Generative AI for Business Model Innovation

Introduction

The arrival of generative AI has triggered a wave of enthusiasm about new products, new capabilities, and new customer experiences. Yet the most consequential impact of generative AI lies elsewhere: in how it changes the way companies create, deliver, and capture value. The real story is business model innovation.

Business model innovation (BMI) is the process of reconfiguring the fundamental logic of a firm (who it serves, what it offers, how it delivers, and how it earns revenue) in ways that create new value for customers and competitive advantage for the firm. Generative AI does more than improve existing business models. It makes entirely new configurations viable, models that would have been economically impossible, operationally impractical, or simply unthinkable five years ago.

Consider a traditional cookbook publisher with 15,000 professionally tested recipes accumulated over four decades. Under its existing business model, this company sells physical books through retailers. Its revenue depends on bestseller cycles, its customer relationship ends at the point of sale, and it has no idea whether anyone actually cooks from its recipes. Now imagine the same company deploying a generative AI personalization engine that turns that recipe database into millions of unique weekly meal plans, adaptive to dietary restrictions, taste feedback, and fitness goals, delivered through a subscription app. The technology is a means, not an end. What changed is the business model: from one-time product sales to recurring subscription revenue, from mass-market publishing to mass customization, from a channel strategy built on retailers to a direct digital relationship with every customer.

This chapter provides a framework for understanding how generative AI enables business model innovation. It introduces the EDGE Framework as a strategic lens for categorizing the value that AI creates, maps eight generative AI-powered business model patterns onto the established St. Gallen Business Model Navigator, and illustrates how these concepts apply in practice.

11.1 Foundations: What Is a Business Model?

Before exploring how generative AI transforms business models, we need a working definition of what a business model is and a way to analyze one systematically.

The Business Model Canvas

The most widely adopted framework for describing business models is the Business Model Canvas (BMC), developed by Alexander Osterwalder and Yves Pigneur. The BMC decomposes any business model into nine interdependent building blocks: Customer Segments, Value Propositions, Channels, Customer Relationships, Revenue Streams, Key Resources, Key Activities, Key Partnerships, and Cost Structure.

The power of the BMC lies not in any single block but in the relationships between them. A change in Value Proposition often forces changes in Customer Segments, which in turn requires different Channels and a different Revenue model. Business model innovation, then, means reconfiguring the relationships between the blocks, not filling in the nine boxes once and framing the result.

The Four-Box Framework

Johnson, Christensen, and Kagermann offer a complementary lens through their Four-Box Framework, which groups the nine BMC blocks into four interdependent systems: the Customer Value Proposition (what job does the product do for the customer?), the Profit Formula (how does the company make money?), Key Resources (what assets are required?), and Key Processes (what operational capabilities are needed?). Their central insight is that these four boxes are structurally interdependent: change one and you typically have to change all four. This helps explain why business model innovation is so difficult. A new idea for a value proposition is not enough; the entire operating system of the firm must be redesigned to support it.

The St. Gallen Business Model Navigator

Gassmann, Frankenberger, and Csik's research at the University of St. Gallen produced a striking empirical finding: approximately 90% of all "new" business models are recombinations of 55 existing patterns. Their Business Model Navigator organizes these patterns around a "Magic Triangle" of four dimensions (Who is the customer? What do we offer? How do we deliver it? How do we make money?) and provides a systematic vocabulary for describing business model configurations.

The St. Gallen approach is particularly valuable for practitioners because it shifts the creative challenge from blank-page invention to pattern recombination. Executives do not need to imagine entirely novel models. They need to recognize which existing patterns, applied in new combinations or new contexts, can create value. Generative AI expands the range of patterns that are economically viable, and that expansion is the subject of this chapter.

11.2 The EDGE Framework: Where Does AI Create Value?

Not all AI applications create the same type of value, and not all types of value call for the same business model response. The EDGE Framework provides a strategic lens for categorizing the value that generative AI creates across four dimensions: Efficiency, Decisions, Growth, and Empowerment. Each dimension corresponds to a different strategic logic and points toward different business model configurations.

