms, suppliers, upgrading, and the distribution of value along the chain. A major multi-disciplinary review consolidated this literature and set an agenda that continues to shape the field. Work published since the pandemic has moved decisively toward questions of #supply_chain_resilience and reconfiguration. Studies have examined how firms redesign global supply chains under compounding disruptions, how trade policy provokes adaptive reconfiguration, and how geographical reconfiguration is bounded by managerial cognition, with firms searching for new locations only within a limited mental space rather than across the full set of possibilities. That last finding is quietly devastating for rational-choice accounts and quietly supportive of the argument advanced here, because it implies that the option set itself is constructed rather than given. Alongside this, a distinct body of work has begun applying ANT directly to supply chains. The most important single contribution argues that supply chain management research has borrowed its assumptions from other fields and, in doing so, has become blind to features that are distinctive to supply chains: their fragility, their dependence on non-human participants, and the way power emerges within them rather than being held in advance. Related work has used ANT with system dynamics modelling to represent agricultural supply chains as systems of human and non-human agents, has examined how artificial intelligence occupies actor roles within resilient supply networks, and has explored how #blockchain interacts with #accounting and #management_control practices in the agro-food sector, showing that the technology does not simply record the chain but reshapes what counts as a fact within it. Meanwhile, empirical studies of Industry 4.0 technologies have examined their effect on the resilience of established cross-border supply chains, and network-based analyses have connected supply chain technologies to sustainability outcomes by taking seriously the mutual dependence of human and non-human participants. The picture that emerges from this stream is of a field discovering, slowly and by several routes, that its objects have been acting all along. 3.3 Stream three: digital platforms and platform-based internationalisation The third stream is the youngest and the fastest moving. Early theoretical work established that digital platforms and ecosystems create new pathways for internationalisation, new ways of building knowledge and relationships across borders, and new modes of value delivery to global customers, and argued that international business theory would have to absorb these implications. Subsequent research has refined the picture considerably. Studies have shown that #network_effects in digital platforms differ from those in physical products, that switching costs are often lower than assumed, and that winner-takes-all outcomes are less automatic than early accounts suggested. Platforms internationalise at very different speeds and intensities, and their strategies for doing so differ systematically from those of traditional firms, in part because internationalisation is built into their core model rather than layered onto it. Work on hyperspecialisation and hyperscaling has argued that digital firms can grow to extraordinary scale precisely by narrowing their scope and relying on external partners for everything else, a structural fact with obvious implications for how such firms perceive and manage risk. Research on platform boundaries has shown that these boundaries are made and remade through the interplay of firm scope, the sides of the platform, and the design of #api and other digital interfaces. Studies of small and medium-sized enterprises have found that digital platforms function as both enablers and constraints of internationalisation, with commercial and social platforms exerting distinct effects, and with the firm's own internal digitalisation moderating those effects. Reviews of platform-based firms have consolidated a rapidly expanding evidence base, and scholars have argued for the integration of platform theory with international business theory through the concept of the platform-based multinational. Critical voices in the same literature have drawn attention to the dark side of digital globalisation, and to the ways that #data_localisation, regulatory divergence, and digital sovereignty claims cut against the frictionless picture that platform enthusiasm tends to produce. 3.4 The gap Each stream has an obvious blind spot from the vantage point of the others. International management theory under-specifies materiality. Supply chain research has begun to correct this but rarely connects it to the strategic apex of the firm. Platform research is intensely aware of infrastructure but tends to study platforms as firms rather than as the environment within which other firms now make strategy. What is missing is a framework that follows a single chain of action across all three: from a physical constraint in a supply network, through a digital representation of that constraint, into an executive decision, and back out again into the world as an instruction that reorganises the network. That chain is what ANT is designed to trace. Reviews of ANT's uptake across management and information systems research have noted that its diffusion has been uneven and that its potential remains underexploited, particularly outside a handful of national research communities. This paper is an attempt to exploit some of it. 4. Conceptual Framework: The Strategy Translation Cycle 4.1 Overview The framework proposed here treats executive strategy not as a decision but as a cycle of translation, in which a heterogeneous network is repeatedly assembled, stabilised, disrupted, and reassembled. It has four phases, corresponding to Callon's four moments, and one background condition. The background condition is punctualisation. In normal times, the executive does not see a network. She sees objects: "our supply base," "the platform," "the region." These are networks that have been folded up and treated as single points. Strategy in normal times operates on these folded objects. Phase one: problematisation. A problem is defined, usually by an alliance of humans and instruments. A dashboard turns amber. A ratings provider downgrades a country. A supplier misses a delivery window. An analyst writes a slide. The problem is not discovered; it is composed. And in composing it, the composers define who the relevant actors are and what their interests must be. Phase two: interessement. The problem definition is defended against rivals and made binding. Budgets are allocated. Task forces are created. Software modules are procured. Crucially, technology does much of this work, because a system that cannot record an alternative makes that alternative expensive to pursue. Phase three: enrolment. Suppliers, subsidiaries, regulators, platforms, and machines take up, refuse, or renegotiate their assigned roles. This is where most strategies actually fail. A supplier agrees to a dual-sourcing arrangement and then quietly subcontracts to the same tier-two firm as before. A legacy system cannot represent the new supplier category. A regulator refuses to accept a certification. Enrolment failure looks like execution failure, but it is not the same thing. Phase four: mobilisation. A few spokespersons come to represent the network. The chief supply chain officer speaks for the network in the boardroom, armed with metrics that speak for thousands of relationships. If the representation holds, the executive can act on the network as though it were an object again. Punctualisation is restored, and the cycle can begin once more. Betrayal. ANT adds one further concept that is indispensable here. Translation can always be betrayed. The represented can refuse the representation. A supplier network that was said to be resilient turns out not to be. A platform that was said to be a partner turns out to be a competitor. When betrayal occurs, punctualisation collapses, the object becomes a network again, and the executive suddenly discovers that "our supply base" was two thousand separate agreements held together by hope and a spreadsheet. 4.2 Proposition one: the option set is inscribed before it is chosen P1. In multinational firms, the range of strategic options that executives perceive as available is substantially determined by the inscription capacities of the firm's calculative devices, such that options which cannot be represented by existing systems are systematically excluded from deliberation regardless of their merit. This proposition reframes a familiar phenomenon. Bounded rationality is usually attributed to cognitive limits in human beings. ANT locates it, at least partly, outside the head. The bound is in the software. If the #enterprise_resource_planning system models suppliers as single legal entities with single locations, then a strategy of geographic diversification within a single supplier is literally unthinkable in the sense that it cannot be entered, costed, or reported. It will not appear on the slide. It will not be rejected; it will simply never arrive. The finding from global value chains research that reconfiguration searches are bounded by managerial cognitive span is consistent with this, but ANT sharpens the mechanism. The span is not merely mental. It is instrumental. Expand the instrument and the span expands with it. 4.3 Proposition two: platforms as obligatory passage points P2. As digital platforms accumulate the functions of discovery, transaction, payment, logistics, and reputation, they become obligatory passage point in cross-border exchange, converting the commercial dependence of participating firms into strategic dependence, and transferring a portion of #strategic_agency from the participating firm to the platform's architecture. An obligatory passage point, in ANT, is the narrow gate through which all traffic must flow if any actor is to achieve its goals. Whoever controls the gate does not need to control anything else. The platform literature has documented the components of this argument separately: the boundary-setting power of interfaces, the asymmetric information positions of platform owners, the difficulty participating firms face in building direct customer relationships, the way that platform rules can be changed unilaterally. ANT allows these to be assembled into a single mechanism. The exporter who reaches the world through a marketplace has not merely chosen a channel. It has accepted a problematisation in which its own identity is defined as "seller," its performance is defined by the platform's metrics, and its customers are defined as the platform's users. Its strategy is now formulated in a language it does not own. The proposition has a corollary that matters for policy. Where states pursue digital sovereignty or data localisation, they are, in ANT terms, attempting to build alternative passage points. Whether they can do so depends less on regulation than on whether they can assemble the network of standards, infrastructure, and users required to make the alternative gate actually passable. 4.4 Proposition three: geopolitical disruption operates through depunctualisation P3. geopolitical risk affects executive strategy primarily by breaking the punctualisation of previously stable networks, forcing executives to engage with heterogeneous detail they had previously delegated to instruments, and this depunctualisation, rather than the geopolitical event itself, is the proximate cause of strategic revision. This proposition makes a specific empirical claim that could be tested. It says that the strategic impact of a shock is not proportional to its economic magnitude but to how thoroughly it dissolves the objects through which executives had been managing. A modest tariff that leaves the mental object "our China operations" intact may change costs without changing strategy. A single export-control listing that makes it impossible to know whether a given component is compliant may be economically smaller and strategically enormous, because it forces the firm to disassemble its own supply object into thousands of individually inspected parts. This is why #chip_supply_chains and #critical_minerals have produced strategic upheaval out of proportion to their share of trade value. They depunctualise. The implication for executives is uncomfortable. The most dangerous exposures are not the ones with the largest numbers attached. They are the ones where nobody has ever had to look inside the box. 4.5 Proposition four: strategic failure is often enrolment failure P4. In globally distributed firms, the failure of announced strategies is more frequently attributable to failed enrolment of specific human and non-human actors than to errors in analysis or intent, and such failures are systematically misdiagnosed as problems of execution, culture, or commitment. This proposition has practical bite. Consider the very common case of a multinational that announces a sustainability programme requiring supplier #traceability to raw material origin. The analysis is sound. The intent is real. Three years later, coverage stands at eleven percent. The conventional diagnosis blames supplier reluctance, weak incentives, or insufficient management attention. The ANT diagnosis asks who and what refused enrolment. Perhaps the tier-two suppliers were never given a role in the problematisation and therefore have no reason to accept one now. Perhaps the traceability platform cannot represent a commodity that is physically blended at a warehouse, so the object being traced does not exist in the form the system requires. Perhaps the #procurement team's own incentives, inscribed in a bonus formula, reward unit cost reduction, and the traceable supplier costs more. Perhaps a #standards body has not yet certified the method, so the data cannot be used for #sustainability_reporting and therefore nobody in finance cares. Each of these is an enrolment failure with a specific address. Each is fixable. None is fixed by exhorting people to care more. 5. Methodology 5.1 Approach of this paper This article is an integrative conceptual review. It does not present new primary data. It synthesises three bodies of literature and develops theory from their intersection. This design was chosen because the gap identified is theoretical rather than empirical: the phenomena are well documented in fragments, and what is missing is a framework that connects them. Sources were identified through structured searching of the international management, operations and supply chain, and information systems literatures, with priority given to peer-reviewed work published within the last five years and to a small number of foundational texts that cannot responsibly be omitted from any account of ANT. Selection favoured work that either applies ANT within management settings or documents the phenomena the framework seeks to explain, namely supply chain reconfiguration, platform-based internationalisation, and geopolitical fragmentation. The reader should treat the propositions as arguments to be tested rather than findings. That is the appropriate status of theory-building work, and it is the honest one. 5.2 Guidance for students conducting ANT-informed research Because this article is written for students, a section on how to actually do this kind of research is warranted. ANT is notoriously easy to invoke and difficult to practise. The following guidance draws on the methodological discussions in the supply chain and information systems literatures. Follow the actors, and mean it. ANT's core methodological instruction is to trace connections empirically rather than to impose a framework. This means you do not begin with a list of stakeholders. You begin with a single controversy, a moment when things stopped working, and you ask what and who was involved. Then you follow each thread outward, and you do not stop when you leave the categories you expected. Choose a controversy, not a topic. ANT works best where a network is visibly under strain, because that is when its components become describable. A smoothly running process is a black box and will tell you nothing. A failed system implementation, a supplier crisis, a contested standard, a regulatory dispute: these are ANT's natural habitats. Take documents and machines as informants. Interview the humans, certainly. But also read the contract, the form, the specification, the error log, the audit report, the dashboard. Ask what these objects prescribe, what they make easy, what they make impossible, and who wrote them. inscription analysis is real analysis. Resist the urge to explain by context. The temptation, always, is to say that something happened because of "the culture" or "the institutional environment" or "market pressure." ANT regards these as things to be explained, not explanations. If institutional pressure was real, show the letter, the inspection, the fine, the meeting. Context must be assembled from traceable connections or it is not context, it is hand-waving. Be honest about ANT's descriptive bias. ANT produces rich, situated accounts. It is weak at prediction and weak at generalisation, and its practitioners have often been cheerfully unbothered by this. If your research question requires generalisable causal claims, ANT alone will not serve you, and you should say so. Methods that fit. #ethnography, longitudinal #case_study, document analysis, and process tracing all sit comfortably with ANT. Some recent work has combined ANT with formal modelling, using system dynamics to represent the interactions among human and non-human agents identified through #qualitative_research, which offers one route toward greater analytical rigour without abandoning ANT's ontology. 6. Analysis 6.1 The supply chain as a heterogeneous network Begin with a container. It is a steel box of standardised dimensions, and those dimensions were fixed decades ago. Every port crane, every truck chassis, every ship's hold, every warehouse door, and a large part of the world's manufacturing geography is arranged around them. The container does not have goals. It nonetheless exerts an enormous force on the strategies of firms, because it makes some things cheap and other things expensive, and it does so invisibly, as a fact of nature rather than a decision. Now add the other participants. A customs classification code, which determines a tariff and can turn a profitable product into a loss-making one through the interpretation of a single word. An incoterm, which allocates risk at a specific physical point. A quality certificate, which lives for a fixed number of months and whose expiry can halt a production line as effectively as a fire. A piece of tooling, physically located in a single factory, which cannot be moved without months of requalification and which therefore holds a firm's sourcing strategy hostage in a way that no contract does. A software system at a port, which if it fails will strand goods that are physically present and legally cleared. Recent supply chain scholarship has argued precisely this: that ANT reveals the dynamic and fragile character of supply chains, the real conduct of managerial work within them, the role of non-human actors in managing the chain, and the way power unfolds rather than pre-exists. The point is not that objects are interesting. The point is that #power in supply networks is frequently held by objects, and that a theory which cannot see this will misattribute that power to whichever human happens to be standing nearby. 6.2 Inscription devices and the manufacture of the strategic view The instruments through which executives see their supply networks deserve particular attention, because they are where ANT most directly touches executive strategy. Consider the #supplier_scorecards. It appears to be a measuring instrument. It is in fact a machine for producing a particular kind of supplier. It defines what counts as performance. It defines the units of comparison. It converts a rich, particular, historically embedded relationship into a number that can be placed alongside other numbers. Once it does so, that number can travel. It can be aggregated, benchmarked, put in a dashboards, and shown to a board. This is mobilisation in its purest form. Two thousand relationships are made to speak with one voice, and that voice says what the scorecard is capable of saying. Everything the scorecard cannot say is now, for strategic purposes, silent. The supplier's political exposure. Its dependence on a single sub-tier vendor. The fact that its plant sits in a flood plain. The fact that its founder is nearing retirement with no succession. These are not on the scorecard, so they are not in the boardroom, so they are not in the strategy, until the day they are the only thing in the strategy. The same logic applies to #carbon_accounting templates, to #ESG ratings, to country risk indices, to #compliance modules, and to the entire apparatus of #due_diligence. Each is an inscription. Each defines a boundary of responsibility and a domain of visibility. Each, in the process of making the world legible, makes most of the world invisible. Research applying ANT to blockchain in agro-food supply chains has shown this concretely, tracing how a new recording technology interacts with existing management control practices and reshapes what the organisation is able to treat as true. This is the deep reason why #supply_chain_visibility initiatives so often disappoint. They are usually framed as data problems. They are actually translation problems. Adding data to a system that inscribes the wrong ontology does not produce visibility. It produces more of the same blindness, at higher resolution. 6.3 Platforms as obligatory passage points Turn now to digital platforms. A platform, described in ANT terms, is an actor that has succeeded in making itself indispensable. It has performed a problematisation so effective that other actors now define their own interests in its terms. The seller wants ranking. The driver wants acceptance rate. The developer wants api access. The advertiser wants impressions. None of these were natural interests. They were manufactured, and they were manufactured by the platform's inscription of what success means. The literature on platform-based internationalisation supports this reading, even where it does not use ANT's language. It documents that platforms provide critical external resources that firms could not otherwise assemble, that they enable extraordinarily rapid cross-border scaling, and that they simultaneously constrain the firms that depend on them. It documents that platform boundaries are actively managed through interface design. It documents that hyperscaling is achieved through hyperspecialisation, which is another way of saying that the platform holds the passage point and lets everyone else do the work. For the executive strategy of a non-platform multinational, three consequences follow. The strategic frontier moves inside someone else's system. A consumer goods firm's most important decisions about pricing, assortment, and promotion may now be executed within a marketplace's rules, and constrained by an algorithm the firm cannot inspect. Its strategy team is optimising within a space it did not design. It is, in the precise ANT sense, enrolled. Data becomes a site of contest rather than an asset. The firm generates transaction data by selling. The platform holds it. The firm's ability to understand its own customers is now a negotiated outcome. This transforms what looked like an operational question into a question of strategic agency. Dependence is discovered late. Because platform relationships begin as channels and become infrastructures, firms typically do not notice the transition. The dependence is punctualisation. It shows up as a line item. It reveals itself as a network only when the platform changes its rules, at which point the firm discovers that a substantial fraction of its revenue was passing through a gate it does not own. There is a further wrinkle worth flagging for students. Platforms are not monoliths. Recent work stresses that network effects in digital platforms are frequently local rather than global, that switching costs are often lower than expected, and that winner-takes-all outcomes are not automatic. In ANT terms, this means passage points are contestable and that the work of holding one requires constant, expensive maintenance. Platforms know this, which is why so much of their behaviour is best understood as interessement: the ceaseless labour of making it difficult for their participants to imagine an elsewhere. 6.4 Artificial intelligence and the delegation of strategic work A new class of participant has entered the network, and it is not obvious that our theory is ready for it. Recent research has begun examining artificial intelligence as an actor within supply networks, asking what social abilities and network roles algorithmic systems occupy and how they contribute to resilience. Other work is beginning to examine what happens when boundary-spanning employees are displaced by AI, and what this does to the innovation capacity of the chain. From an ANT perspective, this is not a new kind of question. It is an old question with higher stakes. Delegation to non-humans has always been how organisations work; a purchase order delegates authority, a specification delegates judgment, a control chart delegates attention. What changes with AI is the scope and the opacity of the delegation. Three specific concerns arise for executive strategy. First, AI systems perform mobilisation at a scale and speed no human spokesperson can match, which means the representations they produce are extraordinarily difficult to challenge. When a model says a supplier is high-risk, the cost of asking why is high and the cost of accepting is zero. enrolment happens by default. Second, AI systems are trained on the inscriptions the firm already possesses, which means they tend to reproduce and harden existing blind spots. If tier-three suppliers were never in the data, the model will not merely fail to see them. It will confidently report that there is nothing there. Third, and most importantly, AI shifts the boundary between the strategy that is deliberated and the strategy that is enacted. Millions of routing, sourcing, pricing, and inventory decisions are now made by systems whose aggregate effect constitutes a strategic posture that no one ever chose. The firm has a strategy in the ordinary sense, written on slides. It also has a revealed strategy, executed by machines. ANT would insist that we study the second at least as carefully as the first. 6.5 Geopolitics: when the network stops being an object The recent literature on geopolitical fragmentation gives us abundant raw material. decoupling is real but partial. #reshoring and #nearshoring are announced more often than accomplished. #friendshoring reallocates investment toward politically aligned regions while leaving peripheral areas behind. industrial policy has shifted from expanding domestic firms abroad to screening and restraining foreign activity at home. Emerging economies face the uncomfortable position of needing to stay integrated while building resilience against fragmentation. The ANT contribution here is to specify the mechanism by which these macro-level shifts become micro-level strategy. They do so by depunctualising. Before an export control, "our semiconductor supply" is an object with a cost, a lead time, and a supplier name. After an export control, it is a network of specific tools, specific nationalities of engineers, specific software licences, specific end-use certifications, and specific chains of ownership that must each be individually traced and justified. The firm is now compelled to know things about itself that it never previously needed to know, and in most cases cannot easily discover, because its systems were never built to record them. This explains a puzzle in the empirical literature. Firms often respond to geopolitical shocks in ways that look excessive relative to the economic magnitude of the shock. From a rational-calculation standpoint, this looks like overreaction. From an ANT standpoint, it is perfectly proportionate: the shock has not merely raised a cost, it has destroyed a category. The firm is not responding to the tariff. It is responding to the sudden discovery that it does not know what it is made of. It also explains why the response so often takes the form of investment in visibility rather than investment in alternatives. Before you can rebuild, you must first be able to see. And seeing, as we have established, requires building new inscription devices, which takes years and which will, in due course, install a new set of blind spots. 6.6 The executive as network builder What, then, does an executive actually do? The framework proposed here suggests an answer that is less flattering and more useful than the conventional one. The executive is not primarily a decider. She is a builder and repairer of networks, and her most consequential work consists of: Composing problems. Choosing which problematisation the organisation will adopt determines almost everything downstream. This is the single greatest exercise of strategic agency available to an executive, and it is usually performed unreflectively, in response to whatever the instruments have made visible. Building interessement devices. Budgets, incentives, structures, and above all systems. Every system procured is a bet on an ontology. Every key performance indicators is a machine for producing a certain kind of behaviour. #board_governance that does not understand this is governing the slides rather than the firm. Doing the unglamorous work of enrolment. Persuading, bargaining, testing, and discovering that the assigned roles do not fit. This is where strategies die, and it is systematically under-resourced relative to the analysis that precedes it. Maintaining mobilisation. Keeping the spokespersons credible, which means periodically checking whether the metrics still represent the thing they claim to represent. This is the discipline that almost no organisation practises, and its absence is why #risk_management functions are so often surprised by risks they were formally monitoring. Watching for betrayal. Building the organisational habit of asking, about any stable object, what it would look like if it turned out to be a network. This is the practical translation of ANT into a management discipline, and it is close to what the best crisis-experienced executives already do by instinct. 