Efficiency

Efficiency is the most immediately visible dimension. AI automates processes, reduces costs, and increases throughput. When Klarna deployed a generative AI customer service agent that handled two-thirds of all customer conversations, equivalent to the workload of 700 full-time agents, it was pursuing an Efficiency play: resolution time dropped from 11 minutes to under 2 minutes. The sequel is as instructive as the headline. By 2025 Klarna's CEO publicly conceded that the cost-first automation push had degraded service quality, and the company began rehiring human agents for a hybrid model. Read as a whole, Klarna is the best short case we have on both the reality of AI efficiency gains and their limit: the constraint is not whether AI can absorb the work, but where quality quietly leaks when it does.

The Efficiency dimension is powerful but carries a strategic risk: cost reduction alone is imitable. If your competitor can deploy the same AI model and achieve the same cost savings, the advantage is temporary. The strategic question for Efficiency plays is always: does this cost reduction fund a structural change elsewhere on the canvas? Klarna's AI agent cuts costs, yes, but it also enables 24/7 multilingual support that would be prohibitively expensive with human agents, changing the Customer Relationship and Channel blocks in the process. The Efficiency gain funds a broader business model reconfiguration.

Decisions

The Decisions dimension captures value created when AI improves the quality of choices, for the firm, its employees, or its customers. This goes beyond faster analytics. Generative AI can synthesize unstructured information, surface non-obvious patterns, and generate scenario analyses that human decision-makers would miss.

Consider how Notion AI operates within its productivity suite. The AI add-on synthesizes meeting notes, extracts action items, and answers questions across a company's entire knowledge base; nobody's job disappears. The user makes better decisions because the information in front of them is better synthesized. The business model implication is the Add-On pattern: the base subscription provides the workspace, and the AI layer charges a premium for decision-quality improvement.

The Decisions dimension is strategically interesting because decision-quality improvements compound over time and are difficult for customers to attribute to a single tool. Once an organization's workflows depend on AI-augmented decision-making, switching costs become substantial, not because of data lock-in, but because of capability lock-in.

Growth

Growth captures value created when AI opens new markets, enables new customer segments, or creates entirely new revenue streams. Unlike Efficiency (which improves existing operations) and Decisions (which improves existing judgment), Growth is about expansion into territories that were previously inaccessible.

The classic Growth play in generative AI is the Razor & Blade model deployed by companies like OpenAI. ChatGPT is free for basic use (the razor). Revenue comes from Plus and Pro subscriptions and API token consumption (the blades). The free tier drives massive adoption, creating a user base of hundreds of millions that converts to paid tiers as needs deepen, while developers pay per token for API access, making every inference call a recurring consumable. The free tier is less a marketing expense than a growth engine, one that expands the total addressable market for AI capabilities.

Growth is also where Mass Customization shines. When generative AI lets a company treat every customer as a segment of one (personalized recommendations, tailored content, individualized pricing), markets that were too fragmented for a one-size-fits-all model become serviceable. A traditional publisher serving a mass market might reach millions but satisfy few deeply. An AI-powered personalization platform can serve millions of individual needs simultaneously, opening segments (corporate wellness, medical nutrition, athletic performance) that the original model could never address.

Empowerment

Empowerment is the deepest of the four dimensions and the one most often underestimated. Empowerment occurs when AI transfers capabilities to users that they previously lacked, enabling them to do things they could not do before, rather than merely accessing things they could not reach before.

The distinction between Growth and Empowerment is subtle but consequential. Growth changes what the customer can access. Empowerment changes what the customer can do. Growth expands the customer's world; Empowerment expands the customer's capability. The practical test is: if you remove the AI, what happens? With Growth, the customer loses access to a market or service, they are back where they started. With Empowerment, the customer retains at least some of the capability. They have leveled up.

Cohere's enterprise AI infrastructure illustrates the Empowerment dimension as a Layer Player. By providing specialized language AI capabilities (embeddings, retrieval-augmented generation, text generation) to organizations across finance, healthcare, legal, and technology, Cohere enables these organizations to build AI capabilities they could not develop internally. The customer consumes AI, and in the process becomes an AI-capable organization. Over time, internal teams learn to design prompts, build retrieval pipelines, and architect AI-native workflows. The capability transfer is real.