7. Discussion 7.1 Implications for international management theory The framework has three implications for theory. Agency should be treated as distributed and as an empirical question. International management has tended to locate agency in firms and in managers, and to treat everything else as context. The evidence reviewed here suggests this is untenable. A more defensible position is that agency is distributed across a network and that its distribution in any given case is something to be discovered rather than assumed. This does not require accepting ANT's more provocative metaphysical claims. It requires only accepting that the question of who is acting should be answered by looking. Strategy content and strategy process should be reunited. Since these traditions separated, content research has studied what firms decide and process research has studied how they decide, with the tacit assumption that the two can be studied independently. ANT denies this. What can be decided is a function of the process, and specifically of the instruments the process employs. Proposition one, if supported, would imply that a great deal of variance in strategic content is explained by variance in calculative devices rather than by variance in environment or intent. The unit of analysis should sometimes be the network, not the firm. international management has a firm-centric habit that is increasingly hard to justify. When a supply network is co-owned, co-governed, and co-dependent, and when a platform's architecture shapes the strategies of thousands of firms simultaneously, the firm is an arbitrary boundary drawn across a continuous field of action. ANT provides one principled way to draw the boundary differently: follow the connections until they stop mattering, and let the empirical material tell you where the edge is. 7.2 Engagement with rival explanations Intellectual honesty requires acknowledging what rival theories would say to all this. A transaction cost economist would object that ANT lacks a mechanism for explaining why firms make the choices they make, and that describing a network is not the same as explaining an outcome. This objection has force. ANT is not a theory of choice, and where the research question is genuinely about choice under specified constraints, other tools are better. An institutional theorist would object that ANT is so committed to following local connections that it loses sight of the broad, durable structures that make certain connections likely in the first place. Also fair. ANT's flat ontology is a methodological discipline, not a description of a world without structure, but it can become the former if practised carelessly. A quantitative scholar would object that ANT resists operationalisation and generalisation. Entirely true. ANT-based work tends to produce compelling single cases and struggles to accumulate. This is a real weakness and it is the main reason the theory has diffused unevenly across research communities. Recent attempts to combine ANT with formal modelling represent one honest response to this criticism. A critical scholar would object, from the opposite direction, that ANT's symmetry between humans and objects risks flattening moral distinctions and letting powerful actors hide behind their machinery. This is the most serious objection of the four. A supply network that exploits workers is not exonerated by the observation that its exploitation is inscribed in software. ANT scholars have generally answered that describing how power is made material is a precondition for challenging it, not an alternative to challenging it. Students should be aware that the debate is live and should not treat symmetry as a licence to avoid normative judgment. The response to all four objections is the same in structure. ANT is a complement, not a replacement. Its value lies precisely in the questions the other theories have agreed not to ask. 7.3 A note on what ANT is not Three misuses are common enough to warrant a warning. ANT is not a synonym for "networks are important." Social network analysis studies ties between people and produces valuable results. It is not ANT and does not become ANT by adding the word. ANT is not a claim that objects have feelings. It is a claim about how we should write explanations, not a claim about the inner lives of forklifts. ANT is not a framework to be applied. This is the deepest and most frequently violated point. Latour's own instruction was to describe, and to be suspicious of any explanation that arrives before the description is complete. A paper that lists four moments of translation and then sorts its interview data into them has, arguably, done the opposite of ANT. 8. Implications for Practice 8.1 For executives Audit your instruments, not just your risks. The most valuable exercise available to a strategy function is to ask, systematically, what its systems cannot represent. Convene the people who maintain the data and ask them what they have to leave out. The answers are usually known, usually mundane, and usually strategically significant. Treat every metric as a spokesperson and ask whether it still speaks truly. A key performance indicators is a representation of a network. Representations drift. The formal question to put to any dashboard is not "what does this say" but "what would have to be true for this to still be an accurate summary of what it claims to summarise, and when did we last check?" Budget for enrolment. The gap between announced strategy and realised strategy is usually a gap in enrolment work. This work is concrete: identifying every actor whose cooperation the strategy requires, including systems and standards bodies, and confirming that each has accepted its role. It is unglamorous, it is cheap relative to the strategy it protects, and it is almost never done. Map your obligatory passage points. Ask which gates your business must pass through and who owns them. Include platforms, certification bodies, single-source components, key software vendors, and specific logistics chokepoints. For each, ask what would happen if the gatekeeper changed the rules tomorrow. Firms routinely perform this exercise for customers and competitors and routinely omit it for infrastructures. Practise depunctualisation deliberately. Once a year, take the three objects on which the strategy most depends, and force the organisation to open them. Not to find something. To keep the habit alive. 8.2 For suppliers and firms in emerging markets The literature on global value chains has established that firms in emerging economies face a particularly acute version of this problem: they cannot afford to decouple from major markets, and yet the reconfiguration of chains around geopolitical blocs threatens their position. The strategic advice that follows from ANT is specific. Do not compete only on cost. Compete on indispensability. In ANT terms, work to become a passage point rather than a link. This means investing in the things that are hard to re-enrol elsewhere: qualifications, certifications, embedded engineering knowledge, tooling, and above all the capacity to be represented well in the lead firm's instruments. A supplier that can produce the data the lead firm's compliance system requires is more valuable than a cheaper supplier that cannot, because it reduces the lead firm's translation costs. Understand that standards and #certification are not administrative overhead. They are the mechanism by which value chains decide who is in and who is out. A supplier that participates in setting a standard is doing strategy at a level that most of its competitors do not know exists. 8.3 For policymakers The framework suggests that policy interventions succeed or fail according to whether they can assemble a network, not according to whether they identify a correct objective. A subsidy that funds a factory without funding the standards, the skills, the equipment supply, and the certification pathways has funded a node without a network. A data localisation requirement that does not also build a credible alternative to the platform it seeks to displace produces friction rather than sovereignty. Research on industrial policy in the current era has begun to make versions of this point empirically, noting that the same policy produces sharply different outcomes depending on whether the surrounding network exists. 9. Limitations and Future Research 9.1 Limitations This paper is conceptual. Its propositions are arguments, not findings, and they have not been tested. The literatures it synthesises are large, and any synthesis of this kind involves selection that another author would make differently. The framework is developed for large multinationals with complex supply networks and may travel poorly to small firms, to purely domestic firms, or to service industries with limited physical infrastructure. The paper also relies, as ANT-informed work generally must, on illustration rather than measurement, and readers who require quantitative demonstration will not find it here. Finally, and it is worth saying plainly, ANT is a demanding theory that is easy to use badly. This paper has tried to use it carefully, but the risk of turning a method of description into a checklist is present in any attempt to make ANT usable for a management audience, including this one. 9.2 Future research Test proposition one directly. Comparative case studies of firms with materially different enterprise resource planning and analytics architectures, facing similar strategic problems, would allow researchers to examine whether the option sets that reach the executive level differ systematically with the instruments. This is a tractable and important study that has not, to our knowledge, been done. Study depunctualisation events. Export control listings, sanctions, and sudden certification failures offer natural experiments in which a stable object becomes a network overnight. Process research tracing what firms then do, and how long it takes them to build a new object, would test proposition three. Follow an inscription across borders. A promising design is to select a single device, a supplier code of conduct, a carbon reporting template, a due diligence questionnaire, and follow it from headquarters through the tiers of a supply network, documenting what it prescribes, how it is resisted, and what it becomes by the time it reaches a tier-three factory. This is classic ANT and it would produce genuinely new knowledge about how ESG commitments do and do not travel. Take AI seriously as an actor. The nascent research treating artificial intelligence as a participant with network roles in supply chains should be extended to the strategic apex. The specific question is whether, and how, algorithmic representation changes the character of executive deliberation. Examine platform betrayal. Cases in which platform-dependent firms discover that their platform has become a competitor are common, consequential, and under-theorised. ANT's concept of betrayal offers a precise vocabulary for studying them. Bridge to modelling. The combination of ANT with system dynamics or agent-based modelling, attempted in recent supply chain work, deserves further development. It offers a route to accumulating knowledge without abandoning ANT's central commitments, and it directly addresses the accumulation problem that has limited the theory's uptake. Examine digital sovereignty as network building. State attempts to construct alternative digital infrastructures are, in ANT terms, attempts to build rival passage points. Studying them as such would connect the international business literature on regulation with the platform literature and with ANT in a way that has not yet been attempted. 10. Conclusion This article began with a chief operating officer looking at a dashboard, and it ends there too. The conventional account of executive strategy invites us to focus on what she does next. This paper has argued that the more revealing question is what happened before she looked: who built the dashboard, what it can and cannot show, which relationships it converted into numbers and which it left silent, and which of her available options were quietly deleted by a system architecture decision made by someone she has never met. actor-network theory offers international management a way of asking that question rigorously. It does so by refusing three comfortable assumptions: that agency belongs to humans, that networks are stable, and that instruments merely report. In their place it offers a picture of strategy as continuous translation work, in which heterogeneous participants are enrolled into fragile alliances that must be constantly maintained and that can always be betrayed. The setting could hardly be more suitable. global supply chains have become visibly, painfully material. digital platforms have become the infrastructures within which other firms make their strategies. geopolitical risk has torn open objects that firms had folded shut decades ago. The world has, in effect, become more ANT-shaped, and our theory should follow it. The practical message for students, and for the executives they will become, is disarmingly simple. Strategy is not chiefly a matter of thinking clearly about a world that is out there. It is a matter of assembling and maintaining a world that will hold. The firms that survive disruption are not the ones with the best analysis. They are the ones that knew what their networks were made of before they had to find out. #actor_network_theory_in_international_management #ANT_and_global_supply_chains #digital_platforms_and_executive_strategy #tracing_translation_in_multinationals #supply_chain_socio_materiality #platform_dependence_and_strategy #global_value_chain_governance #executive_decision_making_research #international_business_theory #strategy_and_technology #STULIB_research #management_students #scopus_style_research #business_research_methods #latourian_management_studies References Autio, E., Mudambi, R., & Yoo, Y. (2021). Digitalization and globalization in a turbulent world: Centrifugal and centripetal forces. Global Strategy Journal, 11(1), 3-16. https://doi.org/10.1002/gsj.1396 Abbasi, I. A., Shamim, A., & Ashari, H. (2026). Theory in action: Operationalizing actor-network theory with system dynamics to transform indigenous chicken supply chains. Journal of Agribusiness in Developing and Emerging Economies, 16(4), 748-767. https://doi.org/10.1108/JADEE-05-2024-0164 Brookbanks, M., & Parry, G. C. (2024). The impact of Industry 4.0 technologies on the resilience of established cross-border supply chains. Supply Chain Management: An International Journal, 29(4), 731-754. https://doi.org/10.1108/SCM-01-2023-0031 Buckley, P. J. (2023). Corporate reactions to the fracturing of the global economy. International Business Review, 32(6), 102154. https://doi.org/10.1016/j.ibusrev.2023.102154 Cennamo, C. (2021). Competing in digital markets: A platform-based perspective. Academy of Management Perspectives, 35(2), 265-291. https://doi.org/10.5465/amp.2016.0048 Chu, Z., Li, X., Bian, C., & Yang, J. (2024). An actor-network theory analysis and modelling of carbon reduction policy coordination in China: A collaborative environmental governance perspective. Journal of Cleaner Production, 442, 140966. https://doi.org/10.1016/j.jclepro.2024.140966 Conde, L., & Munch, C. (2025). Resilient by design: Exploring the social abilities and actor-network roles of artificial intelligence in supply chain management. Journal of Business Logistics, 46(3). https://doi.org/10.1111/jbl.70032 Cornelissen, J., Stephens, V., & Matthews, L. (2024). Unlocking the power of diversity for supply chain knowledge: Is pluralism in theorizing styles the key? Journal of Supply Chain Management, 60(3), 3-17. https://doi.org/10.1111/jscm.12321 Evenett, S. J., & Pisani, N. (2023). Geopolitics, conflict, and decoupling: Evidence of Western divestment from Russia during 2022. Journal of International Business Policy, 6(4), 511-540. https://doi.org/10.1057/s42214-023-00159-y Franco, P., Bajde, D., & Nojgaard, M. O. (2024). (Dis)entangling actor-network theory and assemblage theory in consumer and marketing scholarship: A review and future directions. Journal of Marketing Management, 40(17-18), 1634-1665. https://doi.org/10.1080/0267257X.2024.2426191 Gallivan, M. J. (2024). What has been the uptake of actor-network theory among IS scholars: An analysis of a stalled innovation diffusion. Proceedings of the Americas Conference on Information Systems (AMCIS 2024). Garcia-Canal, E., Guillen, M. F., & Ponte, B. (2024). Catch me if you can: A simulation model of the internationalization of digital platforms. Information and Organization, 34(1), 100501. https://doi.org/10.1016/j.infoandorg.2024.100501 Gawer, A. (2021). Digital platforms' boundaries: The interplay of firm scope, platform sides, and digital interfaces. Long Range Planning, 54(5), 102045. https://doi.org/10.1016/j.lrp.2020.102045 Gereffi, G., Lim, H.-C., & Lee, J. (2021). Trade policies, firm strategies, and adaptive reconfigurations of global value chains. Journal of International Business Policy, 4(4), 506-522. https://doi.org/10.1057/s42214-021-00102-z Giustiziero, G., Kretschmer, T., Somaya, D., & Wu, B. (2023). Hyperspecialization and hyperscaling: A resource-based theory of the digital firm. Strategic Management Journal, 44(6), 1391-1424. https://doi.org/10.1002/smj.3365 Hald, K. S., & Spring, M. (2023). Actor-network theory: A novel approach to supply chain management theory development. Journal of Supply Chain Management, 59(2), 87-105. https://doi.org/10.1111/jscm.12296 Kano, L., Tsang, E. W. K., & Yeung, H. W. (2020). Global value chains: A review of the multi-disciplinary literature. Journal of International Business Studies, 51(4), 577-622. https://doi.org/10.1057/s41267-020-00304-2 Kocabasoglu-Hillmer, C., Roden, S., Vanpoucke, E., Son, B.-G., & Lewis, M. W. (2023). Radical innovations as supply chain disruptions? A paradox between change and stability. Journal of Supply Chain Management, 59(3), 3-19. https://doi.org/10.1111/jscm.12299 Latour, B. (2005). Reassembling the social: An introduction to actor-network-theory. Oxford University Press. Latour, B. (2021). After lockdown: A metamorphosis. Polity Press. Luo, Y., & Van Assche, A. (2023). The rise of techno-geopolitical uncertainty: Implications of the United States CHIPS and Science Act. Journal of International Business Studies, 54(8), 1423-1440. https://doi.org/10.1057/s41267-023-00620-3 Meyer, K. E., Li, J., Brouthers, K. D., & Jean, R.-J. B. (2023). International business in the digital age: Global strategies in a world of national institutions. Journal of International Business Studies, 54(4), 577-598. https://doi.org/10.1057/s41267-023-00618-x Nambisan, S., & Luo, Y. (2021). The digital multinational: Navigating the new normal in global business. MIT Press. Nambisan, S., Zahra, S. A., & Luo, Y. (2019). Global platforms and ecosystems: Implications for international business theories. Journal of International Business Studies, 50(9), 1464-1486. https://doi.org/10.1057/s41267-019-00262-4 Ozkan-Ozen, Y. D., Ozbiltekin-Pala, M., & Kazancoglu, Y. (2026). Bridging supply chain technologies and sustainability outcomes from a network based resource dependency perspective. Sustainable Development. https://doi.org/10.1002/sd.70639 Roscoe, S., Aktas, E., Petersen, K. J., Skipworth, H. D., Handfield, R. B., & Habib, F. (2022). Redesigning global supply chains during compounding geopolitical disruptions: The role of supply chain logics. International Journal of Operations and Production Management, 42(5), 875-900. https://doi.org/10.1108/IJOPM-12-2021-0777 Stallkamp, M., & Schotter, A. P. J. (2021). Platforms without borders? The international strategies of digital platform firms. Global Strategy Journal, 11(1), 58-80. https://doi.org/10.1002/gsj.1336 Verbeke, A., & Hutzschenreuter, T. (2021). The dark side of digital globalization. Academy of Management Perspectives, 35(4), 606-621. https://doi.org/10.5465/amp.2020.0015 Vertinsky, I., Kuang, Y., Zhou, D., & Cui, V. (2023). The political economy and dynamics of bifurcated world governance and the decoupling of value chains: An alternative perspective. Journal of International Business Studies, 54(7), 1351-1377. https://doi.org/10.1057/s41267-023-00618-x Vitale, G. (2023). Understanding supply chain digitalization through actor-network theory: The interplay between blockchain, accounting and management control. Springer Nature. https://doi.org/10.1007/978-3-031-30988-5 Wei, W. (2024). Using actor-network theory to revisit the digitalized tool in social design. The Design Journal, 27(1), 49-67. https://doi.org/10.1080/14606925.2023.2279836 Witt, M. A., Lewin, A. Y., Li, P. P., & Gaur, A. (2023). Decoupling in international business: Evidence, drivers, impact, and implications for IB research. Journal of World Business, 58(1), 101399. https://doi.org/10.1016/j.jwb.2022.101399

Global supply chains have moved through a decade of repeated shocks. A pandemic, a canal blockage, port congestion, export controls on advanced chips, trade wars, sanctions, attacks on shipping lanes, and extreme weather have all tested networks that were designed mainly for cost efficiency. Traditional #risk_management tools, which rely on periodic reviews, static safety stock rules, and human judgement under time pressure, often react too slowly. This article examines whether #Agentic_AI, a class of artificial intelligence systems that can perceive, plan, use tools, act, and learn with limited human instruction, can meaningfully improve #supply_chain_resilience. The study uses an integrative review method, combining a structured reading of peer reviewed literature published between 2020 and 2026 with a conceptual synthesis grounded in the dynamic capabilities view, complex adaptive systems theory, and the viable supply chain model. The article develops a five layer reference architecture for agentic supply chain systems, covering perception, memory, reasoning and planning, tool use and action, and governance. It then maps this architecture onto two families of disruption: geopolitical shocks such as #sanctions, #export_controls, and #tariffs, and logistical shocks such as #port_congestion, carrier failure, and capacity collapse. Seven propositions are developed and eleven capability requirements are identified. The analysis argues that #Agentic_AI shifts resilience from a periodic planning exercise towards continuous #autonomous_adaptation, but only where data foundations, #interoperability, and #AI_governance are strong. The article also warns about new risks that agentic systems create, including #model_monoculture, correlated failure, agent driven amplification of the #ripple_effect, and accountability gaps. A maturity model and an implementation roadmap are offered for practitioners, along with a research agenda for scholars. The contribution is threefold: a clear definition of agentic autonomy in an operations context, a theory linked framework for autonomous disruption response, and a balanced assessment of benefits and systemic hazards. Keywords: agentic artificial intelligence; supply chain resilience; autonomous agents; geopolitical risk; logistics disruption; multi agent systems; digital supply chain twin 1. Introduction For most of the last thirty years, the guiding logic of global supply chain design was simple. Find the lowest landed cost, keep inventory thin, source from a small number of trusted suppliers, and rely on cheap and predictable ocean freight. That logic worked while the world was politically stable and transport capacity was abundant. It stopped working around 2018 and has not recovered since. The reasons are well documented. Trade tensions between major economies introduced tariffs that changed the economics of sourcing overnight. The COVID-19 pandemic closed factories, emptied container yards, and exposed how little most firms knew about their suppliers beyond the first tier. A single grounded vessel in the Suez Canal showed how one #chokepoints failure can propagate through thousands of shipments. Semiconductor shortages halted automotive assembly lines across three continents. Export controls on advanced computing hardware turned a technology market into an instrument of statecraft. Attacks on commercial shipping in the Red Sea forced carriers to reroute around southern Africa, adding weeks of transit time and reshaping freight rates. Each event was different in cause, but each produced the same pattern: a local shock that spread far beyond its origin. Researchers describe this spreading pattern as the #ripple_effect, where a disruption at one node cascades downstream and upstream through the network and degrades performance far from the point of impact. Dolgui and Ivanov (2021) argue that the ripple effect is now the central problem of disruption management, because modern networks are so tightly coupled that a supplier failure in one region can stop production in another within days. The traditional answer to this problem is to add buffers: more inventory, more suppliers, more capacity. Buffers work, but they are expensive, and Aldrighetti, Battini, Ivanov and Zennaro (2021) show that the cost of resilience is often underestimated in network design models, which explains why boards approve resilience projects slowly and cut them quickly. The alternative answer is speed. If a network cannot afford large buffers, it must instead detect problems earlier and reconfigure faster. This is where artificial intelligence enters the discussion. Over the past decade, firms have deployed #machine_learning for #demand_forecasting, anomaly detection, and #supplier_risk_scoring. These systems are useful, but they are fundamentally advisory. They produce a number, a score, or an alert, and a human then decides what to do. The decision loop still runs at human speed. Cohen and colleagues (2022) found a wide gap between resilience theory and resilience practice, and part of that gap is simply the time it takes for an insight to become an action. #Agentic_AI proposes a different arrangement. An agentic system does not stop at prediction. It is given a goal, it forms a plan, it calls external tools and systems to gather information and execute steps, it observes the result, and it revises the plan. Yao and colleagues (2023) demonstrated that interleaving reasoning and acting improves the reliability of language model agents on complex tasks. Schick and colleagues (2023) showed that models can learn to call external tools without hand written rules. Shinn and colleagues (2023) showed that agents can improve through structured self reflection on their own failures. Surveys by Wang and colleagues (2024) and Xi and colleagues (2025) trace how these building blocks combine into autonomous agents capable of long horizon tasks. Acharya, Kuppan and Divya (2025) describe agentic systems as goal directed rather than instruction directed, which is the key conceptual shift. Applied to operations, this means an agent could notice an unusual rise in congestion at a discharge port, check the exposure of open orders, simulate the delay impact on a production schedule, evaluate alternative routings, book capacity on a different lane, notify the affected plants, and log the decision, all within an hour, without waiting for a Monday planning meeting. Whether this is realistic, safe, and desirable is the question this article addresses. The research is guided by four questions. First, what exactly distinguishes agentic autonomy from the predictive and prescriptive analytics already used in supply chain management, and why does the distinction matter for resilience? Second, what architecture and capabilities does an agentic system need in order to act usefully during #geopolitical_disruption and #logistical_disruption? Third, through what mechanisms does autonomous adaptation improve the recognised phases of resilience, namely anticipation, resistance, recovery, and adaptation? Fourth, what new risks does agentic autonomy introduce at the level of the individual firm and at the level of the wider network, and how should these risks be governed? The article proceeds as follows. Section 2 reviews the literature on resilience, on AI in operations, and on agentic systems, and identifies the gap. Section 3 develops the theoretical framework and states seven propositions. Section 4 explains the method. Section 5 presents the findings, including the reference architecture, the disruption response mechanisms, and the capability requirements. Section 6 discusses implications, risks, and governance. Section 7 offers a maturity model and implementation roadmap. Section 8 sets out limitations and a research agenda. Section 9 concludes. 2. Literature Review 2.1 Supply chain resilience: from buffers to viability Supply chain resilience is usually defined as the ability of a network to prepare for unexpected events, respond to disruptions, and recover from them while maintaining continuity of operations at the desired level of connectedness and control. Katsaliaki, Galetsi and Kumar (2022) provide a broad review of disruption and resilience research and note that the field has grown quickly but remains fragmented, with many overlapping constructs. Hagele, Grosse and Ivanov (2023) go further and conduct a tertiary study, that is, a review of reviews, and confirm that resilience research is dominated by conceptual work and case studies rather than by tested interventions. Two capabilities appear repeatedly. The first is #redundancy, which means holding spare resources such as extra inventory, extra suppliers, or extra capacity. The second is #flexibility, which means the ability to switch between resources quickly. Kamalahmadi, Shekarian and Mellat Parast (2022) tested both empirically and found that flexibility generally outperforms redundancy in improving resilience, although the two are complements rather than substitutes. This finding matters for the present article, because flexibility is fundamentally a decision speed problem. A dual sourced supplier base only helps if the firm can actually switch, quickly, when the primary source fails. Ivanov and Dolgui (2020) extended resilience thinking with the concept of viability, arguing that in intertwined supply networks the goal is not only to bounce back from a shock but to survive long periods of instability and to keep serving society. Ivanov (2022) formalised this as the viable supply chain model, which integrates #agility, resilience, and sustainability. Ivanov (2023) later connected #viability to Industry 5.0, arguing that human centricity must sit alongside resilience and sustainability rather than being traded against them. Wieland and Durach (2021) offer a useful corrective by distinguishing an engineering perspective on resilience, which treats the network as a machine to be restored to equilibrium, from a social ecological perspective, which treats the network as a living system that must transform. Agentic systems, as this article argues, are attractive precisely because they promise to support transformation, not only restoration. A third strand of the literature concerns #visibility. Ivanov (2021) argued that #end_to_end_visibility built during the pandemic was one of the few durable gains of the period, because firms that could see their multi tier exposure were able to act while others were still gathering data. Kinra, Ivanov, Das and Dolgui (2020) showed how supplier risk exposure can be quantified through structural analysis of the network. Brintrup and colleagues (2020) demonstrated that supplier disruption can be predicted from historical transactional data in complex asset manufacturing. Kosasih and Brintrup (2022) went a step further and used #graph_neural_networks to infer hidden links in supply networks, which addresses the persistent problem that firms simply do not know who their sub tier suppliers are. Wichmann and colleagues (2020) extracted supply chain maps from news articles using deep neural networks, which is directly relevant to geopolitical monitoring. 