Applying EDGE to Customer Value

The EDGE Framework was initially conceived as a firm-side lens: how does AI create value for the business? But it is equally powerful when applied to the customer side, and holding both perspectives at once reveals strategic tensions that neither can surface alone.

Efficiency for the customer means AI saves them time, effort, or cost. The customer's burden of consuming the service drops. A chatbot that resolves an issue in 30 seconds instead of a 20-minute phone call delivers customer-side Efficiency even if the firm's primary motivation is cost reduction.

Decisions for the customer means AI helps them make better choices. Instead of being overwhelmed by options, the customer is guided toward the right one. Personalized financial advice, AI health risk assessments, and recommendation engines all serve the Decisions dimension on the customer side.

Growth for the customer means AI opens access to things they could not reach before. AI translation opens global content to a non-English speaker. AI tutoring gives a rural student access to personalized education. AI design tools let a small business owner create professional marketing without hiring an agency.

Empowerment for the customer means AI shifts control to them. Through self-service diagnostics, AI-powered legal document drafting, or natural-language product configuration, the customer moves from dependent on the provider to capable on their own.

The strategic tension emerges when firm-side and customer-side EDGE dimensions diverge. A company might pursue Efficiency by automating support, but the customer experiences Empowerment through self-service. A company might pursue Growth by opening a new market, but the customer experiences Decisions by receiving better recommendations. These misalignments are where the most interesting business model design choices live. The strongest business models align both sides of the EDGE, creating value for the customer in a dimension that simultaneously strengthens the firm's competitive position.

11.3 Eight Generative AI-Powered Business Model Innovation Patterns

The following eight patterns are particularly relevant to the generative AI era. Each is grounded in the St. Gallen Business Model Navigator and has been adapted to reflect the specific economics of generative AI: near-zero marginal cost of inference, token-based pricing, unprecedented personalization capabilities, and the data flywheel dynamics that characterize AI-native businesses.

Pattern 1: Razor & Blade

St. Gallen Reference: Razor and Blade (Pattern #40)

The principle is straightforward: offer the base product at low cost or free, and profit from recurring consumables or complementary services. In the generative AI context, the "razor" is typically a free tier of an AI product (ChatGPT, for example) that drives massive adoption. The "blades" are subscriptions (Plus, Pro, Team, Enterprise) and API token consumption. Every inference call becomes a recurring consumable.

This pattern works in generative AI because the marginal cost of serving a free user is low (especially at lower usage tiers), while conversion to paid tiers is driven by genuine capability needs rather than artificial paywalls. The free tier is no act of charity. It works as a funnel, expanding the addressable market while generating training signal (user interactions improve the model).

The BMC blocks most affected are Value Proposition (free AI assistant lowers adoption friction), Revenue Streams (layered subscriptions plus metered API consumption), and Customer Segments (a funnel from free individual users through developers to enterprise buyers). The dominant EDGE dimension is Growth.

Pattern 2: Performance-Based Contracting

St. Gallen Reference: Performance-based Contracting (Pattern #35)

This pattern ties pricing to measurable outcomes rather than inputs consumed. In generative AI, this means charging for results (marketing performance lift, conversion rates, cost savings delivered) rather than for seats, words generated, or API calls made.

The pattern is powerful because it aligns provider and customer incentives. If an AI marketing platform charges based on engagement lift, it has every reason to keep improving output quality. The customer bears less risk, and the provider captures more value when the AI performs well. The catch is measurement: both parties must agree on what counts as a "result" and how it is attributed, and that agreement is harder to reach than it sounds.

The BMC blocks most affected are Revenue Streams (outcome-based fees replace volume pricing), Value Proposition (guaranteed performance rather than tool access), and Key Activities (continuous model optimization becomes existential, not optional). The dominant EDGE dimension is Efficiency.

Pattern 3: Subscription + Add-On (Hybrid)

St. Gallen Reference: Add-On (Pattern #1) + Subscription (Pattern #46)

A base subscription provides core value at a competitive price. AI capabilities are layered on top as premium add-ons. Notion AI ($10/member/month on top of the base workspace subscription) is the canonical example: users adopt the workspace first, discover the AI features over time, and upgrade when they experience value.