2.2 From analytics to autonomy: artificial intelligence in supply chain management The application of AI to operations is not new, but its character has changed. Pournader, Ghaderi, Hassanzadegan and Fahimnia (2021) mapped AI applications across supply chain management and found heavy concentration in forecasting, classification, and optimisation. Sharma, Shishodia, Gunasekaran, Min and Munim (2022) mapped the same territory and identified a shift from descriptive towards #predictive_analytics and #prescriptive_analytics. Modgil, Gupta, Stekelorum and Laguir (2022) studied AI use during COVID-19 and found that firms with mature AI capability recovered faster, mainly because they could re-plan more often. Gupta, Modgil, Meissonier and Dwivedi (2021) reached a similar conclusion regarding information system resilience. Naz, Kumar, Majumdar and Agrawal (2022) reviewed whether AI is genuinely an enabler of resilience after COVID-19 and concluded that the evidence is promising but thin, with most studies reporting perceived rather than measured benefits. Zamani, Smyth, Gupta and Dennehy (2023) reviewed AI and big data analytics for resilience and identified an important limitation: most systems support human decisions but do not close the loop between insight and action. Dubey, Bryde, Dwivedi, Graham and Foropon (2022) showed, in a humanitarian setting, that an analytics culture, not just analytics technology, drives agility and resilience. This is a recurring finding. Technology alone does not create capability. A distinct and highly relevant strand concerns #reinforcement_learning. Oroojlooyjadid, Nazari, Snyder and Takac (2022) applied deep Q networks to the beer game and showed that learned policies can outperform classical heuristics under partial information. Gijsbrechts, Boute, Van Mieghem and Zhang (2022) tested deep reinforcement learning on lost sales, dual sourcing, and multi echelon problems and found competitive performance without problem specific tuning. Boute, Gijsbrechts, van Jaarsveld and Vanvuchelen (2022) offered a roadmap for deep reinforcement learning in inventory control and were careful to note where it does not help. Rolf and colleagues (2023) reviewed reinforcement learning across supply chain management and identified a persistent gap between simulation results and industrial deployment. These studies matter because reinforcement learning gives us a policy, that is, a mapping from state to action, which is the closest classical analogue to what an agent does. Digital twins bridge analytics and autonomy. Ivanov and Dolgui (2021) proposed the #digital_supply_chain_twin as a way to manage disruption risk in real time by mirroring the physical network in a computational model. Ivanov (2023) later developed the intelligent digital twin for #stress_testing, resilience, and viability, arguing that a twin is most valuable when it can be run continuously rather than as an occasional exercise. Spieske and Birkel (2021) reviewed Industry 4.0 technologies and resilience and found that #simulation, visibility, and connectivity technologies produce the clearest gains. Dolgui and Ivanov (2022) argued that 5G and the internet of everything provide the connectivity layer needed for real time twins. Fragapane, Ivanov, Peron, Sgarbossa and Strandhagen (2022) showed how autonomous mobile robots increase flexibility inside production networks, which is a physical form of the same idea. Generative AI is the most recent addition. Jackson, Ivanov, Dolgui and Namdar (2024) developed a capability based framework for generative artificial intelligence in supply chain and operations management, and explicitly raised the possibility of autonomous agents that plan and execute. Dwivedi and colleagues (2023) discussed the wider implications of generative conversational AI across research, practice, and policy. Sheffi (2023) argued that AI will change the nature of supply chain work rather than simply remove it, a point that connects directly to #workforce_transformation. 2.3 Agentic AI: definition, architecture, and capability The word agent has a long history in artificial intelligence. Russell and Norvig (2021) define an agent as anything that perceives its environment through sensors and acts upon it through actuators, and a rational agent as one that acts to maximise expected performance given what it knows. What has changed is not the definition but the capability of the underlying models. Bommasani and colleagues (2021) described the rise of foundation models, which are large models trained on broad data and adaptable to many tasks. When such a model is placed inside a loop that lets it reason, call tools, observe results, and try again, it becomes an agent in a practical rather than merely theoretical sense. Yao and colleagues (2023) introduced ReAct, which interleaves reasoning traces with actions and showed that this reduces hallucination and improves task success. Schick and colleagues (2023) introduced Toolformer, showing that models can teach themselves when and how to call APIs, calculators, and search engines. Shinn and colleagues (2023) introduced Reflexion, in which an agent verbally reflects on failure and stores the lesson for future attempts, producing a simple but effective form of learning without weight updates. Park and colleagues (2023) built generative agents with memory, reflection, and planning modules that produced believable long horizon behaviour. From these components, a consistent architecture emerges. Wang and colleagues (2024) survey large language model based autonomous agents and identify four modules: profile, memory, planning, and action. Xi and colleagues (2025) describe a brain, perception, and action structure. Acharya, Kuppan and Divya (2025) survey agentic systems more broadly and emphasise goal decomposition, tool orchestration, and multi agent collaboration. Across these treatments, five capabilities recur: Perception, meaning the ability to ingest heterogeneous signals from structured systems and unstructured sources. Memory, meaning both short term working context and long term storage of experience, often in vector databases or #knowledge_graphs. Reasoning and planning, meaning the decomposition of a goal into steps and the revision of those steps when the world changes. Tool use and action, meaning the ability to call external systems, including transactional systems of record, and to execute changes in them. Reflection and learning, meaning the ability to evaluate outcomes and update behaviour. #multi_agent_systems extend this further. Instead of one agent doing everything, specialised agents negotiate, delegate, and check one another. This is architecturally attractive in supply chains, which are themselves distributed systems of semi autonomous actors with partially conflicting objectives. The literature is also clear about risk. Chan and colleagues (2023) analyse harms from increasingly agentic algorithmic systems and identify a set of properties, including underspecification, directness of impact, goal directedness, and long term planning, that raise the stakes as autonomy grows. Shavit and colleagues (2023) propose practices for governing agentic systems, including agent identification, human oversight of high impact actions, and interruptibility. Gabriel and colleagues (2024) analyse the ethics of advanced AI assistants and stress alignment with user intent and the avoidance of manipulation. Hendrycks, Mazeika and Woodside (2023) survey catastrophic risks and note that competitive pressure can push organisations to deploy autonomy faster than they can supervise it, which is precisely the dynamic one would expect in a cost driven industry. 2.4 Geopolitics and the reconfiguration of global networks Geopolitical disruption is qualitatively different from a warehouse fire. It is often slow to build, fast to bite, legally complex, and irreversible. Roscoe and colleagues (2022) studied redesign under compounding geopolitical disruptions and found that firms apply different supply chain logics, and that these logics shape which responses they even consider. Gereffi, Pananond and Pedersen (2022) analysed resilience through the lens of global value chains and argued that state capability, firm capability, and chain structure interact, so that a purely firm level response is often insufficient. Handfield, Graham and Burns (2020) examined tariffs and trade wars and predicted an evolutionary redesign of networks, which is broadly what has occurred through #nearshoring and #friendshoring. Panwar, Pinkse and De Marchi (2022) considered the future of global supply chains after COVID-19 and questioned whether decoupling is economically viable at scale. Free and Hecimovic (2021) asked whether the pandemic marked the end of the road for a particular model of globalisation. The technology dimension is central. Miller (2022) documents how #semiconductors became the decisive strategic industry and how control over specific tools and materials creates leverage that is very hard to replicate. Similar dynamics apply to #critical_minerals, where processing capacity is geographically concentrated even where extraction is not. For a supply chain manager, this means the relevant risk is often not the supplier but the supplier of the supplier, and not the country of assembly but the country of a licence. Choi, Netland, Sanders, Sodhi and Wagner (2023) revisited just in time production in turbulent times and argued that the answer is not to abandon lean but to make it context aware. Sodhi and Tang (2021) called for research on supply chain management under extreme conditions. Craighead, Ketchen and Darby (2020) offered a theoretical toolbox for pandemic era research. Shih (2020) argued that firms must know their networks in depth and diversify deliberately. Sarkis (2021) connected the disruption debate to sustainability, warning that resilience measures can increase emissions if they simply add stock and airfreight. 2.5 The research gap Three observations follow from this review. First, resilience research knows what capabilities matter, particularly flexibility, visibility, and rapid re-planning, but it has said relatively little about who or what performs the re-planning when disruptions arrive faster than committees can meet. Second, AI in operations research has produced strong predictive tools and promising reinforcement learning policies, but Zamani and colleagues (2023) and Naz and colleagues (2022) both note the missing link between insight and action. #decision_latency remains a bottleneck. Third, the agentic AI literature is technically rich but domain thin. It demonstrates capability on benchmarks and simulated environments, not in regulated, capital intensive, multi party industrial settings with legal exposure. The gap, therefore, is an integrated account of how agentic autonomy could operate inside real supply networks under geopolitical and logistical stress, what it would require, what it would improve, and what it would endanger. This article addresses that gap. 3. Theoretical Framework and Propositions 3.1 Three theoretical lenses The framework rests on three complementary theories. The dynamic capabilities view holds that competitive advantage in changing environments comes from the capacity to sense opportunities and threats, seize them through timely decisions, and transform the resource base. Applied here, Agentic AI can be read as an instrument that industrialises sensing and compresses seizing. Sensing becomes continuous rather than periodic. Seizing becomes machine speed rather than meeting speed. Transformation, however, remains organisational and human, because it involves contracts, capital, and politics. This suggests that agentic systems will strengthen the first two capabilities far more than the third, at least in the near term. The complex adaptive systems lens treats a supply network as a population of interacting agents whose local behaviour produces global patterns that no single actor designed. The ripple effect is exactly such a pattern. This lens is valuable because it warns us that adding autonomous agents does not simply add speed to an existing system. It changes the system. If many firms deploy similar agents that respond to the same signals with similar logic, the aggregate behaviour may become more synchronised and therefore more volatile, not less. The viable supply chain model of Ivanov (2022) integrates agility, resilience, and sustainability and asks whether a network can survive long term instability while continuing to serve society. This lens keeps the analysis honest. An agentic system that protects one firm's service level by outbidding everyone else for scarce capacity has improved that firm's resilience and possibly reduced the viability of the network. 3.2 Propositions Proposition 1. Agentic systems improve the anticipation phase of resilience primarily by widening #perception_layer coverage, that is, by monitoring a far larger number of weak signals across structured and unstructured sources than human teams can monitor, thereby increasing the probability of #early_warning before a disruption becomes visible in transactional data. Proposition 2. Agentic systems improve the response phase primarily by reducing decision latency, that is, the elapsed time between signal detection and executed countermeasure, and this reduction produces disproportionate benefit in networks with low redundancy, because in such networks the value of a countermeasure decays rapidly with time. Proposition 3. The resilience benefit of agentic autonomy is conditional on the quality and coverage of the underlying data. Where multi tier mapping is absent and master data is inconsistent, agentic autonomy converts slow bad decisions into fast bad decisions, and #data_quality therefore moderates the relationship between agentic capability and resilience outcomes. Proposition 4. Agentic systems are more effective against logistical disruption than against geopolitical disruption in the short term, because logistical disruptions have richer real time telemetry, more standardised remedies, and lower legal ambiguity, whereas geopolitical responses involve licences, contracts, and sovereign decisions that lie outside the agent's action space. Proposition 5. multi agent systems that span organisational boundaries deliver greater network level resilience than single firm agents, but only where #interoperability standards and data sharing agreements exist, since without them agents optimise locally and can worsen the ripple effect. Proposition 6. Widespread adoption of similar agentic systems trained on similar data increases model monoculture and therefore raises #systemic_risk, because correlated agent behaviour synchronises demand for scarce alternative capacity and can transform a local shock into a market wide one. Proposition 7. Effective agentic deployment requires graduated autonomy governed by decision impact, with #human_in_the_loop control retained for irreversible, high value, or legally sensitive actions, and full autonomy reserved for reversible, bounded, and frequent actions. 4. Methodology This study uses an integrative conceptual review. This design is appropriate because the subject is emerging, the empirical base is thin, and the aim is to build a framework rather than test a hypothesis. Search strategy. Peer reviewed literature was identified from major operations, information systems, and computer science outlets. Search terms combined three clusters: a resilience cluster (supply chain resilience, disruption, ripple effect, viability, risk management), a technology cluster (artificial intelligence, agentic AI, autonomous agents, multi agent systems, reinforcement learning, large language models, digital twin), and a context cluster (geopolitical, sanctions, tariffs, export controls, logistics, ports, shipping). Inclusion criteria. Sources were included if they were peer reviewed articles, established scholarly books, or widely cited technical reports; if they were published between 2020 and 2026; and if they contributed either a theoretical construct, an empirical finding, or an architectural pattern relevant to the research questions. Practitioner marketing material was excluded. Sources older than the window were excluded except where they define a foundational construct that has no recent equivalent. Synthesis procedure. The synthesis followed four steps. First, resilience constructs were extracted and grouped into the four phases of anticipation, resistance, recovery, and adaptation. Second, agentic capabilities were extracted from the AI literature and grouped into the five layers described in Section 5.1. Third, the two sets were cross mapped in a capability to phase matrix, and gaps were recorded where a resilience phase had no corresponding agentic capability. Fourth, disruption archetypes were constructed for geopolitical and logistical shocks, and the mapped capabilities were traced through each archetype to identify preconditions and failure modes. Illustrative scenarios. Because deployed agentic systems in industry are rarely documented in the academic record, the article uses illustrative scenarios rather than case studies. These scenarios are constructed from documented disruption patterns reported in the reviewed literature and are used analytically, to test whether the proposed architecture could plausibly act, not as evidence that it has acted. Limitations of the method. The approach cannot establish causality and cannot estimate effect sizes. It is vulnerable to the optimism of a young technical literature, in which reported successes are more visible than failures. These limitations are addressed in Section 8. 5. Findings 5.1 A five layer reference architecture for agentic supply chain systems Synthesising the agent architectures of Wang and colleagues (2024), Xi and colleagues (2025), and Acharya and colleagues (2025) with the operational requirements identified by Ivanov (2023) and Jackson and colleagues (2024), the following five layer architecture emerges. Layer 1: Perception. The perception layer converts the world into machine readable state. It draws on three source families. The first is internal transactional data from enterprise resource planning, warehouse management, and transport management systems: open orders, inventory positions, lead times, bills of material, and shipment status. The second is partner and platform data: carrier schedules, vessel positions, port dwell times, customs status, and supplier confirmations. The third is external and largely unstructured data: news, regulatory notices, sanctions listings, weather and hazard feeds, and social signals. Wichmann and colleagues (2020) showed that supply network structure can be inferred from news text, which is precisely the capability needed here. Kosasih and Brintrup (2022) showed that missing links can be predicted with graph neural networks, which allows the perception layer to reason about tiers it cannot directly observe. The critical design point is that perception must be continuous and event driven, not batch. A weekly extract is not perception. It is a photograph. Layer 2: Memory and world model. Memory holds three things. Semantic memory holds the structure of the world: the supply network graph, product hierarchies, contractual terms, compliance rules, and site capabilities. This is best represented as a knowledge graph, because supply chains are relational and queries are inherently multi hop. Episodic memory holds what has happened: past disruptions, past decisions, and their outcomes. Procedural memory holds how things are done: playbooks, approved substitution rules, escalation paths. The world model is where the digital supply chain twin lives. Ivanov and Dolgui (2021) and Ivanov (2023) argue that a twin turns a data model into a decision instrument because it allows counterfactual questions. An agent without a twin can only pattern match on history. An agent with a twin can ask what would happen if. Layer 3: Reasoning and planning. The #planning_module decomposes a goal such as protect the launch schedule for product line X into subgoals, orders them, and revises them as the situation develops. Three planning styles are relevant. Deliberative planning searches over possible action sequences and is suited to slow, high stakes decisions such as network redesign. Reactive planning maps states to actions quickly and is suited to fast, bounded decisions such as re-routing a container. Learned policy planning, drawing on the reinforcement learning work of Boute and colleagues (2022) and Gijsbrechts and colleagues (2022), is suited to repeated decisions such as replenishment under uncertainty. In practice, these are combined. A large language model provides flexible reasoning over unstructured context and orchestrates the workflow, while classical solvers and learned policies provide reliable optimisation within well defined subproblems. This hybrid pattern is important: it uses the model where flexibility is needed and constrains it where correctness is needed. Layer 4: Tool use and action. The #action_layer is what separates an agent from an analyst. Through #tool_use, the agent invokes external capabilities: it queries the twin, runs an optimisation solver, calls a carrier booking interface, checks a denied party list, raises a purchase requisition, or opens a case in a workflow system. Schick and colleagues (2023) showed that tool invocation can be learned. In an industrial setting it should also be constrained, with each tool exposed through a permissioned interface and each action logged. Actions must be classified by reversibility and impact. Requesting a quote is reversible and low impact. Cancelling a purchase order is not. This classification is the foundation of the graduated autonomy model in Proposition 7. Layer 5: Governance, reflection, and learning. This layer contains everything that makes the system safe to run. It includes #explainable_AI mechanisms that render the agent's reasoning inspectable, following the argument of Ali and colleagues (2023) that explainability is a precondition of #trustworthy_AI rather than an optional extra. It includes audit logging of every observation, decision, and action. It includes guardrails: hard constraints the agent may not violate regardless of its objective, such as trade compliance rules. It includes escalation logic that routes decisions to humans when confidence is low or impact is high. And it includes reflection, in the sense of Shinn and colleagues (2023), so that outcomes feed back into episodic memory and improve future behaviour. The five layers can be summarised in a single sentence. Perception tells the agent what is happening, memory tells it what the world is like and what happened before, planning tells it what to do, tools let it do it, and governance decides whether it is allowed to. 5.2 Autonomous adaptation to logistical disruptions Logistical disruptions are the natural first domain for agentic autonomy because they are frequent, observable, and reversible. Archetype 1: Port congestion and berth delay. Congestion at a discharge port raises dwell times and delays inland delivery. The perception layer detects rising dwell times and vessel queuing. The world model maps affected vessels to affected containers to affected orders to affected production lines, which is the multi hop query that knowledge graphs make tractable. The planning module then evaluates options: wait, discharge at an alternative port and move inland, split the shipment, expedite a subset by #air_freight, or reschedule production around the delayed parts. The action layer executes the chosen option through booking and planning systems. A human approves anything above a cost threshold. The value here is not that the machine finds an option no human could find. It is that the machine finds it on day one rather than day six, when alternative capacity is still available and cheaper. This is Proposition 2 in concrete form. Archetype 2: Carrier or lane failure. A carrier withdraws capacity, an airline cancels a freighter rotation, or a rail corridor closes. The agent recomputes routing across the remaining network, respecting service commitments and cost constraints, and rebooks. #transport_routing is a well structured optimisation problem with mature solvers, so the agent's contribution is orchestration and speed rather than mathematical novelty. Archetype 3: Demand and supply shock in a single node. A supplier plant halts. The agent checks alternative qualified sources, checks inventory across the network including in transit stock, reallocates constrained supply across customers according to a policy the firm has approved, and triggers #dual_sourcing activation where a qualified second source exists. Here the constraint is rarely the algorithm. It is qualification. An agent cannot qualify a supplier, and this is a hard limit on autonomy. Archetype 4: Inventory drift under sustained uncertainty. Not all disruption is an event. Some is a slow degradation of lead time reliability. #autonomous_replenishment policies learned through reinforcement learning can adapt safety stock continuously rather than at quarterly reviews. Boute and colleagues (2022) caution that learned policies need careful validation and can fail silently when the environment shifts outside the training distribution, which argues for keeping a classical policy as a fallback and monitoring divergence. Across these archetypes, three enablers appear consistently. The first is end to end visibility, without which the agent cannot construct the exposure map. The second is interoperability, because the agent's usefulness is limited by which systems it can actually call. The third is a pre agreed decision policy, because an agent that must ask a human what to prioritise every time is not saving anyone time. 5.3 Autonomous adaptation to geopolitical disruptions Geopolitical shocks are harder, and the analysis here supports Proposition 4. Archetype 5: Sanctions and denied party changes. A new listing is published. Compliance exposure must be assessed immediately across suppliers, customers, banks, vessels, and beneficial owners. This is a text heavy, relational, time critical task, which is exactly where agentic systems are strong. The agent monitors regulatory publications, parses new listings, matches them against the entity graph including indirect ownership, flags exposed transactions, freezes what must be frozen, and produces a documented assessment for the compliance officer. Crucially, the agent should not decide the legal question. It should assemble the evidence and propose an action, and a qualified human should decide. This is #trade_compliance as machine assisted diligence, not machine judgement. Archetype 6: Export controls on a critical input. Controls are placed on a category of components or on the tools used to make them. The immediate operational task is to determine exposure across the bill of material, including at tiers the firm cannot see. graph neural networks based link prediction, following Kosasih and Brintrup (2022), can propose likely hidden dependencies for human verification. The agent then runs the twin to simulate the production impact under different licence outcomes and different substitution options. It cannot obtain a licence. It cannot redesign a product. But it can tell the organisation, within hours instead of weeks, which products stop and when. Archetype 7: Tariff changes and trade policy shifts. tariffs change landed cost and therefore change sourcing optimality. An agent can continuously recompute landed cost across the qualified supplier set, model the effect of rules of origin, and propose reallocation of volume between existing sources. What it cannot do is decide whether to build a plant in another country. Handfield and colleagues (2020) and Roscoe and colleagues (2022) both emphasise that such redesign is strategic, political, and slow. Archetype 8: Chokepoint closure and maritime security. When a strait or canal becomes unsafe, carriers reroute, transit times lengthen, and effective global capacity falls because the same ships are at sea for longer. The agent's role is anticipatory. It should detect the security signal early through #situational_awareness monitoring, quantify the exposure of goods in transit and goods planned, and reprioritise bookings before rates spike. Speed matters enormously here, because freight markets reprice quickly once everyone sees the same thing. Archetype 9: Concentration risk in critical inputs. critical minerals processing, advanced semiconductors, and certain #pharmaceutical_supply_chains inputs are concentrated in ways that no single firm can diversify away in the short run. Miller (2022) documents how deep and specific such dependencies can be. Here the honest conclusion is that agentic AI offers visibility and preparedness, not solution. It can identify the exposure, quantify it, and monitor it. Removing it requires capital, time, and often state policy, which is the point made by Gereffi and colleagues (2022). The pattern is clear. In geopolitical disruption, agentic systems excel at exposure discovery, monitoring, simulation, and documentation. They perform poorly, and should not be trusted, on legal interpretation, sovereign negotiation, and irreversible strategic commitment. The action space of the agent is bounded by what can be executed through a system interface, and the most consequential geopolitical responses cannot be. 5.4 Multi agent coordination across organisational boundaries The most interesting, and most difficult, possibility is coordination between agents belonging to different firms. In principle, a buyer's agent and a supplier's agent could negotiate allocation, exchange forecast updates, and jointly re-plan, achieving in minutes what currently takes weeks of email. Ivanov and Dolgui (2020) argue that viability is a property of intertwined networks, not individual firms, so network level coordination is where the largest gains should lie. Three obstacles stand in the way. The first is interoperability. Agents must share a common vocabulary for products, locations, events, and commitments. Without semantic standards, cross firm agent communication degrades into brittle integration projects. The second is trust and confidentiality. Firms do not want to reveal capacity, cost, or exposure to counterparties. Kshetri (2021) discusses how distributed ledger approaches can support traceability with controlled disclosure, and this line of work is relevant to the design of agent to agent trust, although blockchain is not a complete answer and adds its own cost. The third is incentive misalignment. An agent optimising its principal's service level will, under scarcity, compete for the same alternative capacity as everyone else's agent. If a hundred firms deploy agents that all detect a chokepoint closure at the same hour and all book the same alternative capacity at the same minute, the market clears violently. This is Proposition 6, and it deserves emphasis. Individually rational agentic behaviour can be collectively destabilising. This is not a hypothetical concern. It is the standard result in any market where reaction speed increases faster than supply elasticity. 5.5 Human and agent teaming Sheffi (2023) argues that AI will reshape supply chain work rather than eliminate it, moving people from execution towards exception handling, judgement, and relationship management. Fugener, Grahl, Gupta and Ketter (2022) study delegation between humans and AI and find that productive delegation requires people to understand where the machine is strong and where it is weak, which is harder than it sounds and does not happen automatically. Three design principles follow. Graduated autonomy. Actions should be tiered. Tier one actions are automatic: information gathering, simulation, alerting, reversible bookings below a threshold. Tier two actions are automatic with notification and a reversal window. Tier three actions require human approval: order cancellation, supplier switching, large financial commitments, anything with legal exposure. Tier four actions are advisory only: the agent may recommend but never execute, for example anything involving sanctions determination or contract termination. This tiering operationalises Proposition 7 and reflects the governance practices proposed by Shavit and colleagues (2023). Explainability by construction. explainable AI cannot be added afterwards. The agent should emit a reasoning trace, the evidence it used, the options it considered, the option it selected, and the expected consequence. Ali and colleagues (2023) argue that explainability is necessary for trust, and Glikson and Woolley (2020) show empirically that human trust in AI depends heavily on transparency and reliability signals. Deliberate friction. Some slowness is protective. A mandatory pause before irreversible action, a second agent that argues the opposite case, or a required human acknowledgement all reduce the chance of a fast, confident, wrong decision. In a system built for speed, friction must be designed in on purpose, because it will not appear by accident. 5.6 Capability requirements: what an organisation actually needs The analysis yields eleven requirements. They are ordered roughly by how often their absence blocks progress. One, a maintained multi tier supply network map, extending at minimum to tier three for critical parts. Two, clean and governed master data, particularly part, supplier, and site identifiers that reconcile across systems. Three, event driven data pipelines with latency measured in minutes, not days. Four, a digital supply chain twin capable of running counterfactual simulations on demand. Five, machine callable interfaces into planning, procurement, and transport systems, with permissioned write access. Six, an approved decision policy that encodes priorities, thresholds, and allocation rules in advance. Seven, a hard constraint library covering compliance, safety, and contractual limits that the agent may never override. Eight, complete audit logging of observations, reasoning, actions, and outcomes. Nine, an escalation and override mechanism that a human can trigger at any time, including a global stop. Ten, monitoring of the agent itself, tracking decision quality, drift, and the frequency of human overrides. Eleven, a workforce that understands the system well enough to supervise it, which is a #change_management task and not a technical one. Most organisations that attempt agentic deployment fail on requirements one through three. The intelligence layer is rarely the binding constraint. The data layer almost always is, which is the essence of Proposition 3. 6. Discussion 6.1 What genuinely changes The central change is the compression of the decision loop. Classical analytics improved the quality of a decision but left its timing to the organisation's calendar. Agentic systems attack the calendar itself. In a disruption, the value of a countermeasure decays with time, sometimes steeply, because alternatives are consumed by whoever moves first. A firm that reroutes on day one pays a premium. A firm that reroutes on day ten may find nothing to buy. Speed is therefore not a convenience. In a scarce market, speed is the resilience. The second change is the scope of attention. Human teams triage. They monitor what they have time to monitor, which means they monitor their largest suppliers and their loudest problems. An agentic system can monitor the long tail: the small supplier of a small component that stops an entire assembly line. This is a real and underappreciated benefit, because disruption history is full of trivial parts causing serious damage. The third change is institutional memory. Episodic memory, in the sense of Shinn and colleagues (2023), means that the lessons of a disruption survive the departure of the person who handled it. Most organisations lose their disruption learning through turnover. A well designed agent does not. 6.2 What does not change Agentic systems do not create capacity. They do not qualify suppliers. They do not obtain export licences, negotiate with governments, or build factories. They do not resolve the fundamental tension identified by Sarkis (2021) between resilience and #sustainability, in which faster and more flexible responses often mean more air freight and more emissions. Indeed, an agent optimising service level without an emissions constraint will systematically choose the higher carbon option, because that option is usually faster. This is an argument for writing sustainability constraints into the hard constraint library rather than hoping they emerge. Agentic systems also do not fix bad strategy. A network with a single point of failure in a politically exposed jurisdiction remains a network with a single point of failure. The agent will simply tell you about it more often and more precisely. 6.3 New risks introduced by autonomy Correlated behaviour and systemic risk. This is the most serious concern and follows directly from the complex adaptive systems lens. If agents across an industry share similar architectures, similar training data, similar signal sources, and similar objectives, their responses will correlate. Correlated responses concentrate demand for scarce alternatives, amplify price movements, and can convert a manageable regional problem into a market wide one. This is model monoculture, and it is a network level externality that no individual firm has an incentive to fix. Regulators and industry bodies may eventually need to address it, in the same way that financial regulators addressed correlated algorithmic trading. Amplification of the ripple effect. Dolgui and Ivanov (2021) describe how disruptions propagate. Autonomous agents propagate decisions as well as disruptions. If a buyer's agent cancels orders automatically, that cancellation becomes a demand shock for the supplier, whose agent may then cancel its own upstream orders. A cascade of automated cancellations can travel upstream faster than any human can intervene. Rate limits, cooling off periods, and cross tier notification requirements are plausible mitigations. Accountability gaps. When an agent makes a decision that causes loss, who is responsible? Chan and colleagues (2023) identify this as a core problem of agentic systems. The practical answer is that responsibility must remain with the deploying organisation, and this requires that every agent action be attributable, logged, and reviewable. #algorithmic_accountability is not achievable without infrastructure built for it in advance. Security and adversarial manipulation. An agent that reads external text and acts on it can be attacked through that text. A false report of a port closure, injected into a monitored feed, could trigger expensive and unnecessary rerouting. Prompt injection through supplier documents, invoices, or emails is a live threat surface. #cyber_security for agentic systems must include input provenance verification, cross source corroboration before consequential action, and strict separation between data the agent reads and instructions the agent follows. Silent degradation. Learned policies fail quietly when the world moves outside their training distribution. Boute and colleagues (2022) warn about this in inventory control. The organisational danger is that the agent keeps producing confident output that is quietly wrong, and because it has been reliable for months, nobody checks. Over trust and skill erosion. If humans stop practising disruption response because the machine does it, the human fallback degrades. When the machine fails, and eventually it will, the organisation discovers that the people who once knew how to handle a crisis have not done so in three years. 6.4 Governance Governance should be proportionate to impact, and it should exist before deployment rather than after the first incident. Four elements are essential. Autonomy tiering by reversibility and value. Described in Section 5.5. This is the single most important control. Constraint hardening. Compliance, safety, and contractual limits must be enforced outside the model, in the tool layer, not requested politely inside a prompt. An instruction can be argued with. A permission cannot. Continuous evaluation. Agent decisions should be sampled and reviewed, override rates tracked, and outcomes compared against a counterfactual baseline. If the override rate rises, something has drifted. Alignment with external frameworks. Public governance instruments, including the risk based approach of the European Union AI Act (2024) and the National Institute of Standards and Technology AI Risk Management Framework (2023), give organisations a defensible structure for classifying, documenting, and monitoring AI systems. Firms operating internationally will need to satisfy multiple regimes, and building for the strictest is usually cheaper than retrofitting. 6.5 Implications for theory For the dynamic capabilities view, the findings suggest that sensing and seizing can be substantially automated while transforming cannot, which implies that the competitive differentiator will migrate towards transformation capability once sensing and seizing become commodities available to everyone through similar tools. For complex adaptive systems theory, agentic deployment offers a rare natural experiment in what happens when the agents in a socio technical system become faster and more similar at the same time. The prediction from theory is increased synchronisation and reduced diversity of response, which means lower system level resilience even as firm level resilience improves. This is a testable and important claim. For the viable supply chain model, the findings suggest that viability requires network level rules of engagement for autonomous agents, not only firm level capability. Ivanov (2023) links viability to human centricity in Industry 5.0, and the graduated autonomy principle developed here is one concrete expression of #Industry_5 thinking in an operations context. 7. Implementation: A Maturity Model and Roadmap 7.1 Five levels of agentic maturity Level 0, Manual. Disruption response is entirely human. Data is gathered on demand. Response time is measured in days or weeks. Level 1, Assisted. Dashboards and alerts exist. Analytics predict risk. Humans decide and execute. Response time is measured in days. Level 2, Recommending. The system proposes specific actions with quantified consequences drawn from a twin. Humans approve and the system executes. Response time is measured in hours. Most advanced organisations that describe themselves as having a #control_tower are here. Level 3, Supervised autonomy. The system executes tier one and tier two actions independently within a defined envelope and escalates the rest. Humans supervise by exception. Response time is measured in minutes for bounded decisions. Level 4, Networked autonomy. Agents coordinate across organisational boundaries under shared protocols and interoperability standards. Response is coordinated rather than merely fast. Very few, if any, organisations are here today, and reaching this level requires collective action rather than individual investment. The important insight from the maturity model is that levels cannot be skipped. An organisation at Level 1 that attempts Level 3 will discover that its data foundation cannot support autonomous action, which returns us to Proposition 3. 7.2 A phased roadmap Phase one, foundations, roughly six to twelve months. Build the multi tier map for critical parts. Fix master data. Establish event driven pipelines. Stand up the twin. This phase produces no autonomy and is therefore politically difficult to fund, which is why so many programmes stall here. It is nonetheless the phase that determines whether everything after it works. Phase two, perception and simulation, roughly three to six months. Deploy agents in read only mode. Let them monitor, detect, simulate, and recommend. Measure their early warning performance against what the human team would have caught. This phase builds trust and produces evidence, and it carries almost no operational risk because the agent cannot act. Phase three, bounded action, roughly six to twelve months. Grant write access for a narrow, reversible, high frequency decision class, for example intra network stock transfers below a value threshold or expedite requests within an agreed budget. Instrument everything. Track override rates. Expand the envelope only when performance justifies it. Run #pilot_projects in a single region or product family rather than enterprise wide. Phase four, supervised autonomy, ongoing. Expand the action envelope by decision class, not by department. Formalise the escalation policy. Establish the review board. Begin measuring #return_on_investment against a counterfactual, which is difficult but necessary, since the natural comparison, namely what would have happened otherwise, is unobservable and must be estimated through simulation. Phase five, network extension, ongoing and collective. Engage suppliers, carriers, and industry bodies on shared protocols. This cannot be done alone. 7.3 Measurement Traditional metrics such as service level and inventory turns are necessary but insufficient, because they do not isolate the agent's contribution. Four additional measures are proposed. Time to detect, meaning the elapsed time between the real world onset of a disruption and its recognition by the system. Time to decide, meaning the elapsed time between recognition and a committed countermeasure. Exposure coverage, meaning the proportion of the bill of material for which multi tier exposure is actually known rather than assumed. Counterfactual loss avoided, meaning the estimated difference between realised outcome and simulated outcome under the previous decision process. This last measure is the hardest and the most valuable, and it requires the twin to be maintained precisely so that the counterfactual can be run. 8. Limitations and Future Research 8.1 Limitations This article is conceptual. It builds a framework from literature and reasons through scenarios. It does not test the framework against deployed systems, because such systems are, at present, either absent or undisclosed. The propositions are therefore plausible rather than established. The literature base is also asymmetric. The resilience literature is mature and empirical. The agentic AI literature is young and heavily weighted towards benchmarks, demonstrations, and surveys, with limited evidence from regulated industrial environments. There is a real risk that the capabilities of agentic systems are overstated in the sources available, since negative results are less likely to be published or publicised. Readers should treat the technical claims as capability ceilings rather than typical performance. Finally, the geopolitical analysis is necessarily general. Sanctions regimes, export controls, and tariff structures differ by jurisdiction and change frequently. The article deliberately describes mechanisms rather than specific measures, because specific measures would be out of date almost immediately. 8.2 A research agenda Empirical validation of decision latency effects. The core claim of this article, that reduced decision latency produces disproportionate resilience gains in low redundancy networks, is testable. Simulation studies using established ripple effect models could quantify the relationship between response time and total disruption cost across different network topologies and redundancy levels. Systemic effects of correlated agents. Agent based simulation should be used to test Proposition 6 directly. If many firms deploy similar agents, does aggregate volatility rise? Under what conditions? What diversity of policy is sufficient to prevent synchronisation? This is possibly the most important open question in the field, and it cannot be answered by studying single firms. Human agent teaming under stress. Fugener and colleagues (2022) studied delegation in controlled settings. Supply chain crises are not controlled settings. Field or experimental work is needed on how planners actually behave when an agent recommends an unpopular action under time pressure, and on when they override correctly versus incorrectly. Evaluation standards for operational agents. The field currently has no shared benchmark for agent performance on supply chain tasks. A public, disruption rich benchmark environment, analogous to those that accelerated progress in other AI domains, would allow claims to be compared rather than merely asserted. Explainability that planners actually use. Ali and colleagues (2023) survey explainability techniques, but few are designed for the specific question a planner asks, which is not how does the model work but why this option and not the one I would have chosen. Domain specific contrastive explanation is an open problem. Legal and accountability structures. Who is liable when an agent cancels an order that turns out to have been essential? Work at the intersection of operations, law, and AI governance is needed, and it should proceed faster than deployment, not behind it. Sustainability constrained autonomy. Given Sarkis (2021), research is needed on how emissions constraints change agentic behaviour under disruption, and on whether resilience and decarbonisation can be jointly optimised or must be traded. Small and medium enterprise access. Almost all of the capability requirements in Section 5.6 assume resources that large firms have and smaller ones do not. If agentic resilience is available only to the largest firms, network level fragility may increase, since the weakest nodes will become relatively weaker. This distributional question deserves serious attention from #policy_makers. 9. Conclusion Global supply chains now operate in an environment where disruption is normal rather than exceptional. Geopolitical fragmentation, technology controls, contested shipping lanes, and climate driven logistics failures are not passing turbulence. They are the operating conditions. Networks built for cost efficiency in a stable world must now perform in an unstable one, and they must do so without the option of unlimited buffers. This article has argued that Agentic AI represents a real change in what is possible, but a narrower one than the enthusiasm surrounding it suggests. The change is in the decision loop. Agentic systems can perceive more widely, remember more reliably, simulate more often, and act more quickly than human teams working through weekly planning cycles. Where a disruption is observable, where the remedy can be executed through a system interface, and where the action is reversible, autonomy delivers genuine value, and it does so most where redundancy is thinnest. That is the core of the argument. But the limits are equally real. Agents cannot qualify a supplier, obtain a licence, negotiate with a state, or build a factory. In the geopolitical domain, where the most consequential disruptions now originate, their contribution is exposure discovery, monitoring, simulation, and documentation, not resolution. Their performance depends entirely on a data foundation that most organisations do not yet have, which means the binding constraint on agentic resilience is not intelligence but plumbing. And autonomy creates its own hazards. Correlated agents can synchronise markets. Automated cancellations can cascade upstream faster than humans can intervene. Learned policies can fail silently. External text can be weaponised against systems that read it. Accountability can dissolve if it is not designed in from the start. These are not reasons to avoid agentic systems. They are reasons to govern them, through graduated autonomy, hard constraints enforced outside the model, complete auditability, continuous evaluation, and deliberate friction before irreversible action. The most useful way to think about Agentic AI in this context is therefore not as a replacement for human judgement but as an extension of organisational attention and reaction speed, bounded by rules that humans set and can revoke. The firms that benefit will be those that fix their data first, start in read only mode, expand autonomy by decision class rather than by ambition, and keep their people close enough to the system to know when it is wrong. The networks that benefit will be those that develop shared standards and shared rules of engagement, because #supply_chain_resilience, in the end, is a property of the network and not of any single firm within it. Hashtags #AgenticAI #Agentic_AI_in_Supply_Chains #Global_Supply_Chain_Resilience #Autonomous_Supply_Chains #Geopolitical_Risk #Logistics_Disruption #Self_Healing_Supply_Chains #AI_Agents #Supply_Chain_Research #Resilient_Operations #Supply_Chain_Digital_Twins #Autonomous_Decision_Making #Supply_Chain_Technology #Operations_Management #Future_of_Logistics References Acharya, D. B., Kuppan, K., and Divya, B. (2025). Agentic AI: Autonomous intelligence for complex goals, a comprehensive survey. IEEE Access, 13, 18912 to 18936. Aldrighetti, R., Battini, D., Ivanov, D., and Zennaro, I. (2021). Costs of resilience and disruptions in supply chain network design models: A review and future research directions. International Journal of Production Economics, 235, 108103. Ali, S., Abuhmed, T., El-Sappagh, S., Muhammad, K., Alonso-Moral, J. M., Confalonieri, R., Guidotti, R., Del Ser, J., Diaz-Rodriguez, N., and Herrera, F. (2023). Explainable Artificial Intelligence (XAI): What we know and what is left to attain trustworthy artificial intelligence. Information Fusion, 99, 101805. Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S., and others (2021). On the opportunities and risks of foundation models. Center for Research on Foundation Models, Stanford University, Technical Report. Boute, R. N., Gijsbrechts, J., van Jaarsveld, W., and Vanvuchelen, N. (2022). Deep reinforcement learning for inventory control: A roadmap. European Journal of Operational Research, 298(2), 401 to 412. Brintrup, A., Pak, J., Ratiney, D., Pearce, T., Wichmann, P., Woodall, P., and McFarlane, D. (2020). Supply chain data analytics for predicting supplier disruptions: A case study in complex asset manufacturing. International Journal of Production Research, 58(11), 3330 to 3341. Chan, A., Salganik, R., Markelius, A., Pang, C., Rajkumar, N., Krasheninnikov, D., and others (2023). Harms from increasingly agentic algorithmic systems. Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency (FAccT), 651 to 666. Choi, T. Y., Netland, T. H., Sanders, N., Sodhi, M. S., and Wagner, S. M. (2023). Just in time for supply chains in turbulent times. Production and Operations Management, 32(7), 2331 to 2340. Cohen, M. A., Cui, S., Doetsch, S., Ernst, R., Huchzermeier, A., Kouvelis, P., Lee, H. L., Matsuo, H., and Tsay, A. A. (2022). Bespoke supply chain resilience: The gap between theory and practice. Journal of Operations Management, 68(5), 515 to 531. Craighead, C. W., Ketchen, D. J., and Darby, J. L. (2020). Pandemics and supply chain management research: Toward a theoretical toolbox. Decision Sciences, 51(4), 838 to 866. Dolgui, A., and Ivanov, D. (2021). Ripple effect and supply chain disruption management: New trends and research directions. International Journal of Production Research, 59(1), 102 to 109. Dolgui, A., and Ivanov, D. (2022). 5G in digital supply chain and operations management: Fostering flexibility, end to end connectivity and real time visibility through internet of everything. International Journal of Production Research, 60(2), 442 to 451. Dubey, R., Bryde, D. J., Dwivedi, Y. K., Graham, G., and Foropon, C. (2022). Impact of artificial intelligence driven big data analytics culture on agility and resilience in humanitarian supply chain: A practice based view. International Journal of Production Economics, 250, 108618. Dwivedi, Y. K., Kshetri, N., Hughes, L., Slade, E. L., Jeyaraj, A., Kar, A. K., and others (2023). Opinion paper: So what if ChatGPT wrote it? Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, practice and policy. International Journal of Information Management, 71, 102642. European Union (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union. Fragapane, G., Ivanov, D., Peron, M., Sgarbossa, F., and Strandhagen, J. O. (2022). Increasing flexibility and productivity in Industry 4.0 production networks with autonomous mobile robots and smart intralogistics. Annals of Operations Research, 308, 125 to 143. Free, C., and Hecimovic, A. (2021). Global supply chains after COVID-19: The end of the road for neoliberal globalisation? Accounting, Auditing and Accountability Journal, 34(1), 58 to 84. Fugener, A., Grahl, J., Gupta, A., and Ketter, W. (2022). Cognitive challenges in human artificial intelligence collaboration: Investigating the path toward productive delegation. Information Systems Research, 33(2), 678 to 696. Gabriel, I., Manzini, A., Keeling, G., Hendricks, L. A., Rieser, V., Iqbal, H., and others (2024). The ethics of advanced AI assistants. Google DeepMind Technical Report. Gereffi, G., Pananond, P., and Pedersen, T. (2022). Resilience decoded: The role of firms, global value chains, and the state in COVID-19 capabilities. California Management Review, 64(2), 46 to 70. Gijsbrechts, J., Boute, R. N., Van Mieghem, J. A., and Zhang, D. J. (2022). Can deep reinforcement learning improve inventory management? Performance on lost sales, dual sourcing, and multi echelon problems. Manufacturing and Service Operations Management, 24(3), 1349 to 1368. Glikson, E., and Woolley, A. W. (2020). Human trust in artificial intelligence: Review of empirical research. Academy of Management Annals, 14(2), 627 to 660. Gupta, S., Modgil, S., Meissonier, R., and Dwivedi, Y. K. (2021). Artificial intelligence and information system resilience to cope with supply chain disruption. IEEE Transactions on Engineering Management, 71, 10496 to 10506. Hagele, S., Grosse, E. H., and Ivanov, D. (2023). Supply chain resilience: A tertiary study. International Journal of Systems Science: Operations and Logistics, 10(1), 2106936. Handfield, R. B., Graham, G., and Burns, L. (2020). Corona virus, tariffs, trade wars and supply chain evolutionary design. International Journal of Operations and Production Management, 40(10), 1649 to 1660. Hendrycks, D., Mazeika, M., and Woodside, T. (2023). An overview of catastrophic AI risks. Center for AI Safety, Technical Report. Ivanov, D. (2021). Supply chain viability and the COVID-19 pandemic: A conceptual and formal generalisation of four major adaptation strategies. International Journal of Production Research, 59(12), 3535 to 3552. Ivanov, D. (2022). Viable supply chain model: Integrating agility, resilience and sustainability perspectives, lessons from and thinking beyond the COVID-19 pandemic. Annals of Operations Research, 319, 1411 to 1431. Ivanov, D. (2023). Intelligent digital twin (iDT) for supply chain stress testing, resilience, and viability. International Journal of Production Economics, 263, 108938. Ivanov, D. (2023). The Industry 5.0 framework: Viability based integration of the resilience, sustainability, and human centricity perspectives. International Journal of Production Research, 61(5), 1683 to 1695. Ivanov, D., and Dolgui, A. (2020). Viability of intertwined supply networks: Extending the supply chain resilience angles towards survivability. A position paper motivated by COVID-19 outbreak. International Journal of Production Research, 58(10), 2904 to 2915. Ivanov, D., and Dolgui, A. (2021). A digital supply chain twin for managing the disruption risks and resilience in the era of Industry 4.0. Production Planning and Control, 32(9), 775 to 788. Jackson, I., Ivanov, D., Dolgui, A., and Namdar, J. (2024). Generative artificial intelligence in supply chain and operations management: A capability based framework for analysis and implementation. International Journal of Production Research, 62(17), 6120 to 6145. Kamalahmadi, M., Shekarian, M., and Mellat Parast, M. (2022). The impact of flexibility and redundancy on improving supply chain resilience to disruptions. International Journal of Production Research, 60(6), 1992 to 2020. Katsaliaki, K., Galetsi, P., and Kumar, S. (2022). Supply chain disruptions and resilience: A major review and future research agenda. Annals of Operations Research, 319, 965 to 1002. Kinra, A., Ivanov, D., Das, A., and Dolgui, A. (2020). Ripple effect quantification by supplier risk exposure assessment. International Journal of Production Research, 58(18), 5559 to 5578. Kosasih, E. E., and Brintrup, A. (2022). A machine learning approach for predicting hidden links in supply chain with graph neural networks. International Journal of Production Research, 60(17), 5380 to 5393. Kshetri, N. (2021). Blockchain and sustainable supply chain management in developing countries. International Journal of Information Management, 60, 102376. Miller, C. (2022). Chip War: The Fight for the World's Most Critical Technology. Scribner, New York. Modgil, S., Gupta, S., Stekelorum, R., and Laguir, I. (2022). AI technologies and their impact on supply chain resilience during COVID-19. International Journal of Physical Distribution and Logistics Management, 52(2), 130 to 149. National Institute of Standards and Technology (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). NIST, United States Department of Commerce. Naz, F., Kumar, A., Majumdar, A., and Agrawal, R. (2022). Is artificial intelligence an enabler of supply chain resiliency post COVID-19? An exploratory state of the art review for future research. Operations Management Research, 15, 378 to 398. Oroojlooyjadid, A., Nazari, M., Snyder, L. V., and Takac, M. (2022). A deep Q network for the beer game: Deep reinforcement learning for inventory optimization. Manufacturing and Service Operations Management, 24(1), 285 to 304. Panwar, R., Pinkse, J., and De Marchi, V. (2022). The future of global supply chains in a post COVID-19 world. California Management Review, 64(2), 5 to 23. Park, J. S., O'Brien, J., Cai, C. J., Morris, M. R., Liang, P., and Bernstein, M. S. (2023). Generative agents: Interactive simulacra of human behavior. Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology (UIST), 1 to 22. Pournader, M., Ghaderi, H., Hassanzadegan, A., and Fahimnia, B. (2021). Artificial intelligence applications in supply chain management. International Journal of Production Economics, 241, 108250. Rolf, B., Jackson, I., Muller, M., Lang, S., Reggelin, T., and Ivanov, D. (2023). A review on reinforcement learning algorithms and applications in supply chain management. International Journal of Production Research, 61(20), 7151 to 7179. Roscoe, S., Aktas, E., Petersen, K. J., Skipworth, H. D., Handfield, R. B., and Habib, F. (2022). Redesigning global supply chains during compounding geopolitical disruptions: The role of supply chain logics. International Journal of Operations and Production Management, 42(9), 1407 to 1434. Russell, S., and Norvig, P. (2021). Artificial Intelligence: A Modern Approach, 4th edition. Pearson, Harlow. Sarkis, J. (2021). Supply chain sustainability: Learning from the COVID-19 pandemic. International Journal of Operations and Production Management, 41(1), 63 to 73. Schick, T., Dwivedi-Yu, J., Dessi, R., Raileanu, R., Lomeli, M., Zettlemoyer, L., Cancedda, N., and Scialom, T. (2023). Toolformer: Language models can teach themselves to use tools. Advances in Neural Information Processing Systems, 36. Sharma, R., Shishodia, A., Gunasekaran, A., Min, H., and Munim, Z. H. (2022). The role of artificial intelligence in supply chain management: Mapping the territory. International Journal of Production Research, 60(24), 7527 to 7550. Shavit, Y., Agarwal, S., Brundage, M., Adler, S., O'Keefe, C., Campbell, R., and others (2023). Practices for governing agentic AI systems. OpenAI Technical Report. Sheffi, Y. (2023). The Magic Conveyor Belt: Supply Chains, A.I., and the Future of Work. MIT CTL Media, Cambridge, Massachusetts. Shih, W. C. (2020). Global supply chains in a post pandemic world. Harvard Business Review, 98(5), 82 to 89. Shinn, N., Cassano, F., Gopinath, A., Narasimhan, K., and Yao, S. (2023). Reflexion: Language agents with verbal reinforcement learning. Advances in Neural Information Processing Systems, 36. Sodhi, M. S., and Tang, C. S. (2021). Supply chain management for extreme conditions: Research opportunities. Journal of Supply Chain Management, 57(1), 7 to 16. Spieske, A., and Birkel, H. (2021). Improving supply chain resilience through Industry 4.0: A systematic literature review under the impressions of the COVID-19 pandemic. Computers and Industrial Engineering, 158, 107452. Wang, L., Ma, C., Feng, X., Zhang, Z., Yang, H., Zhang, J., and others (2024). A survey on large language model based autonomous agents. Frontiers of Computer Science, 18(6), 186345. Wichmann, P., Brintrup, A., Baker, S., Woodall, P., and McFarlane, D. (2020). Extracting supply chain maps from news articles using deep neural networks. International Journal of Production Research, 58(17), 5320 to 5336. Wieland, A., and Durach, C. F. (2021). Two perspectives on supply chain resilience. Journal of Business Logistics, 42(3), 315 to 322. Xi, Z., Chen, W., Guo, X., He, W., Ding, Y., Hong, B., and others (2025). The rise and potential of large language model based agents: A survey. Science China Information Sciences, 68(2), 121101. Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K., and Cao, Y. (2023). ReAct: Synergizing reasoning and acting in language models. Proceedings of the 11th International Conference on Learning Representations (ICLR). Zamani, E. D., Smyth, C., Gupta, S., and Dennehy, D. (2023). Artificial intelligence and big data analytics for supply chain resilience: A systematic literature review. Annals of Operations Research, 327, 605 to 632.