The pattern is elegant because AI is positioned as an enhancement to a product the customer already trusts and uses, not as the product itself. That lowers adoption friction and creates natural upsell dynamics. The subscription provides recurring baseline revenue, and the add-on captures the incremental value that AI creates.

The critical design question: what does the AI add-on deliver in month two that it did not deliver in month one? If the answer is nothing, if the AI is a static feature, the add-on pricing will face churn pressure. The strongest implementations lean on the Decisions dimension of EDGE, with the AI getting better at synthesizing information and sharpening judgment as it learns the user's specific context.

Pattern 4: Self-Service

St. Gallen Reference: Self-Service (Pattern #44)

The customer performs tasks previously handled by employees, enabled by AI. Generative AI transforms this from traditional self-service (ATMs, self-checkout) into intelligent self-service where the AI handles complex, unstructured interactions that previously required skilled humans.

Klarna's AI agent demonstrates the potential: two-thirds of all customer service conversations handled by AI, resolution time reduced from 11 minutes to under 2 minutes, equivalent to the workload of 700 full-time agents. Critically, the AI resolves inquiries rather than deflecting them: it processes refunds, handles disputes, and answers complex policy questions. Two caveats keep the case honest (its full arc, including the 2025 quality correction and partial rehiring of humans, is discussed earlier in this chapter): self-service becomes an experience advantage only while quality holds, and quality only holds where it is measured. With those caveats, the pattern stands: faster, always available, and increasingly personalized service at a marginal cost near zero.

The BMC blocks affected span almost the entire canvas. Key Activities shift from human service delivery to AI orchestration. Cost Structure changes dramatically as labor costs drop. Customer Relationships become automated but paradoxically more personalized (the AI remembers every interaction). Channels collapse to an AI-first interaction layer. The dominant EDGE dimension is Efficiency, though the customer-side experience is often one of Empowerment.

Pattern 5: Mass Customization

St. Gallen Reference: Mass Customization (Pattern #30)

Deliver individually customized products or services at near mass-production cost. Generative AI is the technology that finally makes this pattern work at real scale, because AI can generate unique outputs (meal plans, styling recommendations, learning paths, marketing copy) for each customer without a matching increase in human labor.

Stitch Fix illustrates the mechanism in retail, and also its limits: AI analyzes individual style preferences, purchase history, body measurements, and fit feedback to curate unique selections for millions of customers simultaneously. The personalization engine worked; the business shrank anyway for years, a reminder that mass customization is a capability, not a business model, and cannot rescue weak demand economics on its own. What once required a human stylist per client now runs algorithmically at scale.

The strategic insight is that Mass Customization redefines Customer Segments. Instead of demographic or psychographic clusters, every customer becomes a segment of one. This opens markets that mass-market approaches cannot serve: individuals with specific dietary needs, niche style preferences, or unusual learning requirements. The dominant EDGE dimension is Growth, because personalization at scale expands the addressable market.

Pattern 6: Pay Per Use

St. Gallen Reference: Pay Per Use (Pattern #34)

Customers pay only for what they actually consume, metered precisely. In generative AI, this translates to token-based pricing, per-API-call billing, or per-inference charges. AWS Bedrock, Google Vertex AI, and Azure OpenAI Service all follow this model: a startup and a Fortune 500 company access the same foundation models, paying only for what they use.

This pattern mirrors the cloud computing revolution that preceded it. AI becomes a utility. Token-level pricing means costs scale linearly with value extracted, making generative AI accessible to companies of any size without upfront capital commitment.

The BMC implications are broad: Revenue Streams become metered consumption, Value Proposition centers on "enterprise AI without upfront investment," and Cost Structure becomes purely variable. The dominant EDGE dimension is Efficiency; precise cost alignment eliminates waste and lowers the barrier to AI adoption.