The cross border movement of universities is one of the most visible changes in global higher education over the last three decades. Institutions no longer wait for students to travel to them. They travel to the students. This article examines #educational_entrepreneurship as the organising idea behind that movement, and it develops a practical strategic framework for founding, operating and growing #international_branch_campuses. The study uses an integrative review of scholarly work published mainly between 2021 and 2025, combined with documented sector evidence from host countries such as the United Arab Emirates, China, Malaysia and Uzbekistan. Three theoretical lenses guide the analysis: the resource based view, institutional theory and the dynamic capabilities perspective. From these, the article builds a nine stage model called the Entrepreneurial Branch Campus Lifecycle, covering opportunity sensing, feasibility and entry mode selection, regulatory design, financial modelling, academic architecture, legitimacy building, governance, scaling and exit planning. The findings suggest that failure in cross border campus ventures is rarely caused by weak academic quality alone. It is more often caused by optimistic demand forecasts, thin capital reserves, poor partner selection and a mismatch between what the home institution promises and what the host environment allows. The article argues that successful branch campus development depends on a balance between entrepreneurial ambition and academic responsibility, and that this balance must be designed into governance from the beginning rather than repaired later. The framework is intended for university leaders, ministries, investors, quality assurance bodies and postgraduate students studying international education management. It also offers a checklist of warning signs that predict underperformance, and a research agenda for scholars working on #transnational_education. Keywords: educational entrepreneurship; international branch campus; transnational education; internationalisation of higher education; entry mode strategy; quality assurance; education hubs; higher education management 1. Introduction For most of their history, universities were tied to a single place. A university was a set of buildings in a city, and the people who wanted its degrees had to come to those buildings. That assumption has weakened. Today a student in Dubai can study for a British degree without leaving the country, a student in Suzhou can graduate from a joint venture institution with an American name above the door, and a student in Tashkent can enrol in a campus operated by a Korean or Russian university. This shift is part of a broader change in how #higher_education organises itself, and it is closely tied to the rise of #internationalization as a deliberate strategy rather than an accidental by product of academic exchange (de Wit & Altbach, 2021). The vehicle at the centre of this change is the international branch campus. Wilkins (2021) documents that the number of such campuses rose from fewer than one hundred at the turn of the century to well over two hundred and fifty by the late 2010s, and later counts place the figure above three hundred. These campuses are physical operations located in a host country, carrying the name of a foreign parent institution, and awarding qualifications that carry the parent brand. They are not simply satellite classrooms. They are ventures. They require capital, market analysis, staffing plans, regulatory approvals and a clear view of who will pay the bills and why. This is where the idea of #educational_entrepreneurship becomes useful. Entrepreneurship in the university context has usually been discussed in terms of spin out companies, technology transfer and incubators, and a large body of work has examined how the #entrepreneurial_university creates economic value from research (Correia, Marques, Silva & Ramadani, 2024; Meek & Gianiodis, 2023). Yet there is a second form of academic enterprise that receives less theoretical attention. It is the business of exporting the institution itself. Establishing a campus abroad involves opportunity recognition, resource assembly under uncertainty, risk taking, and the creation of a new organisation in an unfamiliar environment. Those are the classic features of entrepreneurial behaviour, applied to a public good rather than a private product. The problem is that many universities enter this space with enthusiasm and without a framework. Some open a campus because a competitor did. Some accept an invitation from a host government without testing the assumptions behind it. Some sign a partnership because a senior leader had a good meeting on a trade mission. The consequences are visible. Campuses close, partnerships collapse, and students are left holding qualifications from institutions that no longer exist in their country. Wilkins (2021) notes that while several campuses have grown into large purpose built facilities, others have failed to reach their recruitment and financial targets. Regulators have also become stricter, with host authorities revoking licences from providers that could not meet quality expectations. This article responds to that gap. It sets out to answer four questions. First, what does #educational_entrepreneurship mean when applied to cross border campus development, and how does it differ from ordinary internationalisation activity? Second, which theoretical perspectives best explain why institutions succeed or fail when they open campuses abroad? Third, what practical stages, decisions and controls make up a defensible strategic framework for establishing and expanding a branch campus? Fourth, what does the recent evidence from major host regions tell us about the conditions under which such ventures thrive? The article is written for an audience of students, early career researchers and practitioners. It therefore uses plain language while keeping the structure of a research paper. Section two reviews the literature. Section three sets out the theoretical foundations. Section four explains the method. Section five presents the core framework, the Entrepreneurial Branch Campus Lifecycle. Section six examines regional evidence. Section seven analyses risk and failure. Section eight discusses implications. Section nine addresses limitations and future research, and section ten concludes. 2. Literature Review 2.1 From academic entrepreneurship to institutional export The literature on academic enterprise has grown quickly. Reviews of the field identify a set of recurring themes: entrepreneurial ecosystems, organisational structures, capacity building, impact evaluation, contextual influence and sustainability (Correia et al., 2024). Most of this work looks inward. It asks how a university can turn its own knowledge into economic value inside its home region. Meek and Gianiodis (2023) go further and question whether the classic model of the entrepreneurial university has run its course, arguing that the model needs rebuilding around broader social outcomes rather than narrow commercial ones. What this body of work rarely addresses is the university as an exporter of itself. Yet the export decision has all the features that entrepreneurship scholars care about. It requires an institution to recognise an unmet demand, to commit resources before the outcome is known, to build a new organisation in a place where its reputation may mean little, and to accept the possibility of loss. The literature on #transnational_education has documented these features in detail without always naming them as entrepreneurial. This article treats the two literatures as complementary. 2.2 Definitions and forms of transnational provision Transnational education refers to arrangements in which learners study for a qualification awarded by an institution based in a different country. The category is broad. It includes franchised programmes, validated programmes, articulation agreements, joint and dual degrees, distance provision and full branch campuses. Wilkins and Huisman (2025) have argued that the definitions used in the field need updating, because the older categories no longer capture hybrid arrangements that mix online delivery, local partners and foreign awarding bodies. For the purposes of this article, the following forms are treated as a spectrum of #entry_mode choices, ordered by rising commitment and rising control: Distance and online delivery with no local presence. Validation, where a local provider teaches and the foreign institution awards. Franchise, where the foreign curriculum is delivered locally under licence. Articulation and progression pathways, where students transfer after local study. Joint programmes and joint institutes, often co governed. #joint_venture campuses, where ownership and control are shared with a local partner. #wholly_owned_campus operations, where the foreign institution carries full risk and full authority. The choice among these is the first strategic decision of any cross border venture, and it shapes almost everything that follows. 2.3 Why institutions go abroad Research consistently identifies a mix of motives. Financial motives are real but rarely sufficient on their own. Institutions also seek reputational reach, access to talent, resilience against downturns in inbound student mobility, and closer links with regions where their graduates already work. Studies of the determinants of cross border campus location show that economic ties, cultural proximity and institutional similarity all increase the likelihood that a campus will be established between two countries, and that these factors influence not only whether a campus appears but also how many programmes it offers (Paniagua, Villó & Escrivà-Beltran, 2022). In other words, #institutional_distance matters. Universities do not scatter campuses at random. They follow trade routes, language links and diaspora networks. Host countries have their own motives. Many governments treat foreign campuses as instruments of national development. They want to widen access without building new public universities, to keep talented students and their tuition fees inside the country, and to signal that the nation is a serious participant in the #knowledge_economy. Ennew (2023) describes how the growth of foreign campuses has been enabled by shifts in policy and regulation as much as by institutional ambition, and notes that what once looked like an unusual and risky choice is now a normal strategic option. 2.4 What the evidence says about outcomes The record is mixed, and the literature is honest about this. Wilkins (2021) reviews two decades of activity and reports that some campuses have grown steadily into large facilities while others have failed to meet their targets. Tran, Amado and Santos (2023) conduct a systematic review of challenges and success factors in transnational provision and identify recurring problems: differences in academic culture, difficulty maintaining consistent standards, staffing instability, weak local research capacity and unclear lines of authority between the parent institution and the overseas operation. The pandemic sharpened these issues. Merola, Coelen, Hofman and Jansen (2023) found that the disruption changed both the academic experience of branch campus students and the relationship between the overseas campus and its home institution. Campuses that had been treated as distant appendages suddenly needed real decision making authority, and those that had it coped better. More recent work examines the economics of small operations. Wilkins and colleagues note that many campuses begin on a very small scale, sometimes occupying only one or two floors of an office building, and often enrol fewer than one hundred students in their first year. Small size limits the student experience and constrains profitability, because tuition income alone may not cover the cost of a full academic and administrative structure. This creates a difficult trade off between prudence and viability, which the framework in section five addresses directly. 2.5 The shifting boundary between physical and digital provision A further complication has emerged in the last few years. The clean distinction between a campus abroad and a distance programme is dissolving. Institutions now run arrangements in which students attend a small local centre for seminars and support while receiving lectures from the home country, or in which a local partner provides facilities and pastoral care while the awarding institution provides everything academic. Some of these arrangements meet the traditional definition of a branch campus. Some do not. Many sit uncomfortably between the categories, which is why scholars have called for the terminology of the field to be revised (Wilkins & Huisman, 2025). This matters for more than tidiness. Regulation follows definitions. A provider classified as a distance operator may face lighter licensing requirements than one classified as a campus, even where the student experience is broadly similar. Quality assurance regimes designed for physical institutions may not examine the parts of a hybrid operation where quality is actually determined. And students may not understand what they are buying, because marketing tends to emphasise the foreign brand and to say less about who will actually teach and where. For an institution planning an entry, the practical implication is that the physical campus is now one option among several rather than the default expression of ambition. A hybrid model can be a genuine strategy rather than a compromise, and in some markets it is the only model that works, because the numbers required to support a full campus are simply not present. What a hybrid model cannot do is deliver the community, the employer proximity and the local legitimacy that a physical presence creates. The choice between them is a strategic judgement about what the institution is trying to achieve, and it deserves to be made explicitly rather than by default. 2.6 The gap this article addresses Three gaps stand out. First, much of the existing work is descriptive rather than prescriptive. It tells us what happened but not what a leadership team should do on a Monday morning. Second, the entrepreneurship literature and the transnational education literature have developed separately, which means the tools of venture design have not been applied systematically to campus development. Third, there is limited integrated guidance that covers the full lifecycle from first idea to eventual exit. This article addresses those gaps by building a framework that is grounded in published evidence and usable in practice. 3. Theoretical Framework Three theories are used together. Each explains part of the picture, and none is sufficient alone. 3.1 The resource based view The #resource_based_view holds that an organisation performs well when it controls resources that are valuable, rare, hard to imitate and supported by organisational systems. For a university considering expansion, the relevant resources include its academic brand, its curriculum, its accreditation status, its faculty, its alumni network and its financial reserves. The important question is not whether the institution has these resources at home, but whether they travel. Some resources travel well. A curriculum can be copied. An accreditation status can often be extended. Others travel badly. A research culture built over decades cannot be shipped in a container. A campus community depends on scale and on the accidental encounters that a small rented facility cannot provide. Many failures in this sector come from assuming that a resource which creates #competitive_advantage at home will do the same abroad. #brand_equity in particular behaves unpredictably. A university that is a household name in its own country may be unknown to parents making decisions in the host market, and unknown brands cannot charge premium fees. 3.2 Institutional theory and legitimacy #institutional_theory explains behaviour that the resource view cannot. Organisations do not act only to maximise efficiency. They act to appear appropriate to the audiences that matter to them. Wilkins and Huisman have long argued that decisions to open campuses abroad are shaped by concerns about legitimacy, status and risk as much as by revenue calculations. #legitimacy is a currency. A new campus must earn it from several audiences at once: the host regulator, local employers, prospective students and their families, the academic staff it hopes to recruit, and its own home community, which may be sceptical about the venture. These audiences do not want the same thing. The regulator wants compliance. Employers want #graduate_employability. Families want a recognised qualification and a safe environment. Home academics want assurance that standards are not being diluted. A campus that satisfies only one of these audiences will struggle. Institutional theory also explains imitation. When several universities open campuses in the same city within a short period, the later entrants are often responding to the earlier ones rather than to independent analysis of demand. This is a known source of oversupply and eventual #campus_closure. 3.3 Dynamic capabilities The third lens is the #dynamic_capabilities perspective, which focuses on an organisation's ability to sense opportunities, seize them and reconfigure itself as conditions change. This is the closest theoretical match to entrepreneurship as a practice, and it is particularly relevant to cross border education because host environments change quickly. Visa rules shift. Local competitors emerge. Currencies move. Governments change their development priorities. A campus designed for the conditions of its founding year will be obsolete within five. The capability to reconfigure is therefore not a luxury. It is a survival trait. It shows up in practical ways: the ability to close a weak programme quickly, to add a new one without waiting two years for home approval, to switch from one recruitment channel to another, or to renegotiate a lease. Institutions that centralise every decision in the home country lose this capability, and the evidence from the pandemic period supports that conclusion (Merola et al., 2023). 3.4 Bringing the three together The three lenses map onto three questions that every cross border venture must answer: What do we have that is worth exporting? (resource based view) Who must accept us, and on what terms? (institutional theory) How will we change as the environment changes? (dynamic capabilities) The framework presented in section five is organised so that each stage forces the leadership team to answer at least one of these questions explicitly. 4. Methodology This study is conceptual and uses an integrative review method. An integrative review gathers findings from studies that use different designs, synthesises them, and produces a new conceptual output rather than a statistical estimate. It is appropriate here because the evidence base on cross border campuses is fragmented across education research, management research, economics and policy analysis. The review proceeded in four steps. Step one: identification. Peer reviewed literature published mainly between 2021 and 2025 was identified using academic databases, focusing on the terms international branch campus, transnational education, cross border higher education, education hub, academic entrepreneurship and entrepreneurial university. Earlier foundational works were included only where later research continues to rely on them. Step two: screening. Sources were retained where they addressed institutional strategy, governance, quality, finance or outcomes in cross border provision, or where they contributed theory that could be applied to those questions. Sources dealing only with student mobility or curriculum internationalisation at home were excluded. Step three: synthesis. Findings were coded into categories reflecting the practical decisions that institutions face. These categories were refined iteratively and eventually became the nine stages of the framework. Step four: framework construction. The stages were arranged as a lifecycle rather than a linear plan, because evidence suggests that institutions revisit earlier decisions repeatedly, particularly when they expand or when regulation changes. Two limitations of the method should be stated at the outset. First, the sector suffers from survivorship bias in reporting. Successful campuses publicise themselves. Failed ones disappear quietly, and the reasons for their failure are seldom documented in the open literature. Second, much of the useful operational detail sits in consultancy reports and government documents whose methods are not always transparent. The framework therefore treats such material as supporting evidence rather than as primary proof. 5. The Entrepreneurial Branch Campus Lifecycle The framework has nine stages. They are presented in order, but in practice they overlap and repeat. Each stage includes the decisions to be made, the evidence that supports them, and the failure modes to watch for. Stage 1: Opportunity sensing and market intelligence Every venture begins with a claim about demand. The claim is usually optimistic. Serious #market_research at this stage asks five questions. Who is the student? Not a demographic abstraction, but a person with a budget, a family, a set of alternatives and a set of fears. In many host markets the buyer is not the student at all. It is the parent. That changes the message, the price sensitivity and the importance of safety and reputation. What are they buying? Some students buy a foreign degree because it opens migration pathways. Some buy it because the local system cannot offer the subject they want. Some buy it because their family cannot afford to send them overseas but wants an international qualification. These are different products with different #value_proposition statements, and a campus that does not know which one it is selling will price and market badly. What is the realistic addressable market? #demand_forecasting in this sector is notoriously weak. A common error is to take national population figures, apply a participation rate, and assume a share. A more defensible approach works upward from observable behaviour: current outbound student numbers to the parent country, enrolment at comparable existing providers in the host city, and the number of families in the relevant income bracket. If the projected first year intake exceeds what comparable providers achieved in their first three years, the projection is probably wrong. Who else is coming? Oversupply is a real risk in popular hubs. The #first_mover_advantage in this sector is genuine but temporary, and it belongs to institutions that establish regulatory relationships and employer links early, not simply to those that arrive first. What is the government actually trying to achieve? Host policy is rarely neutral. A country that wants to retain its own students will support providers who teach locally at moderate fees. A country that wants to become a regional #education_hub will support providers who can attract students from neighbouring states. These two policy goals lead to very different campus designs. Failure mode: the invitation trap. A senior official invites the university to open a campus. The invitation feels like validation. It is not evidence of demand. Invitations are cheap. Enrolments are not. Stage 2: Feasibility, risk assessment and entry mode selection Once the opportunity is described, it must be tested. A rigorous #feasibility_study covers four domains. Academic feasibility. Can the proposed programmes be delivered to the standard required by the home institution and the host regulator? Do they require laboratories, clinical placements or industry links that do not exist locally? Programmes with heavy infrastructure requirements are poor candidates for early launch. Business, computing, engineering foundations, and some social sciences travel more easily than medicine or fine art, which is why they dominate the early portfolios of most campuses. Regulatory feasibility. What licences are required, how long do they take, and what happens if they are refused? #regulatory_compliance is not a formality. Some jurisdictions require local ownership. Some require that the campus teach national curriculum components. Some restrict the repatriation of surplus. Some require the campus to be located in a designated zone. Financial feasibility. This is covered in stage four, but the feasibility stage must at least establish the order of magnitude and the tolerance for loss. Political and reputational feasibility. Would association with the host state create difficulties at home? Questions about #academic_freedom, labour conditions, or the treatment of particular groups have derailed proposed campuses before construction began. These questions should be answered before commitment, not during a public controversy. The output of this stage is an #entry_mode decision. The table below summarises the trade offs. Mode Capital required Control Speed Risk to brand Best suited to Online only Low High Fast Moderate Testing demand Validation Low Low Fast High Mature partners, strong QA Franchise Low Medium Fast High Standardised programmes Joint institute Medium Shared Medium Medium Regulated markets such as China Joint venture campus Medium to high Shared Slow Medium Markets requiring local ownership Wholly owned campus High High Slow Low if managed well Deep, long term commitment The general rule is that control and risk move together. Low commitment modes protect capital but expose the brand, because the institution is trusting others with its name. High commitment modes protect the brand but expose the balance sheet. Many institutions now use a staged approach, beginning with a low commitment mode to test the market and converting to a campus only if the early cohorts perform. Failure mode: choosing the mode before doing the analysis, usually because a partner has already been identified and the relationship has developed political momentum inside the institution. Stage 3: Partner selection, regulation and quality architecture Where a partner is required, #partner_selection is the single highest leverage decision in the venture. Effective #due_diligence on a prospective partner covers ownership structure, financial strength, reputation with the regulator, other education interests that may conflict, litigation history, and the personal reliability of the individuals who will actually run the operation. It also covers motive. A partner whose primary interest is real estate will behave differently from one whose primary interest is education, and both are common. #contract_design should address, at minimum: the division of academic authority, the appointment and removal of the campus head, the approval process for new programmes, the ownership of #intellectual_property, the treatment of surpluses and deficits, the handling of #transfer_pricing between entities, dispute resolution, and the conditions under which either party may withdraw. A contract that does not specify an exit is a contract that guarantees a bad one. #quality_assurance architecture must satisfy two regulators at once. The home regulator will normally require that the overseas award is equivalent in standard to the home award. The host regulator will have its own requirements, which may include local content, local staffing quotas, or specific facilities. Where the two regimes conflict, the conflict must be resolved in writing before students are admitted. Practical design principles for #accreditation and quality: Assessment standards should be set and moderated by the home institution, with external examiners who see scripts from both locations. Programme approval should follow the home process, with an added host context review. Data on progression, retention and attainment should be reported by campus and compared, not aggregated into a single institutional