Pattern 7: Two-Sided Platform

St. Gallen Reference: Two-Sided Market (Pattern #49)

A platform connects two interdependent user groups, capturing value from facilitating transactions between them. The generative AI era has created an explosion of platform opportunities because the number of AI artifacts (models, datasets, prompts, fine-tuned variants, adapters) that can be shared, traded, and deployed has expanded enormously.

Hugging Face exemplifies this pattern: model creators (researchers, companies) upload pre-trained models, and model consumers (developers, enterprises) discover, test, and deploy them. The platform earns through enterprise hosting, private model hubs, and compute services. Network effects compound as more models attract more developers, which attracts more model creators.

The hard strategic problem for AI platforms is the chicken-and-egg question: how do you attract both sides at once? The most successful approach is to subsidize one side (typically creators, through free hosting and community recognition) to build supply, then monetize the other side (enterprises, through premium features and SLAs) once the network reaches critical mass. The dominant EDGE dimension is Growth.

Pattern 8: Layer Player

St. Gallen Reference: Layer Player (Pattern #27)

Specialize in one specific step of the value chain and serve it across multiple industries. Cohere focuses exclusively on the language AI layer (embeddings, retrieval-augmented generation, text generation) without building end-user applications. It serves this single layer across finance, healthcare, legal, and technology.

The Layer Player pattern is compelling in generative AI because the AI stack is maturing into distinct layers (infrastructure, foundation models, fine-tuning, application) with different competitive dynamics at each one. By specializing, a Layer Player achieves a depth of enterprise-grade security, regulatory compliance, and domain-specific optimization that horizontal competitors cannot match.

The BMC insight is that Key Activities narrow to one capability, perfected, while Customer Segments broaden across industries. Key Partnerships become essential because the Layer Player must integrate into others' full-stack solutions. The dominant EDGE dimension is Empowerment: the Layer Player enables other organizations to build AI capabilities they could not develop alone.

11.4 Pattern Combinations and Strategic Design

In practice, the most powerful generative AI business models combine two or more patterns rather than implementing a single one in isolation. The St. Gallen research confirms this: the majority of successful business model innovations are pattern recombinations, not single-pattern implementations.

Several combinations are particularly potent in the generative AI context.

Self-Service + Mass Customization creates a model where the customer serves themselves, but the AI makes every self-service interaction individually tailored. Picture an AI-powered nutrition platform where each user receives a unique meal plan generated from their preferences, allergies, and goals, without ever speaking to a dietitian. The customer does the work (self-service), but the output is unique to them (mass customization).

Platform + Subscription combines network effects with recurring revenue. A two-sided marketplace that also charges a monthly subscription for premium access (analytics, priority matching, exclusive listings) captures value from both the transaction flow and the ongoing relationship. This is the model that many AI talent platforms and AI-powered marketplaces are pursuing.

Razor & Blade + Layer Player creates a model where the free tier (razor) attracts developers to a specialized infrastructure layer, and API consumption (blades) generates recurring revenue at scale. The free tier builds ecosystem lock-in, and the specialization creates switching costs.

The design question for executives is not "which pattern should we choose?" but "which combination of patterns, applied to our specific assets and market position, creates a configuration that is difficult to replicate?" The answer almost always involves building on existing data, customer relationships, or brand trust: legacy assets that pure-play AI startups cannot easily acquire.

11.5 From Theory to Practice: The Role of Legacy Assets

A misconception I hear from executives almost weekly is that incumbents are disadvantaged in the generative AI era, that agile startups will disrupt them with superior AI capabilities. The pattern analysis suggests the opposite. In each of the eight patterns, the most defensible implementations combine generative AI with an asset the incumbent already possesses and a new entrant cannot easily replicate.

A cookbook publisher's 15,000 professionally tested, photographed recipes are the moat. A recruitment agency's database of placement outcomes (not just candidate profiles, but data on which hires succeeded and which failed) is training data that job boards cannot match. A hotel chain's 80 physical properties in prime urban locations are irreplaceable no matter how good a booking AI becomes. A driving school network's two million lessons of dashcam footage is a training dataset for computer vision that no rival can reproduce.

The strategic lesson: generative AI does not render legacy assets obsolete. It makes them more valuable, provided the business model is reconfigured to exploit them in a new way. The recipes are worthless as static cookbook content in a declining print market. They are invaluable as the training corpus for an AI personalization engine in a subscription model. The asset did not change. The business model did.

This is the core of business model innovation: seeing existing resources through the lens of new configurations. The EDGE Framework helps executives decide what type of value their AI initiative should create. The St. Gallen patterns supply a vocabulary for describing the configuration. And the BMC provides the canvas on which the reconfiguration is designed, tested, and communicated.

11.6 The Economics of Generative AI Business Models

Generative AI introduces several economic dynamics that business model designers must understand.

Token economics and marginal cost. The cost of serving an additional user or generating an additional output in generative AI is not zero, but it is very low and declining rapidly. Inference costs have dropped by an order of magnitude in the past two years and are expected to continue falling. This has profound implications for pricing strategy: models that charge per token (Pay Per Use) will face margin pressure as costs decline, while models that charge for outcomes (Performance-Based Contracting) or subscriptions capture more of the value created.

The data flywheel. Generative AI models improve with use. More users generate more data, which improves the model, which attracts more users. This creates winner-take-most dynamics in many AI markets. The business model implication is that customer acquisition is a product improvement play as much as a revenue play, which justifies aggressive pricing on the razor (free tier) to accelerate the flywheel.

Inference cost as a variable cost floor. Unlike traditional software (where the marginal cost of serving an additional user is near zero), generative AI has a real marginal cost: every API call, every generated response, every personalized recommendation costs something to compute. This means that purely free models are unsustainable without a monetization path, and that the Cost Structure block of the BMC must explicitly account for inference costs that scale with usage.

Compounding personalization. AI-powered business models that learn from individual user behavior create increasing returns to the customer relationship. The longer a customer uses an AI meal planner, the better it knows their preferences. The longer a company uses an AI recruitment platform, the better it predicts hiring outcomes. This creates natural retention and switching costs, not through lock-in, but through accumulated value.

11.7 Lessons from the AI Valley: Capital, Concentration, and Survival

The business model patterns in this chapter assume that a company gets to choose its position. The journalist Gary Rivlin's book AI Valley (2025) is a useful corrective, because it documents what happens when the economics of the underlying technology choose for you. Rivlin spent more than a year embedded inside Inflection AI, the startup Reid Hoffman and Mustafa Suleyman founded in 2022 to build Pi, a chatbot designed for emotional intelligence rather than raw capability. Inflection raised 1.5 billion dollars from investors including Microsoft and Nvidia. It was not enough. By Rivlin's account the company needed roughly another 2 billion dollars just to operate for twelve months, against competitors holding tens of billions in reserve.

The ending has become a template. In March 2024, Microsoft hired Suleyman and most of Inflection's staff and paid a licensing fee widely reported at around 650 million dollars, in what the industry now calls a reverse acqui-hire. Suleyman became CEO of Microsoft AI. Inflection, technically still alive, ceased to matter. One founder quoted in the book predicted that within five to ten years none of the startups in the consumer AI space would survive as independent companies. For strategists, the lesson is not that startups are doomed. It is that any business model built directly on frontier model development requires capital that only a handful of firms possess. Everyone else must build one layer up, on top of models they do not own, in exactly the way this chapter's patterns describe.

Rivlin's reporting also illustrates why incumbents hesitate at the moments that matter most. Google had conversational AI comparable to ChatGPT well before OpenAI released it, and sat on it, in part because of what happened to Microsoft's Tay chatbot in 2016, which users manipulated into producing hateful content within a day of launch. The reputational caution of a trillion-dollar company created the opening a smaller, hungrier rival exploited. Legacy assets, as we argued in section 11.5, are a genuine advantage. Legacy risk aversion is the tax on that advantage, and it comes due at moments of technological discontinuity.

Then there is the efficiency shock. The Chinese lab DeepSeek reported a final-training-run cost of about $5.6 million for its near-frontier V3 model, a figure that excludes research, prior experiments, and infrastructure, so it understates the true all-in cost, but even multiplied several-fold it sits an order of magnitude below the all-in cost of Western frontier runs. Rivlin describes the episode as an existential scare for Silicon Valley, and its strategic meaning for business model design is worth spelling out: capital moats built on compute can invert with a single algorithmic insight. A cost structure that looks like a permanent barrier to entry may be one clever training recipe away from becoming a stranded asset.