figure that hides divergence. Student complaints and academic appeals should have a route that does not depend solely on local management. #data_governance is increasingly part of this picture. Student records cross borders, and privacy law in the host country may not permit the arrangements the home institution assumes. Failure mode: treating quality assurance as a compliance exercise to be completed once, rather than a continuing control system. Stage 4: Financial modelling and resource commitment This is where most optimistic ventures break. The #cost_structure of a branch campus is unusual. A high proportion of costs are fixed and must be incurred before the first student arrives: premises, licensing, senior staff, marketing, and the systems needed to run an institution. Revenue, by contrast, builds slowly, because enrolment accumulates cohort by cohort. A three year undergraduate programme reaches full enrolment only in its third year. This produces a deep and predictable cash trough. A defensible financial model should include the following components. #capital_expenditure. Premises, fit out, laboratories, library, technology, and the reserve required by the regulator. In many jurisdictions a financial guarantee or bond must be lodged, and this capital is not available for operations. #operating_expenditure. Academic salaries, which in expatriate heavy models include housing, schooling and flights; administrative salaries; rent; marketing; and a management fee or royalty paid to the home institution, which is often set as a percentage of gross tuition and which can quietly make a campus unviable if set too high. Revenue. #tuition_revenue is the dominant line. Scholarship discounting is the most commonly underestimated cost. Campuses seeking to build enrolment often discount heavily in early years, and headline fees bear little relationship to realised income per student. Models should be built on net fee, not list fee. #break_even analysis. The key question is not when the campus becomes profitable but how much cash it will consume before it does. Many campuses require five to seven years to reach operational break even, and some never do at scale. Sensitivity testing. The model should be stressed against a first intake at half the forecast, a fee discount ten percentage points deeper than planned, a delay of one year in licensing, and a significant currency movement. #currency_risk deserves specific attention where costs are incurred in one currency and fees are collected in another. #economies_of_scale. The evidence on small campuses is instructive. Small operations reduce initial exposure but struggle to fund the services that students expect and that regulators require. The result can be a trap: too small to be good, too weak to grow. A credible model therefore identifies the minimum viable scale and the pathway to reach it, rather than treating small size as a permanent virtue. Failure mode: funding the venture from the parent institution's operating surplus without a ring fenced budget, so that the true cost is invisible until it becomes a crisis. A worked illustration. Consider a modest campus planning three undergraduate programmes with a target intake of one hundred and fifty students in year one, rising to two hundred and fifty by year three, with a three year degree structure. Full enrolment is therefore not reached until year three, when the first cohort is in its final year and two later cohorts sit below it. If the list fee is set at a level competitive with local private universities, and average discounting reaches twenty five per cent through scholarships and sibling concessions, realised income per student is materially lower than the headline figure suggests. Meanwhile, the campus must pay for a full academic leadership team, programme leaders for each subject, professional services staff, premises and marketing from day one. In this structure, the cumulative cash requirement typically peaks somewhere in year three or year four, not year one, which is precisely when institutional attention has moved on and when internal patience is thinnest. Three consequences follow. First, the sponsor must commit the full multi year funding envelope at the outset, not year by year, because a venture that is starved in year three fails for reasons that have nothing to do with its underlying viability. Second, governance must define in advance what evidence would justify additional investment and what evidence would justify withdrawal, so that the decision in year three is made against a standard agreed in year zero rather than against the mood of the moment. Third, any royalty or management fee charged by the parent should be waived or deferred until operational break even, since charging a percentage of gross tuition to an operation that is not yet covering its costs simply moves the loss from one balance sheet to another while making the campus look worse than it is. The wider lesson is that the finances of a cross border campus are not merely a smaller version of the finances of the home institution. The cost base is front loaded, the revenue is back loaded, and the gap between them is the venture. Institutions that understand this build reserves. Institutions that do not build excuses. Stage 5: Academic model, curriculum and staffing A campus is not a building. It is a teaching operation. #curriculum_design faces a tension between consistency and relevance. The home institution needs the overseas award to be equivalent to its own. The host market needs graduates who can work in the local economy. The answer is not to choose one, but to specify which elements are fixed and which may vary. Learning outcomes, assessment standards and credit structures are normally fixed. Case studies, examples, industry projects, guest speakers and elective content should vary. This distinction between fixed and flexible components is the practical content of #localisation, sometimes described as #glocalisation. #learning_outcomes must be demonstrably comparable across locations. Comparability is proved by evidence, not by assertion. External examiner reports, moderated samples, and comparative attainment data are the evidence. #faculty_recruitment is the hardest operational problem. Cross border campuses compete for staff against home institutions, against local providers and against industry. Research shows that academic careers at overseas campuses carry real rewards, including autonomy, career acceleration and international experience, but also real costs, including isolation from the research community, heavy teaching loads, insecure contracts and limited progression. Practical responses that appear in successful operations include: Multi year contracts rather than annual renewals, which reduce turnover in the second and third year when institutional memory matters most. A guaranteed research allocation, even a modest one, so that #academic_staff do not see the posting as the end of their scholarly life. Structured mobility between campuses, so that home staff spend time overseas and overseas staff spend time at home. Local hiring where the labour market permits, because a campus staffed entirely by short term expatriates cannot build continuity. #student_experience must be planned, not assumed. Students at overseas campuses often report that they receive the academic content of the parent institution without the community. Clubs, sport, careers services, counselling and physical space are not luxuries. They are part of the product, and they influence retention, which influences finance. Failure mode: exporting the curriculum and forgetting to export the institution. Stage 6: Brand, legitimacy and student recruitment A campus must persuade people who have never heard of it to spend a significant sum on a promise. #reputation_management begins with clarity about what the campus is. Students and families need to understand whether the qualification is identical to the home award, whether transfer to the home campus is possible, and what the qualification is worth in the local labour market. Ambiguity on these points is corrosive, because it is quickly exposed and it damages trust. #student_recruitment in host markets usually depends on a mix of channels: direct enrolment from local schools, agent networks, corporate sponsorship, government scholarship schemes, and regional recruitment from neighbouring countries. Each has a different cost per enrolled student and a different quality profile. Agent networks deliver volume but require careful management, because agents optimise for commission, not for fit, and poor fit produces attrition. #stakeholder_engagement builds the legitimacy that marketing cannot buy. The most effective mechanisms are practical: #industry_partnerships that produce internships and placements, which give families concrete evidence of employability. Advisory boards containing respected local employers and professionals. Professional body recognition of programmes, which in many markets matters more than university rankings. Public engagement through open lectures, school outreach and community projects, which builds #social_impact and gives the campus a civic identity. #alumni_networks, which take years to become useful but eventually become the strongest recruitment asset the campus has. Failure mode: relying on the parent institution's global ranking as if it were a local marketing strategy. Rankings impress ministries. They do not, on their own, fill classrooms. Stage 7: Governance, control and the home to host relationship The relationship between the parent institution and the overseas campus is a permanent negotiation between control and responsiveness. Too much central control produces an operation that cannot respond to local conditions. Programme approvals take eighteen months while competitors launch in three. Marketing copy must be approved in another time zone. Salaries are benchmarked against the home country rather than the local labour market. The campus becomes slow and uncompetitive. Too little control produces drift. Standards diverge quietly. The campus takes on activities the parent did not authorise. Financial reporting becomes unreliable. In the worst cases the parent discovers a problem only when the regulator raises it. Effective #governance separates the decisions that must be reserved from those that must be delegated. Typically reserved to the parent: award of qualifications, academic standards and assessment regulations, programme approval, appointment of the senior campus academic officer, use of the institutional name, and any commitment above a defined financial threshold. Typically delegated to the campus: operational budget within approved limits, local hiring below senior level, marketing execution, timetabling, student services, and local partnerships that do not involve the award of credit. #performance_indicators should be agreed in advance and reported on a fixed cycle. A workable set includes applications, conversion rate, net fee per student, cohort retention, progression, attainment compared with the home campus, staff turnover, graduate employment at six months, cash position and regulatory standing. #benchmarking against comparable providers in the same host market is more useful than benchmarking against the home institution. #mission_drift deserves separate attention. It occurs when the financial logic of the campus begins to override its academic purpose: entry standards fall to fill seats, marginal programmes are launched because they are cheap to run, or assessment becomes lenient because attrition threatens cash flow. The defence against drift is structural, not moral. It requires that #academic_integrity decisions sit with people whose performance is not measured by enrolment. Failure mode: a governance structure that exists on paper and is bypassed in practice because the campus is too far away and the parent is too busy. Stage 8: Scaling, diversification and portfolio expansion Once a campus is established and stable, the entrepreneurial question changes. It is no longer whether to enter, but how to grow. Growth has four directions. Deepen the programme portfolio. Add postgraduate provision on top of an undergraduate base. Postgraduate students are usually cheaper to recruit and more willing to pay, but they demand stronger faculty and often expect research supervision, which raises cost. Widen the student base. Recruit regionally rather than only locally. This is the point at which a campus becomes part of an #education_hub strategy and begins to serve the host government's ambition rather than merely operating within it. Add delivery modes. #blended_learning and #digital_learning allow a campus to serve students who cannot attend daily, to share specialist teaching across campuses, and to test demand in neighbouring countries before committing physical resources. Hybrid provision is the fastest growing part of transnational activity, and it is also the least regulated, which creates both opportunity and hazard. Replicate. Some institutions build a network of campuses. This introduces new capabilities: managing a portfolio, allocating scarce academic leaders across sites, and preventing campuses from competing with one another for the same students. #scalability should be designed in from the beginning. Practically, this means signing leases with expansion options, building systems that can handle several times the current enrolment, writing partnership agreements that permit new programmes without renegotiation, and appointing leaders capable of running an institution three times the size of the one they inherit. The #business_model of a mature campus usually diversifies beyond tuition. Executive education, professional short courses, contract research, corporate training and consultancy can stabilise income and deepen employer relationships. They also strengthen the campus's case for #financial_sustainability without pressing on undergraduate fees. Failure mode: scaling before the core operation is sound, so that expansion multiplies weaknesses instead of strengths. Stage 9: Contingency, exit and responsible closure Every venture should be designed with the possibility of ending. This is uncomfortable to discuss and essential to plan. An #exit_strategy is not an admission of expected failure. It is a condition of responsible operation, because the people most harmed by an unplanned closure are students who have completed part of a degree. A credible plan includes: Trigger points. Agreed thresholds that prompt a formal review, such as enrolment below a stated percentage of forecast for two consecutive years, or the loss of a licence. Teach out provision. A funded commitment to allow enrolled students to complete their qualification, either at the campus, at a partner institution, or at the home campus. Records continuity. A guarantee that transcripts and award records remain accessible after closure, since graduates may need them decades later. Staff obligations. Notice periods, repatriation and severance, which in expatriate heavy operations carry visa consequences as well as financial ones. Regulatory notification. Host authorities normally require formal notice and may impose conditions. #political_risk and regional instability make contingency planning more urgent, not less. Institutions operating in volatile regions have had to consider the safety of staff and students, the continuity of teaching during disruption, and the reputational effect of appearing to abandon a market. Failure mode: closing a campus in a way that damages students, and therefore damaging the institutional brand far more than the original financial loss. 6. Regional Evidence The framework is general. Its application is not. This section examines four settings that illustrate different logics. 6.1 The United Arab Emirates and the hub model #Dubai is the clearest example of a deliberately constructed #education_hub. The emirate created dedicated zones for education providers, offered regulatory arrangements designed to attract foreign institutions, and treated higher education as part of a wider development strategy alongside trade, aviation and finance. The result is one of the largest concentrations of foreign campuses in the world. Two lessons emerge. First, the hub model works when the host offers not merely permission but infrastructure: premises, visa arrangements, marketing of the destination and regulatory clarity. Second, hubs are competitive environments. A high concentration of providers means that students have many options and that weak providers are exposed. Regulators in the region have shown willingness to withdraw licences from institutions that fail quality reviews, which is a healthy signal for the market and a warning for entrants who assume that a licence, once granted, is permanent. 6.2 China and the regulated joint model #China has taken a different path. Foreign provision is generally organised through joint institutes and joint venture universities in partnership with Chinese institutions, within a regulatory framework that specifies ownership, governance and curriculum requirements. This produces high commitment ventures with shared control. Research on governance in Sino foreign joint institutes highlights the importance of the people who work between the two partners, and the way their perceptions shape how the institution actually operates. The formal agreement matters, but so does the everyday translation work done by staff who must reconcile two academic cultures. For institutions considering this market, the implication is clear: the quality of the working relationship is not a soft factor. It is the operating system of the venture. 6.3 Malaysia and the mature market #Malaysia has hosted foreign campuses for decades and offers a picture of what maturity looks like. Campuses there have grown to substantial size, developed research activity and built local staffing. They also faced the full force of pandemic disruption, and studies of that period show how the relationship between overseas campuses and their parents changed under stress, with campuses that had genuine autonomy adapting faster (Merola et al., 2023). A mature market rewards different capabilities from a new one. Growth comes from reputation, employer relationships and postgraduate provision rather than from being new and foreign. 6.4 Uzbekistan and the rapid entrant #Uzbekistan is the most striking recent development. Between 2018 and 2022 more than twenty foreign institutions established campuses there, moving the country to third place globally as a host of branch campuses, behind China and the United Arab Emirates (Muratov & Wilkins, 2024). The government drove this expansion deliberately, seeking rapid expansion of higher education capacity and international standards. The case is instructive for #policy_makers and institutions alike. Rapid growth compresses the timeline for everything: licensing, staffing, quality systems and student recruitment. It also concentrates risk, because a large number of providers enter a market whose absorptive capacity has not been tested. The early outcomes reported in the research are cautiously positive, but the honest conclusion is that it is too early to judge. What can be said is that the entrants who arrived with realistic scale expectations, local partnerships and patient capital appear better placed than those who arrived expecting immediate volume. 6.5 Cross cutting observations Across these settings, several patterns hold. Government support accelerates entry but does not guarantee demand. Regulatory strictness, though inconvenient, protects serious providers by removing weak competitors. Institutions that build local staff capacity outperform those that rely permanently on rotating expatriates. And campuses that connect to the local economy through employers and professional bodies build legitimacy faster than those that rely on the parent's international reputation. The OECD's recent analysis of policies for transnational collaboration observes that institutions operating across borders must navigate a patchwork of policy levers spanning quality assurance, visas, security and trade, developed at different levels of government and sometimes pulling in different directions (OECD, 2025). This is a useful corrective to the assumption that a single ministry can clear the path. 7. Risk Analysis and Failure Modes The framework is easier to justify when the failure patterns are stated plainly. #risk_management in this sector is not about eliminating risk but about knowing which risks will actually arrive. Demand risk. The most common and the most damaging. Forecasts are built on assumptions that were never tested against comparable providers. The remedy is external validation of forecasts and a financial model that survives an intake at half the projection. Capital risk. Ventures are funded thinly, on the assumption that revenue will arrive quickly. The remedy is to fund the full cash trough at the outset, plus a contingency reserve, and to treat the venture as an investment with a defined loss tolerance. Partner risk. The local partner's interests diverge from the university's. The remedy is thorough due diligence, contractual clarity on control and exit, and a relationship that does not depend on a single individual on either side. Regulatory risk. Rules change, licences are reviewed, requirements are added. The remedy is continuous engagement with the regulator, conservative compliance and a policy of over documenting rather than under documenting. Quality risk. Standards diverge from the home institution and are discovered late. The remedy is comparative data reported by campus, external examining that spans both locations, and a reporting line for academic standards that bypasses local management. Reputational risk. A problem at the campus becomes a story about the university. The remedy is to treat the overseas campus as fully part of the institution in communications and crisis planning, not as a subsidiary that can be disowned. People risk. High staff turnover destroys continuity. The remedy is contracts, career paths and research time. Political and security risk. Instability disrupts operations. The remedy is contingency planning, insurance, and clear duty of care arrangements for staff and students. Student risk. This is the one that matters most ethically. When a venture fails, students bear the consequences. The remedy is a funded teach out commitment established before the first student enrols. A simple diagnostic can be applied to any proposal. If the answer to three or more of the following is uncertain, the venture is not ready: Can we name the first two hundred students and explain why they will choose us over the alternatives? Have we modelled the outcome if the first intake is half of forecast? Do we know exactly who has authority to approve a new programme, and how long it takes? Have we agreed, in writing, what happens if we want to leave? Do we have a funded plan to allow enrolled students to complete their degrees if we close? Is anyone in the governance chain responsible for academic standards whose pay is not linked to enrolment? 8. Implications 8.1 For university leaders Treat cross border expansion as a venture, not as a project. Ventures have sponsors, capital limits, staged approvals and defined exits. Projects have enthusiasm and deadlines. The evidence suggests that institutions which apply venture discipline to campus development make fewer catastrophic errors. Build the governance before the building. It is much easier to agree the division of authority when there is no money at stake and no crisis in progress. Accept that #strategic_planning for an overseas campus is planning under deep uncertainty. Plan in stages, with real decision points at which the institution can stop. 8.2 For host governments and regulators Clarity attracts quality. Providers with strong reputations avoid markets where the rules are ambiguous, because ambiguity threatens their brand. A clear, demanding and predictable regulatory framework is therefore an instrument of attraction, not a barrier. Regulate the outcome, not only the input. Requirements about square metres and library holdings matter less than progression, attainment and graduate employment. Protect students explicitly. Require a funded teach out plan as a condition of licence. 8.3 For students Students considering an overseas campus should ask concrete questions. Is the qualification identical to the home award? Who teaches the programme, and are they permanent staff? Is transfer to the home campus possible, and at what cost? Is the programme recognised by the professional body that governs the career you want? What happens to your degree if the campus closes? These questions are reasonable, and an institution that cannot answer them clearly has told you something important. 8.4 For the field #educational_entrepreneurship deserves to be recognised as a distinct research area, sitting between entrepreneurship studies and higher education studies. It concerns the creation of new educational organisations under uncertainty, and it raises questions that neither parent field answers alone: how to balance mission and margin, how to build legitimacy across cultures, and how to grow without diluting the thing that made growth possible. 9. Limitations and Future Research This study has clear limits. It is conceptual and does not test the framework empirically. It draws on published literature which under reports failure. It cannot fully capture markets where information is scarce, and its regional discussion is illustrative rather than exhaustive. The framework itself is a heuristic, not a formula, and no framework can substitute for judgement about a specific place at a specific time. Four research directions follow. Failure studies. The field needs systematic, well documented analysis of closed campuses, including access to financial and governance records. Without this, the sector will keep repeating errors that it cannot see. Longitudinal outcome studies. What happens to graduates of overseas campuses over ten and twenty years, compared with graduates of the home institution and with local alternatives? This is the question that matters most and is asked least. Hybrid and digital models. As delivery becomes more mixed, the definition of a campus becomes unstable. Research is needed on how quality is assured, how legitimacy is built and how students experience institutions that exist partly in physical space and partly online (Wilkins & Huisman, 2025). New host regions. Africa, Central Asia and parts of South East Asia are becoming active hosts. The conditions there differ from those in the established hubs, and frameworks derived from the Gulf and East Asia may transfer poorly. Sustainability and social contribution. Beyond financial performance, what is the contribution of a foreign campus to the host society? Does it build local capacity or compete with it? #sustainability in this sense is not a marketing term. It is an empirical question that the field has largely avoided. 10. Conclusion Establishing a university campus in another country is one of the boldest things an academic institution can do. It commits money, reputation and people to a place where the institution has no history and no automatic authority. Done well, it widens access, builds local capacity, and creates opportunities for #students who would otherwise have none. Done badly, it wastes public money, damages a reputation built over centuries, and leaves students holding worthless promises. This article has argued that the difference between these outcomes is not luck. It is design. The nine stages of the Entrepreneurial Branch Campus Lifecycle set out the decisions that must be made and the evidence that should support them: sensing the opportunity honestly, testing feasibility rigorously, choosing an entry mode that matches the institution's appetite for risk, selecting partners with care, funding the venture properly, designing an academic model that is both consistent and relevant, earning legitimacy from the audiences that matter, governing with a clear division of authority, growing deliberately, and planning for an ending that protects students. The deeper argument is that entrepreneurship and academic responsibility are not opposites. An institution that ignores the commercial realities of cross border operation will fail, and its students will pay the price. An institution that thinks only about the commercial realities will drift from its purpose, and its students will pay a different price. The task of leadership is to hold both at once. The framework offered here is an attempt to make that task a little more manageable, and to give the next generation of #higher_education leaders something more reliable than optimism to work with. #educational_entrepreneurship #international_branch_campus #transnational_education #cross_border_higher_education #internationalisation_of_higher_education #university_expansion_strategy #education_hubs #higher_education_management #global_campus_strategy #academic_governance #university_business_model #international_education_policy #branch_campus_quality #entrepreneurial_universities #global_education_research References Correia, M. P., Marques, C. S., Silva, R., & Ramadani, V. (2024). Academic entrepreneurship ecosystems: Systematic literature review and future research directions. Journal of the Knowledge Economy, 15(4), 17498-17528. https://doi.org/10.1007/s13132-024-01819-x de Wit, H., & Altbach, P. G. (2021). Internationalization in higher education: Global trends and recommendations for its future. Policy Reviews in Higher Education, 5(1), 28-46. https://doi.org/10.1080/23322969.2020.1820898 Ennew, C. T. (2023). The international branch campus: Motives and mechanisms. In L. Engwall (Ed.), Internationalization in higher education and research (Higher Education Dynamics, Vol. 62). Springer. https://doi.org/10.1007/978-3-031-47335-7_11 Meek, W. R., & Gianiodis, P. T. (2023). The death and rebirth of the entrepreneurial university model. Academy of Management Perspectives, 37(1), 55-71. Merola, R. H., Coelen, R. J., Hofman, W. H. A., & Jansen, E. P. W. A. (2023). Through the looking glass: How the COVID-19 pandemic changed international branch campuses' academic experience and home campus relationship. Journal of Studies in International Education, 27(1), 3-21. https://doi.org/10.1177/10283153211070112 Muratov, B., & Wilkins, S. (2024). The development of Uzbekistan as a transnational higher education hub: Government and institution rationales, and early outcomes. Journal of Higher Education Policy and Management, 46(5), 494-511. https://doi.org/10.1080/1360080X.2024.2324404 OECD. (2025). Policies for transnational collaboration in higher education. OECD Publishing. Paniagua, J., Villó, C., & Escrivà-Beltran, M. (2022). Cross-border higher education: The expansion of international branch campuses. Research in Higher Education, 63, 1037-1057. https://doi.org/10.1007/s11162-022-09674-y Tran, N. H. N., Amado, C. A. F., & Santos, S. P. (2023). Challenges and success factors of transnational higher education: A systematic review. Studies in Higher Education, 48(1), 113-136. https://doi.org/10.1080/03075079.2022.2121813 Wilkins, S. (2021). Two decades of international branch campus development, 2000-2020: A review. International Journal of Educational Management, 35(1), 311-326. https://doi.org/10.1108/IJEM-08-2020-0409 Wilkins, S., & Huisman, J. (2025). Transnational education redefined. International Higher Education, 121, 16-17.

Startups are usually described as engines of novelty. They are praised for breaking rules, ignoring old habits, and inventing new ways of organizing work. Yet when we look closely at how startups are actually governed once they raise money from professional investors, we find something surprising. Their internal structures look remarkably alike. A software venture in Nairobi, a fintech venture in Dubai, a logistics venture in Jakarta, and a health venture in Sao Paulo often carry the same board design, the same class of preferred shares, the same vesting rules, the same reporting cycle, and often the same legal home. This paper asks a simple question with complicated answers. Why do organizations that are supposed to be different end up looking the same, and what role do #global_investors play in producing that sameness? The paper uses #neo_institutional_theory, and in particular the concept of #institutional_isomorphism, to explain the standardization of #startup_governance. It develops a conceptual framework that connects three isomorphic pressures, namely #coercive_isomorphism, #mimetic_isomorphism, and #normative_isomorphism, to the specific mechanisms through which venture capital funds, accelerators, development finance institutions, sovereign funds, and limited partners transmit governance templates across borders. The study is an integrative conceptual review. It synthesizes recent scholarship on entrepreneurial finance, comparative corporate governance, and organization theory, and it uses widely documented industry practices as illustrative material rather than as original primary data. Three findings are advanced. First, standardization is not mainly the result of law. It is the result of #capital_dependence. The most powerful governance rules in a startup are usually contractual, not statutory, and they travel with the money. Second, standardization delivers real benefits, including faster access to capital, cheaper due diligence, credible exits, and protection for minority holders, but it also produces costs that are rarely counted, including #ceremonial_compliance, weak fit with local realities, and a narrowing of organizational imagination. Third, founders are not passive. They respond with selective adoption, symbolic conformity, and #decoupling, which means that a structure can exist on paper while the real decision process runs somewhere else. The paper concludes that #startup_governance is best understood as a negotiated settlement between the need for #legitimacy and the desire for #autonomy. It offers implications for founders, investors, regulators, and business educators, and it proposes a research agenda that treats the venture capital contract as a global institution in its own right. Keywords: institutional isomorphism, corporate governance, venture capital, startups, legitimacy, decoupling, emerging markets, entrepreneurial finance 1. Introduction 1.1 The puzzle of sameness in a world of difference Entrepreneurship literature tends to celebrate variety. New ventures are described as experiments. They test business models, they discover markets, and they are expected to differ from one another because the problems they solve differ. If this were the full picture, we would expect the internal machinery of startups to look as varied as their products. The reality is different. Once a startup takes institutional money, its internal machinery converges on a familiar shape. There is a board of directors with a defined number of seats. There is at least one seat reserved for the lead investor. There is a class of #preferred_shares carrying liquidation preference and anti dilution protection. There is a #founder_vesting schedule, usually four years with a one year cliff. There is an #employee_stock_option_pool carved out before the round closes. There is a set of #protective_provisions that requires investor consent for a list of major actions. There are #information_rights that require monthly or quarterly reporting in a defined format. There is a #capitalization_table maintained in a standard structure. Very often there is also a holding company registered in a jurisdiction that the investor recognizes and trusts. This convergence is the phenomenon the present paper studies. The pattern is not trivial. It means that the governance of the entrepreneurial economy is being written less by parliaments and more by the private contracts of a small number of #capital_providers whose expectations circulate globally. 1.2 Why institutional theory is the right lens Economics gives one explanation. Contracts converge because they are efficient. They solve #agency_problems, they reduce #information_asymmetry, and they align incentives between founders who have private information and investors who supply cash and cannot watch every decision. That explanation is powerful and partly correct, but it is incomplete. It cannot easily explain why the same governance forms appear in places where they do not solve any local problem, why they appear before the problems they are supposed to solve exist, or why they persist even when they clearly do not work. #Neo_institutional_theory offers a second explanation that complements the first. Organizations adopt structures not only because those structures work, but because those structures are expected. Structures function as signals of seriousness. They tell the outside world that this organization is a real, investable, and responsible entity. In the classic formulation, organizations become similar because they operate inside the same #organizational_field, and fields exert pressure toward homogeneity through force, imitation, and professional norms. The startup world is a nearly perfect case for this lens. It is young, it is uncertain, it is dependent on external resources, and it is populated by professional intermediaries who share training, vocabulary, and templates. These are precisely the conditions under which isomorphism is predicted to be strongest. 1.3 Research questions The paper is organized around four questions. Through which specific mechanisms do #global_investor_expectations produce standardized governance structures in startups? Which governance elements converge most strongly, and which retain local variation? What are the consequences of that convergence for the venture, for the founder, and for the wider entrepreneurial ecosystem? How do founders and local ecosystems respond, and under what conditions does convergence become symbolic rather than substantive? 1.4 Contribution The contribution is threefold. Theoretically, the paper extends isomorphism arguments from mature and listed firms, where most governance research sits, into the private and early stage domain where governance is contractual rather than statutory. It proposes that the venture capital term sheet should be treated as a carrier of institutional order, comparable in influence to a national code of corporate governance. Conceptually, it distinguishes between #structural_convergence, which is the visible adoption of forms, and #practice_convergence, which is the actual behavior inside those forms, and argues that the gap between the two is where most of the interesting action lies. Practically, it gives founders and policymakers in capital importing regions a clear map of which pressures are negotiable and which are not. 1.5 Structure of the paper Section two develops the theoretical background. Section three describes the global venture field and its dominant governance template. Section four presents the conceptual framework and propositions. Section five explains the method. Section six presents the analysis. Section seven discusses the findings. Section eight sets out implications, and section nine addresses limitations and a future research agenda before the conclusion. 2. Theoretical Background 2.1 The foundations of neo institutional theory Classical organization theory assumed that structure follows function. Firms build the structures that make them efficient. Institutional theory challenged that assumption. It argued that many formal structures are adopted because they are socially expected, not because they demonstrably improve output. Organizations are embedded in environments full of rules, beliefs, and taken for granted assumptions about what a proper organization looks like. Conforming to those assumptions buys #legitimacy, and legitimacy buys survival, because it unlocks resources, partners, employees, and customers. The seminal statement of #institutional_isomorphism argued that once a field is established, powerful forces push organizations within it toward similarity. Three mechanisms were identified. Coercive isomorphism arises from formal and informal pressure exerted by organizations on which a focal organization depends, and from cultural expectations in the wider society. The pressure may be legal, such as a regulation, or it may be contractual, such as a funding condition. The key feature is dependence. When one party controls a resource that the other party needs, that party can impose structure. Mimetic isomorphism arises from uncertainty. When goals are ambiguous, when technology is poorly understood, and when the environment produces confusing signals, organizations copy others that appear successful. Copying is cheap and it is defensible. Nobody is blamed for doing what the market leader does. Normative isomorphism arises from professionalization. Members of a profession share education, credentials, conferences, associations, and career paths. They carry a common set of ideas about the right way to organize. As professionals move between organizations, they carry templates with them, and the field becomes more uniform. These three mechanisms are analytically separate but empirically entangled. In practice a single governance clause can be simultaneously imposed by an investor, copied from a peer, and recommended by a lawyer. 2.2 Legitimacy and the liability of newness New ventures suffer from what organizational sociology calls the #liability_of_newness. They lack a track record. They lack routines. They lack relationships with suppliers, employees, and financiers. They cannot yet prove that they will still exist next year. Under such conditions, external audiences must judge the venture using proxies, and structure becomes a proxy. A startup that presents a clean cap table, a properly constituted board, audited or at least reviewed accounts, and standard shareholder agreements is not necessarily better at building products. It is, however, easier to believe in. It reduces the perceived risk of the person on the other side of the table. This is the sense in which #legitimacy operates as currency. It is the social equivalent of collateral. Legitimacy has several forms. #Regulative_legitimacy comes from compliance with laws and rules. #Normative_legitimacy comes from conformity with professional values about what is right and proper. #Cognitive_legitimacy comes from being understandable and taken for granted, so that the audience does not even have to think about whether the form is acceptable. Startups pursuing global capital need all three, and governance structures are the cheapest way to buy them quickly. 2.3 Resource dependence as the engine of coercion Institutional theory explains why forms spread. #Resource_dependence_theory explains who has the power to spread them. The core proposition is straightforward. The party that controls a critical and scarce resource, and that cannot easily be replaced, holds power over the party that needs it. Power is then converted into influence over structure. Startups are, almost by definition, resource dependent. They consume cash before they generate it. The supply of early stage risk capital is concentrated among a relatively small number of funds, and in many markets outside a few dominant hubs, the number of credible investors is small enough to count on one hand. That concentration converts investor preferences into effective rules. When a founder in a thin capital market is told that a certain structure is standard, the practical meaning is that the alternative is no funding. 2.4 Agency theory and the economic rationale for the template It would be a mistake to treat governance standardization as pure ceremony. #Agency_theory provides a serious efficiency logic. The investor is a principal who supplies capital and cannot observe daily decisions. The founder is an agent with superior information and different incentives. The classic problems follow. There is a risk of #moral_hazard, where the founder consumes perquisites or pursues personal glory rather than value. There is a risk of #adverse_selection, where the weakest ventures are the most eager to raise. There is a risk of #hold_up, where the founder threatens to leave once the money is committed. The standard venture template answers each risk with a specific instrument. #Founder_vesting answers hold up. #Board_seats and #information_rights answer monitoring. #Liquidation_preference answers downside risk. #Anti_dilution answers valuation risk. #Protective_provisions answer the risk of unilateral action. #Staged_financing, in which capital is released in tranches against milestones, answers the risk of throwing good money after bad. The intellectual point of this paper is not that these instruments are irrational. It is that their diffusion has outrun their justification. Once a package of instruments becomes the definition of a professional deal, it is adopted even where the underlying agency problem is mild, because not adopting it is read as a warning sign. Efficiency creates the template. Institutional pressure spreads it far beyond the conditions that created it. 2.5 Decoupling and the gap between form and practice The most useful idea for understanding what actually happens inside standardized startups is #decoupling. Decoupling occurs when an organization adopts a structure to satisfy external audiences while its real work continues according to a different logic. The board exists, but the decisions are made in a founder chat group. The audit committee exists, but it meets once a year and reads a document prepared the night before. The ESG policy exists as a document that no operating manager has read. Decoupling is not always cynical. It can be a rational response to conflicting demands. A venture that must simultaneously satisfy a global investor, a local regulator, a demanding customer, and a small team cannot fully honor every expectation, so it honors some symbolically. Recent institutional scholarship also emphasizes #means_ends_decoupling, where the organization genuinely implements a practice but the practice does not produce the outcome it promises. A startup can hold every board meeting on schedule and still receive no useful oversight, because the directors lack sector knowledge or the information pack is designed to reassure rather than inform. 2.6 Optimal distinctiveness as the counterweight If conformity were the only pressure, all ventures would be identical and none would be interesting. In reality, ventures face a second pressure to be different, because differentiation is what attracts attention, talent, and premium valuation. The concept of #optimal_distinctiveness captures this tension. Ventures must be similar enough to be understood and trusted, and different enough to be worth backing. The resolution observed in practice is a division of labor between domains. Startups conform sharply in the domains that investors use to judge risk, which are legal form, capital structure, reporting, and board process. They differentiate sharply in the domains that investors use to judge upside, which are product, market approach, technology, and narrative. Conformity in governance is what buys the right to be radical in strategy. This is a crucial and underappreciated point, and it explains why founders who despise bureaucracy nonetheless sign standard documents without much resistance. 3. The Global Venture Field and Its Governance Template 3.1 Venture capital as an organizational field An #organizational_field is the set of organizations that, taken together, constitute a recognized area of institutional life. It includes suppliers, consumers, regulators, and rival producers, and it becomes structured when interaction increases, when a hierarchy of status emerges, when information flow intensifies, and when participants develop a shared awareness that they are involved in a common enterprise. Global venture finance satisfies every one of these criteria. It has a defined membership of funds, angels, corporate venture arms, accelerators, development finance institutions, sovereign wealth vehicles, and the law firms, auditors, and data providers that serve them. It has a status hierarchy that is publicly known, in which a small number of firms and programs confer disproportionate credibility on whatever they touch. It has dense information flows through rankings, newsletters, databases, demo days, and conferences. It has a shared vocabulary that is instantly recognizable and almost identical from one continent to another. Participants speak of seed rounds, bridges, runway, burn, traction, the cap table, and the exit, regardless of whether they are sitting in London, Riyadh, Lagos, or Singapore. Crucially, the field is transnational. National corporate law varies enormously, but the venture field carries its own private ordering across those borders. A fund that raises capital from #limited_partners in one region, invests in a startup in a second region, and plans an exit in a third region has strong reasons to impose one consistent structure everywhere. Uniformity lowers its own transaction costs, simplifies reporting to its own investors, and makes its portfolio comparable across geographies. 3.2 The dominant template The template that has become the global default originated largely in one legal and financial tradition, spread through the model documents of industry associations, and then travelled with capital. Its main components can be listed with reasonable precision. Legal form and domicile. The venture is expected to be a limited liability company with a clean ordinary share base for founders, capable of issuing multiple share classes. Very often the venture is expected to establish a #holding_company in a jurisdiction that international investors recognize, with the operating entity in the home country becoming a subsidiary. The practice of restructuring a local company under a foreign parent, commonly called a #corporate_flip or redomiciliation, is the sharpest single example of coercive isomorphism in the field. Share structure. Investors receive #convertible_preferred_shares rather than ordinary shares. The preference typically includes a liquidation preference at one times the investment amount, a right to convert into ordinary shares, and protection against dilution in a lower priced future round. Board composition. The board is small, usually three to five members at early stages. It typically contains at least one founder director, at least one investor director appointed by the lead investor, and increasingly at least one #independent_director whose appointment requires the agreement of both sides. The independent seat is a striking case. It is imported directly from listed company governance codes into firms that have no public shareholders at all. Reserved matters. A schedule of #protective_provisions lists decisions that cannot be taken without investor consent, including issuing new shares, changing the constitution, selling the company, taking on debt above a threshold, changing the business, and approving budgets or founder compensation. Founder commitments. Founders accept #reverse_vesting on their own shares, non compete and non solicitation clauses, and full assignment of intellectual property to the company. The venture is expected to hold clean title to all code, brands, and inventions. Employee incentives. An option pool of a customary size is created, with standard vesting and a defined exercise process, so that the venture can compete for talent in a way the investor considers normal. Reporting. Monthly or quarterly management accounts, a defined set of operating metrics, an annual budget, and an annual audit or independent review, delivered in a format that the investor can consolidate across a portfolio. Dispute and exit machinery. Drag along and tag along rights, rights of first refusal, pre emption rights, and an agreed dispute resolution forum, often arbitration in a neutral seat. 3.3 Standard documents as institutional carriers The diffusion of this template is not accidental. It is embedded in artifacts. Industry associations in several regions publish model financing documents that are used, with light adaptation, in thousands of transactions. Simple instruments for early rounds, including convertible notes and the various forms of simple agreement for future equity, are circulated as free downloadable templates and used almost verbatim across dozens of countries. Accelerators publish standard terms and require them as a condition of admission. These artifacts are what institutional scholars call carriers. They embody rules in a portable object. A founder in a market with no venture history can download, in a few minutes, the entire governance apparatus of a distant capital market. The cost of adopting the global template is close to zero, and the cost of designing a local alternative is very high. That asymmetry alone would produce convergence even without any deliberate coercion. 3.4 Gatekeepers and the certification chain Between the founder and the capital there is a chain of #gatekeepers. Accelerators screen and train. Angel networks and syndicates aggregate. Lead investors set terms. Law firms draft. Auditors verify. Data platforms record. Later stage funds and acquirers inspect. Each link in the chain has an interest in standardization, because each link must be able to hand the venture to the next link without friction. A venture with an unusual structure imposes cost on every subsequent participant. This creates a powerful ratchet. Even an investor who personally believes a clause is unnecessary will insist on it, because a future buyer or a future fund will ask for it, and cleaning it up later is expensive. Governance becomes standardized not because any single actor demands it, but because everyone anticipates that someone else will. 4. Conceptual Framework and Propositions 4.1 The framework in outline The framework proposed here has four elements arranged in sequence. First, there are #isomorphic_pressures generated by the field and transmitted through capital. Second, there are #transmission_mechanisms, which are the concrete channels through which pressure becomes structure, including term sheets, due diligence checklists, accelerator curricula, professional advisers, peer imitation, and investor reporting requirements. Third, there are #governance_outcomes, which include structural convergence and, to a lesser and uneven degree, practice convergence. Fourth, there are #consequences, both positive and negative, for the venture and for the ecosystem, moderated by the venture stage, sector, capital dependence, and institutional distance from the investor home context. The framework treats the founder as an active agent rather than a passive recipient. Founders can comply fully, comply symbolically, negotiate selectively, or exit the field by choosing alternative funding sources such as revenue based finance, grants, family capital, or bootstrapping. Their choice depends on bargaining power, which depends in turn on traction, competition among investors for the deal, and the availability of local alternatives. 4.2 Propositions Proposition 1. The strength of #coercive_isomorphism in startup governance increases with the venture dependence on external capital and decreases with the number of credible funding alternatives available to it. Ventures in thin capital markets therefore adopt the global template more completely and negotiate less than ventures in deep capital markets, even when their local conditions differ more sharply from the assumptions built into that template. Proposition 2. Governance convergence proceeds faster in elements that are legible to outsiders, such as legal domicile, share class structure, cap table format, and reporting cadence, than in elements that are internal and hard to verify, such as the quality of board deliberation, the true independence of the independent director, and the seriousness of risk oversight. Proposition 3. #Mimetic_isomorphism is strongest at the earliest stages, where uncertainty is highest and where founders have the least information about what is normal. Early stage founders copy the structures of visible successful peers and of accelerator alumni, often without understanding the trade offs embedded in those structures. Proposition 4. #Normative_isomorphism operates mainly through the professional service layer. The entry of a corporate lawyer, a professional chief financial officer, an audit firm, or an experienced independent director into a venture predicts a step change in the adoption of standardized governance, independent of any new investor demand. Proposition 5. The greater the #institutional_distance between the investor home context and the venture operating context, the greater the likelihood of #decoupling. Structures are adopted to satisfy the distant audience, while operational decisions continue to follow local logics of family, trust, informal networks, and personal relationships. Proposition 6. #Ceremonial_adoption is more likely when a governance element imposes ongoing cost with no visible near term benefit, such as formal risk registers, whistleblowing channels, or environmental policies in a twelve person company, and less likely when the element protects the founder's own interests, such as intellectual property assignment or a clean cap table. Proposition 7. Structural convergence improves access to follow on capital and to cross border exits, and therefore raises the expected value of the venture to its shareholders, but it does not reliably improve operational performance in the short run and may impose real costs on speed and experimentation. Proposition 8. At the ecosystem level, sustained isomorphic pressure narrows the range of organizational forms that are considered fundable. This reduces the diversity of business models that can attract capital and may systematically disadvantage models that are locally appropriate but globally unfamiliar, including cooperative, community owned, family embedded, and deliberately slow growth models. 5. Methodology 5.1 Research design This study is a conceptual and integrative review. It does not generate new primary data. Its purpose is to build a coherent theoretical account of a phenomenon that is widely observed in practice but scattered across separate literatures. Integrative review is appropriate when a topic sits at the intersection of several fields, when the empirical work is fragmented, and when the main need is synthesis and framework building rather than hypothesis testing. Three literatures are brought together. The first is #organization_theory, especially neo institutional work on isomorphism, legitimacy, and decoupling. The second is #entrepreneurial_finance, especially work on venture capital contracting, staged financing, board formation, and investor value added. The third is #comparative_corporate_governance, especially work on the convergence and divergence of governance systems across countries and on governance in emerging markets. 5.2 Literature identification The review drew on peer reviewed journal articles and scholarly books, prioritizing work published within the last five years, supplemented by a small number of foundational texts whose theoretical contribution is unavoidable and cannot be replaced by recent citations. Search terms combined the theoretical vocabulary, including institutional isomorphism, legitimacy, decoupling, and institutional distance, with the empirical vocabulary, including venture capital, startup, board of directors, term sheet, and emerging markets. Inclusion required that a source address at least one of the following: the mechanisms by which governance practices diffuse, the content of venture financing contracts, the behavior of boards in private and early stage firms, or the interaction between global capital and local institutional context. Sources dealing only with listed firm governance in a single mature market, with no diffusion or comparative element, were excluded unless they supplied a necessary theoretical anchor. 5.3 Analytical procedure Sources were coded along three dimensions. The first dimension is the pressure type, which classifies the mechanism described as coercive, mimetic, or normative, allowing for overlap. The second dimension is the governance element affected, such as board, share class, reporting, or domicile. The third dimension is the reported consequence, distinguishing between resource outcomes such as funding and exit, and behavioral outcomes such as decision speed, oversight quality, and founder discretion. The coded material was then arranged into the framework presented in section four. Where the literature is thin, most notably on the internal behavior of early stage boards, which is difficult to observe because private company meetings are not public, the paper marks the gap explicitly rather than filling it with speculation presented as evidence. 5.4 Illustrative material The paper uses widely documented and publicly discussed industry practices as illustration. These include the use of model financing documents by industry associations, the standard structure of accelerator investment terms, the common practice of restructuring ventures from emerging markets under holding companies in investor friendly jurisdictions, and the growing inclusion of environmental, social, and governance conditions in the investment agreements of development finance institutions and impact funds. These practices are treated as illustrations of theoretical mechanisms, not as tested findings. 