Rivlin's summary judgment is that AI is simultaneously overhyped and underhyped, in the same way the internet was in 1999: the eighteen-month claims are inflated, the fifteen-year claims are probably too modest. I find that framing more useful for business model work than either the boosters or the skeptics. It argues for patient architecture: build the model-agnostic layers now, capture the efficiency gains that are real today, and position the balance sheet so you are still present when the deeper transformation of medicine, education, and research arrives. And his warning about Pi deserves the last word, because it applies to every company building companionship or advice into its value proposition: the performance of empathy is not empathy. Customers may not notice the difference immediately. Regulators, and eventually markets, will.

11.8 Implications for Strategy

Several strategic implications emerge from this analysis.

First, business model innovation matters more than technology innovation in the generative AI era. The underlying technology (large language models, diffusion models, multimodal architectures) is increasingly commoditized, and multiple providers offer comparable foundation models at comparable prices. The differentiation is in how the technology is configured into a business model: who is served, what is offered, how it is delivered, and how value is captured.

Second, the EDGE Framework should guide business model design, not follow it. Before selecting a St. Gallen pattern or drawing a BMC, executives should determine which EDGE dimension is strategically most important. If the priority is Efficiency, patterns like Self-Service and Pay Per Use are most relevant. If the priority is Growth, patterns like Razor & Blade, Mass Customization, and Two-Sided Platform dominate. If the priority is Empowerment, the Layer Player pattern and capability-transfer models deserve attention. The EDGE dimension is a strategic choice, not a retrospective label.

Third, the depth of AI integration into the business model determines the depth of the competitive moat. An AI feature added to an existing product (the Subscription + Add-On pattern) creates moderate switching costs. An AI engine that restructures Key Activities and Key Resources (the Self-Service or Mass Customization pattern) creates substantial structural barriers. An AI platform with network effects (the Two-Sided Platform pattern) creates the most defensible position of all. The question is not "are we using AI?" but "how deeply has AI reconfigured our business model?"

Fourth, legacy assets are the foundation of defensible AI business models, not dead weight. The most common strategic error in the generative AI era is assuming that incumbents should compete on AI capability. They should not. They should compete on the combination of AI capability with the proprietary data, customer relationships, brand trust, physical assets, and domain expertise that pure-play AI companies cannot replicate.

Finally, business model innovation is not a one-time event. The most successful companies execute several business model shifts over time, each reconfiguring different blocks of the BMC while keeping what works. The technology may stay the same. What changes is the configuration of value creation, delivery, and capture.

11.9 Conclusion

Generative AI is the most powerful enabler of business model innovation since the internet. But the word "enabler" carries the weight. AI does not innovate business models. Leaders do, by recognizing which configurations of customers, value, channels, resources, activities, partnerships, costs, and revenues are newly viable, and by having the courage to reconfigure their organizations accordingly.

The EDGE Framework provides the strategic compass: where should AI create value? The St. Gallen patterns provide the building blocks: what configurations are available? The BMC provides the design canvas: how do the pieces fit together? And the economic dynamics of generative AI (token economics, data flywheels, compounding personalization) provide the fuel.

The executives who will lead in this era are not those who deploy the most sophisticated AI models. They are those who see most clearly that the business model, not the model, is the unit of competition.

References

• Gassmann, O., Frankenberger, K., & Csik, M. (2014). The Business Model Navigator: 55 Models That Will Revolutionise Your Business. Pearson.
• Johnson, M. W., Christensen, C. M., & Kagermann, H. (2008). Reinventing your business model. Harvard Business Review, 86(12), 50-59.
• Osterwalder, A. & Pigneur, Y. (2010). Business Model Generation: A Handbook for Visionaries, Game Changers, and Challengers. Wiley.
• Yu, S. (2026). EDGE Framework for AI-Driven Business Model Innovation. HEC Paris Working Paper.