5.5 Quality, validity, and limitations of the method A conceptual review cannot establish causality. It can establish plausibility, coherence, and testable propositions. Three limitations deserve to be stated openly. First, there is a #visibility_bias. Governance structures that are recorded in documents are easy to observe, while governance behavior is not. The literature therefore over reports form and under reports practice, and any synthesis inherits that imbalance. Second, there is a #survivorship_bias. Most of what is known about startup governance comes from ventures that succeeded in raising institutional capital. Ventures that refused the template, or that were refused by it, largely disappear from the record, so the true cost of conformity is systematically underestimated. Third, there is a #geographic_bias. A disproportionate share of the empirical evidence comes from a small number of mature markets, which makes it difficult to know how far conclusions travel. The paper attempts to correct for this by treating capital importing regions as a central case rather than a footnote, but it cannot fully overcome the underlying imbalance in the evidence base. 6. Analysis and Findings 6.1 Coercive pressure: the term sheet as private legislation The single most powerful instrument of governance standardization in the startup world is not a statute. It is the #term_sheet. A term sheet is a short document, often no more than a few pages, that sets out the main commercial and control terms of an investment before the long form agreements are drafted. Although usually described as non binding on price, it is treated in practice as close to final, and it determines the governance architecture of the venture for years. Coercion here rarely takes the form of a threat. It takes the form of a description. The investor says that a clause is market standard. That phrase performs an enormous amount of institutional work. It converts a contestable preference into an objective fact about the world. It shifts the burden of proof onto the founder, who must now explain why this venture should be an exception. And it invokes the authority of an invisible collective, the market, which no individual can argue with. Several channels of coercive pressure can be distinguished. Investor conditions precedent. Before money moves, the venture must complete a list of conditions. It must clean its cap table, formalize employment contracts, assign intellectual property, register trademarks, adopt new articles, appoint a board, and sometimes restructure its group. Each condition is a governance transplant. Limited partner pressure. Funds are themselves accountable to their own investors, including pension funds, endowments, family offices, sovereign vehicles, and development finance institutions. Those limited partners impose their own requirements on the fund, covering anti money laundering checks, sanctions screening, environmental and social risk management, exclusion lists, and impact reporting. The fund cannot absorb those requirements alone. It passes them down the chain into portfolio company agreements. This is a cascade, and it explains why a small company with fifteen employees can find itself adopting a policy framework designed for large industrial firms. Regulatory spillover. Rules that formally apply to listed firms or to financial institutions leak into private markets through investor expectations. Sustainability disclosure regimes are the clearest current example. A startup is not obliged by law to report on its emissions or its diversity, but if its investor must report on the aggregate profile of its portfolio, the startup will be asked to supply the numbers, and asking soon becomes requiring. Exit anticipation. The most under discussed source of coercion is the future. Governance is built today for a transaction that may happen in five years. Acquirers conduct legal due diligence, and any irregularity discovered at that point reduces price or kills the deal. Investors therefore front load conformity. The venture is governed as if it were already being sold. The cumulative effect is that a startup is subject to a body of private legislation that is more detailed, more enforceable, and more consequential for its daily life than the company law of its home country. 6.2 Mimetic pressure: copying under uncertainty Founders are not merely coerced. Very often they arrive at the negotiating table already wanting the standard structure, because they have absorbed it from the environment. The startup world produces an extraordinary volume of public advice. Blog posts, podcasts, founder handbooks, accelerator curricula, and open sourced legal templates all describe what a normal company looks like. Founders read this material long before they meet an investor. By the time they raise, they have internalized the template as common sense. Uncertainty amplifies imitation. A first time founder has no way to evaluate whether a one times non participating liquidation preference is fair, whether a four year vesting schedule is right for a two person team, or whether an option pool of a given size is generous or stingy. Faced with that uncertainty, the rational move is to do what successful ventures did. Imitation is a heuristic for survival. Three mimetic channels stand out. The peer channel. Founders talk to other founders, especially those one or two steps ahead. Structures spread horizontally through these conversations faster than through any formal mechanism. The accelerator channel. Accelerators compress the transmission of the template into a few months. They teach the vocabulary, supply the documents, introduce the lawyers, and stage a demo day where conformity is rewarded with attention. A cohort emerges with near identical structures. The success story channel. The most visible ventures in a region become models. Their structures are studied and copied, including the decision to redomicile abroad. This produces a path dependence in which the first successful venture in an ecosystem sets a precedent that later ventures follow, whether or not it suits them. An important consequence follows. Mimetic pressure means that convergence would occur even if investors were passive. The template is pulled as much as it is pushed. 6.3 Normative pressure: the professional layer The third mechanism is quieter and, over time, probably the most durable. It works through people rather than documents. As a venture grows, it hires and appoints professionals. It retains a corporate law firm. It appoints an auditor. It hires a finance leader who has previously worked in a larger company or in a bank. It recruits an independent director with public company experience. It engages consultants for compliance, data protection, and human resources. Each of these professionals brings a template that was formed elsewhere. The lawyer drafts the documents that the lawyer has always drafted. The auditor expects the controls that the auditor has always expected. The finance leader builds the reporting pack that the finance leader built in the previous firm. None of them is coercing anyone. They are simply doing their jobs according to the norms of their profession, and those norms are international. Business education plays a similar role. Governance is taught in a fairly uniform way in graduate management programs across the world. The board is presented as a monitoring body. Independence is presented as a virtue. Separation of chair and chief executive is presented as good practice. Students absorb this as a universal truth rather than as one particular tradition, and they take it into ventures. Normative pressure has a distinctive property. It survives changes in ownership. Investors come and go, but once a professionalized management layer exists, the template reproduces itself internally. 6.4 The outcomes: what convergence delivers It would be a caricature to present standardization as pure imposition without value. The benefits are real and should be stated fairly. Access to capital. A venture that presents a familiar structure is easier to fund. Due diligence is faster and cheaper. Legal fees fall. Time to close shortens. In a competitive market, speed is itself an advantage. Reduced transaction costs. A shared template means that negotiations focus on the few terms that matter rather than reinventing the entire contract. This is a genuine public good for the ecosystem. Protection for minority holders and employees. Standard structures protect people who would otherwise be vulnerable. Employees with options benefit from a clean cap table. Small investors benefit from pre emption and information rights. Founders themselves benefit from vesting when a co founder leaves early. Credible exits. Cross border acquisitions and public offerings require documented, defensible governance. A venture that has been governed properly from the start can be sold; one that has not may be unsellable at any price. Discipline and survival. Regular reporting forces founders to look at the numbers. Board meetings force reflection. Budget approval forces planning. For many young companies, this externally imposed rhythm is the first serious management system they have ever had. 6.5 The costs: what convergence takes away The costs are equally real and much less often counted. Founder discretion. Protective provisions can make a founder unable to act at the speed the business requires. A pivot that requires changing the business may need investor consent, and consent takes time. Ceremonial governance. Where the structure is adopted only for external audiences, the venture pays for a governance apparatus that produces no oversight. Board packs are prepared to impress. Meetings become performances. Real problems are discussed elsewhere. This is the classic pattern of #decoupling, and it is arguably worse than having no board, because it creates a false sense of assurance. Misfit with local context. A governance structure designed for a legal environment with strong courts, deep disclosure, and liquid markets can perform poorly in an environment with slow courts, informal enforcement, and family ownership. Formal minority protections mean little if enforcement takes eight years. Meanwhile the local mechanisms that actually work, such as reputation within a community, are treated as unprofessional and pushed aside. Capital flight of legal value. The corporate flip is efficient for the investor but it moves the holding entity, and therefore the tax base, the legal jurisdiction, the arbitration forum, and often the eventual listing, out of the country where the value was created. Local ecosystems build companies and then export their legal skeletons. Homogenization of ambition. The template embeds an implicit business model. It assumes rapid growth, multiple financing rounds, dilution, and a liquidity event within a defined horizon. Ventures that would prefer to grow slowly, distribute profits, or remain family held do not merely lack funding, they become unintelligible to the field. This is the deepest cost, because it operates at the level of what can be imagined rather than what can be financed. Cost burden. Compliance is not free. Audit fees, legal fees, secretarial services, data protection tooling, and reporting time consume a meaningful share of an early stage budget. For a company with a small team, the governance overhead can equal the cost of an engineer. 6.6 Variation across regions and stages Convergence is strong but it is not uniform, and the exceptions are theoretically informative. Stage. At pre seed, structures are light and often improvised. Convergence intensifies sharply at the first priced round, when preferred shares and a board appear, and again at the growth stage, when audit, committees, and independent directors appear. The template is therefore not adopted at once. It is layered on in a predictable sequence, each layer added when a new class of external audience appears. Sector. Ventures in regulated sectors, especially financial services and health, converge earlier and more completely, because they face regulatory as well as investor pressure, and because their customers demand assurance. Capital source. Ventures funded by development finance institutions or impact funds adopt environmental and social governance apparatus that purely commercial ventures do not. Ventures funded by corporate venture arms adopt reporting structures that mirror the parent corporation. Ventures funded by local family capital converge least of all, which is exactly what the theory predicts, since local family capital does not answer to a distant limited partner. Institutional context. In markets where the local legal system does not offer the instruments the template assumes, such as flexible share classes, enforceable option schemes, or reliable share registries, ventures do not abandon the template. They relocate to a jurisdiction that supplies it. This is a critical observation. When local institutions cannot support the global template, the template does not adapt to the locality. The venture leaves. The template wins by exit rather than by adjustment. Founder bargaining power. Repeat founders with a prior successful exit negotiate materially better terms and retain more control. Convergence, in other words, is stratified. The template is enforced most strictly on those with the least power, which are precisely first time founders in capital scarce markets. 6.7 Responses: how founders and ecosystems push back Founders are not helpless. Four response strategies can be identified. Selective compliance. Founders accept the elements that are cheap or self serving, such as intellectual property assignment, and resist or postpone those that are costly, such as full audit. Symbolic compliance. The venture builds the structure and reports on it while continuing to operate informally. This buys legitimacy at low cost, but it accumulates risk, because a decoupled structure will be exposed during due diligence at the next round. Genuine adaptation. Some ventures use the imposed structure as a real management tool, converting an external requirement into an internal capability. This is the outcome the investor hopes for, and it appears more often in ventures whose founders have prior corporate experience. Exit from the field. A growing group of founders declines institutional venture capital altogether and funds growth through revenue, debt, grants, or community and customer financing. These ventures escape the template, but they also escape the visibility, the network, and the scale that come with it. At the ecosystem level, responses include the creation of local legal vehicles designed to be compatible with international expectations, the reform of company law to permit multiple share classes and enforceable option schemes, the development of regional startup acts, and the establishment of domestic funds intended to keep governance and value at home. These reforms illustrate an important dynamic. Rather than resisting isomorphism, many states have chosen to accelerate it, by rewriting their own institutions so that local ventures no longer need to leave. 7. Discussion 7.1 The term sheet as a transnational institution The central theoretical claim of this paper is that the venture financing contract functions as a #transnational_institution. It performs the work that, in the public company world, is performed by company law, listing rules, and national governance codes. It defines who holds authority, which decisions require consent, how information flows, how conflicts are resolved, and how the organization may be dissolved or sold. It is enforced not by a regulator but by the threat of capital withdrawal and by the anticipation of future scrutiny. This has an important implication for how governance research is framed. Comparative corporate governance has traditionally been organized around countries. Scholars ask whether the governance systems of different national economies are converging. In the venture domain, the country is often the wrong unit of analysis. Two startups in the same country, one funded by international capital and one funded by local family capital, may operate under governance systems that are further apart than two internationally funded startups on different continents. Governance in this field is organized by #capital_network rather than by nation. 7.2 Why isomorphism is not the same as improvement A recurring temptation in this literature is to treat convergence as progress, on the assumption that the diffusing model is the best one. That assumption should be resisted for three reasons. First, the diffusing template was designed for a specific environment characterized by deep capital markets, rapid technology cycles, strong minority protection, and frequent liquidity events. Where those conditions are absent, the same structure may produce cost without benefit. A liquidation preference in a market with almost no exits is a claim on a distribution that will never happen, but it still shapes founder incentives today. Second, adoption is often ceremonial. If the form is present but the substance is absent, the measured convergence in the data is an illusion. Studies that count governance structures without observing behavior may be measuring nothing more than the reach of a set of templates. Third, isomorphism has a cost that does not appear in any individual firm's accounts. It is borne by the ecosystem in the form of reduced variety. Variety is the raw material of adaptation. An ecosystem in which every fundable venture has the same governance structure, the same growth expectations, and the same exit horizon is efficient in the short term and fragile in the long term, because it has fewer alternative models to draw on when conditions change. 7.3 Reconciling agency and institutional explanations The two explanations offered at the start of this paper can now be reconciled rather than opposed. Agency theory explains the origin of each instrument. Institutional theory explains its diffusion, its persistence, and its reach into contexts where the original agency problem is absent or mild. A useful way to express this is to distinguish between the #functional_core of the template and its #institutional_shell. The functional core consists of the instruments that solve a live problem in the specific venture, which usually includes vesting, IP assignment, clean records, and some form of monitoring. The institutional shell consists of everything adopted because it is expected, which may include an independent director in a six person company, a formal audit committee with no audit worth discussing, or an environmental policy in a purely digital business. The core earns its keep. The shell buys legitimacy. Most governance debates in practice are arguments about where the boundary lies, conducted by parties who do not realize that this is the argument they are having. 7.4 Boundary conditions The argument advanced here has limits that should be acknowledged. It applies most strongly to ventures that seek equity finance from professional intermediaries and that aspire to cross border growth. It applies weakly to small businesses, to family firms that do not seek external equity, and to ventures financed by revenue or by the public sector. It also assumes a reasonably competitive investor market at the top of the chain. In markets where a single state backed investor dominates, the pressures may be coercive without being isomorphic in the global sense, because the template imposed is local rather than international. Such settings are theoretically interesting precisely because they break the usual pattern, and they deserve dedicated study. 7.5 A note on the role of the state States are not bystanders. In many capital importing regions, governments have responded to the offshore drift of their most promising ventures by reforming domestic law, creating special economic or financial zones with familiar legal systems, permitting flexible share classes, and clarifying the tax treatment of employee options. In effect, states are competing to supply the institutional environment that global investors expect. This is isomorphism operating at the level of the nation rather than the firm. The pressure travels upward. Investors shape ventures, ventures shape the demands placed on the state, and the state adjusts its institutions so that its ventures can remain at home. Understanding this upward channel is one of the most promising directions for future work, because it links a micro level contracting practice to macro level institutional change. 8. Implications 8.1 For founders Founders should read the template as a political document, not a technical one. Each clause allocates power. The practical advice that follows from this analysis is concrete. Understand which terms are truly standard and which are merely presented as standard. Economic terms and control terms are different, and control terms are usually where the long term consequences live. Negotiate the list of reserved matters carefully, because that list determines how fast the company can move. Treat board composition as a design problem rather than a formality, and prefer a director who understands the business over a director who decorates the letterhead. Do not accept a structure that the venture cannot actually operate, because a decoupled structure will be discovered later, and it will be discovered at the worst possible moment. Above all, founders should recognize that conformity in governance can be traded for freedom in strategy. Fighting the template on ordinary points wastes credibility that could be spent on the points that matter. 8.2 For investors Investors should ask whether each imported requirement produces oversight or only produces paper. A twelve person company that spends a week each quarter preparing a board pack for an audience that reads it in ten minutes is being harmed rather than governed. Investors should also consider proportionality by stage and by context. Requirements designed for a growth stage firm can be scaled down for a seed stage firm without abandoning discipline. And where the local institutional context supplies working substitutes for formal controls, such as strong professional reputation networks, investors might treat those substitutes as complements rather than as evidence of backwardness. 8.3 For policymakers and regulators The policy implication is uncomfortable but clear. If a country cannot supply the legal instruments that global capital expects, its best ventures will leave, and the country will retain the operations while exporting the ownership. Reform of company law to permit multiple share classes, enforceable employee option schemes, reliable share registries, and efficient dispute resolution is therefore not a technical matter. It is industrial policy. At the same time, policymakers should be careful about the assumption that the imported model is the only legitimate one. There is room for public funding instruments, guarantee schemes, and legal forms that support ventures pursuing steady growth, community ownership, or profit distribution rather than the exit driven model. Without such instruments, an entire class of economically valuable organizations remains unfundable. 8.4 For business educators and students Governance is usually taught as a set of best practices. This paper suggests it should also be taught as a set of social processes. Students should learn to ask where a practice came from, whose interests it serves, what problem it originally solved, and whether that problem exists in the case before them. For students who intend to found or join startups, the practical lesson is that the documents signed at the first financing round will shape their working life for years, and that these documents are negotiable in ways that are rarely advertised. 9. Limitations and Future Research 9.1 Limitations This paper is conceptual. It proposes rather than tests. Its propositions are derived from theory and from documented practice, and they remain to be examined against systematic data. The evidence base it synthesizes is uneven across regions and skewed toward ventures that succeeded in attracting institutional capital. The paper also treats the global template as relatively coherent, when in reality there are meaningful differences between the contracting traditions of different regions, and those differences deserve finer analysis than is offered here. 9.2 A research agenda Six directions are proposed. Measure the gap between form and practice. The most valuable empirical contribution would be to study what early stage boards actually do, using meeting observation, director interviews, or document analysis, and to compare that behavior with the formal structure. This would allow the field to distinguish real governance from ceremonial governance. Track redomiciliation. The corporate flip is a measurable event with measurable consequences. Studies could examine which ventures flip, when, under what investor pressure, and with what effect on taxation, employment, and eventual exit location. Examine stratification. If the template is enforced more strictly on less powerful founders, then governance terms become a channel through which existing inequalities are reproduced. Comparative analysis of contract terms across founder characteristics and regions would test this directly. Study the cost of compliance. Almost nothing is known about how much of an early stage budget is consumed by governance and reporting, or about the point at which that cost begins to reduce the venture's chance of survival. Investigate reverse diffusion. Are there any governance innovations that travel from capital importing regions back to the centers of the venture field? If not, that absence is itself an important finding about how power operates in the field. Assess sustainability related pressure. The recent extension of environmental and social requirements into private markets is an isomorphic process happening in real time. It offers an unusual opportunity to observe institutionalization as it occurs, rather than reconstructing it after the fact. 10. Conclusion Startups are celebrated for their difference and governed by their sameness. This paper has argued that the sameness is not accidental, and that it is not primarily produced by law. It is produced by capital. Global investors, acting through term sheets, due diligence checklists, reporting requirements, and the anticipation of a future exit, transmit a coherent governance template across borders. That transmission is reinforced by imitation among uncertain founders and by the international norms of the lawyers, accountants, and directors who staff the venture as it matures. The result is a field in which organizations that operate in radically different social, legal, and economic environments carry almost identical internal architectures. That architecture delivers genuine value. It reduces the cost of transacting, it protects vulnerable parties, it imposes useful discipline, and it makes cross border investment and exit possible at all. But it also carries costs that are rarely recorded. It transfers authority from founders to financiers, it exports legal value from the countries where economic value is created, it can produce ceremony in place of oversight, and it quietly narrows the definition of what a fundable company can be. The most accurate description of #startup_governance is therefore not compliance and it is not resistance. It is negotiation. Founders trade conformity in structure for freedom in strategy, and they trade legitimacy for autonomy at a price they usually do not fully understand at the time of signing. Understanding that trade, making it explicit, and giving founders and policymakers the vocabulary to argue about it, is the practical purpose of institutional analysis in this field. The task for scholarship is now to look inside the structures rather than merely to count them. Whether the global governance template produces real oversight or elaborate theatre remains, after four decades of institutional research, an open and urgent question. References Aguilera, R. V., Aragon-Correa, J. A., Marano, V., and Tashman, P. A. (2021). The corporate governance of environmental sustainability: A review and proposal for more integrated research. Journal of Management, 47(6), 1468-1497. Ararat, M., Claessens, S., and Yurtoglu, B. B. (2021). Corporate governance in emerging markets: A selective review and an agenda for future research. Emerging Markets Review, 48, 100767. Bellavitis, C., Fisch, C., and McNaughton, R. B. (2022). COVID-19 and the global venture capital landscape. Small Business Economics, 59(3), 781-805. Bromley, P., and Powell, W. W. (2012). From smoke and mirrors to walking the talk: Decoupling in the contemporary world. Academy of Management Annals, 6(1), 483-530. Chliova, M., Mair, J., and Vernis, A. (2020). Persistent category ambiguity: The case of social entrepreneurship. Organization Studies, 41(7), 1019-1042. Cumming, D., Meoli, M., and Vismara, S. (2021). Does equity crowdfunding democratize entrepreneurial finance? Small Business Economics, 56(2), 533-552. DiMaggio, P. J., and Powell, W. W. (1983). The iron cage revisited: Institutional isomorphism and collective rationality in organizational fields. American Sociological Review, 48(2), 147-160. Dua, G. K. (2022). Analysis on institutional theory, mimetic isomorphism, and firm performance. International Journal of Health Sciences, 6(S3), 5821-5832. Ewens, M., and Farre-Mensa, J. (2022). Private or public equity? The evolving entrepreneurial finance landscape. Annual Review of Financial Economics, 14, 271-293. Filatotchev, I., Aguilera, R. V., and Wright, M. (2020). From governance of innovation to innovations in governance. Academy of Management Perspectives, 34(2), 173-181. Greenwood, R., Oliver, C., Lawrence, T. B., and Meyer, R. E. (Eds.). (2017). The SAGE Handbook of Organizational Institutionalism (2nd ed.). Sage Publications. Kaplan, S. N., and Stromberg, P. (2003). Financial contracting theory meets the real world: An empirical analysis of venture capital contracts. Review of Economic Studies, 70(2), 281-315. Kostova, T., Beugelsdijk, S., Scott, W. R., Kunst, V. E., Chua, C. H., and van Essen, M. (2020). The construct of institutional distance through the lens of different institutional perspectives: Review, analysis, and recommendations. Journal of International Business Studies, 51(4), 467-497. Lerner, J., and Nanda, R. (2020). Venture capital's role in financing innovation: What we know and how much we still need to learn. Journal of Economic Perspectives, 34(3), 237-261. Meyer, J. W., and Rowan, B. (1977). Institutionalized organizations: Formal structure as myth and ceremony. American Journal of Sociology, 83(2), 340-363. Nicholas, T. (2019). VC: An American History. Harvard University Press. Pfeffer, J., and Salancik, G. R. (2003). The External Control of Organizations: A Resource Dependence Perspective. Stanford Business Books. Scott, W. R. (2014). Institutions and Organizations: Ideas, Interests, and Identities (4th ed.). Sage Publications. Singhania, M., and Saini, N. (2023). Institutional framework of ESG disclosures: Comparative analysis of developed and developing countries. Journal of Sustainable Finance and Investment, 13(1), 516-559. Stinchcombe, A. L. (1965). Social structure and organizations. In J. G. March (Ed.), Handbook of Organizations (pp. 142-193). Rand McNally. Suchman, M. C. (1995). Managing legitimacy: Strategic and institutional approaches. Academy of Management Review, 20(3), 571-610. Tricker, B. (2019). Corporate Governance: Principles, Policies, and Practices (4th ed.). Oxford University Press. Zhao, E. Y., Fisher, G., Lounsbury, M., and Miller, D. (2017). Optimal distinctiveness: Broadening the interface between institutional theory and strategic management. Strategic Management Journal, 38(1), 93-113. #institutional_isomorphism #startup_governance #venture_capital #corporate_governance #entrepreneurial_finance #legitimacy_theory #coercive_pressure #mimetic_pressure #normative_pressure #decoupling_in_organizations #emerging_market_startups #term_sheet_negotiation #founder_control #global_investors #governance_convergence