In the boardrooms of global brands, the same question keeps surfacing: Where should we place our generative AI bets? Stakes are rising fast. Once the domain of small pilot projects, gen AI is now being embedded into supply chains, underwriting, product design, customer service and more. However, early enthusiasm is giving way to a more sobering reality: many projects are failing to pay the expected returns. The technology may be dazzling, but it’s also disruptive, complex, and unforgiving of half-measures.
Alongside colleagues from IMD (Konstantinos Trantopoulos, Amit Joshi, and Michael Wade) and ETH Zurich (Jingqi Liu), we studied 100 corporate gen AI implementations across various industries, from finance to retail. Our research uncovered four distinct strategic approaches or archetypes that, intentionally or not, most companies are adopting in their approach to integrating gen AI: Bold Innovators, Disciplined Integrators, Fast Followers, and Strategic Builders.
Four ways companies are playing the game
Bold Innovators use gen AI to reshape their markets, taking on higher risk for the chance to lead. Heidelberg Materials, for example, relies on AI-driven chemistry simulations to develop low-carbon cement — not just improving a product, but potentially transforming the market.
Disciplined Integrators take a slower path, prioritising trust, control, and compliance. For instance, Roche applies gen AI in clinical trial monitoring, under layers of regulatory oversight, to protect patient data while maintaining precision.
Fast Followers are opportunists. They look for low-cost, high-impact wins — like CarMax, which uses gen AI to summarise used-car reviews, speeding up customer decisions without costly infrastructure changes.
Strategic Builders think in decades, not quarters. Group Allianz is building its gen AI stack to manage claims, detect fraud, and underwrite policies across global markets — a long game aimed at owning intellectual property and sustaining competitive advantage.
These archetypes aren’t fixed. Zurich Insurance began as a Disciplined Integrator, then rolled out gen AI chatbots, behaving more like a Fast Follower. Novartis shifted from buying off-the-shelf tools to building proprietary models for clinical trials. Within a single company, the R&D team might be innovating boldly while the legal department adopts a more cautious stance.
The trade-offs that define strategy
Our study also noted that what distinguishes success from failure is not whether companies use gen AI, but how they navigate the trade-offs that accompany it.
The first aspect is “expected benefit versus cost”. For instance, Carrefour uses gen AI to predict and prevent spoilage, boosting margins; Novartis to accelerate clinical trial design; and Zurich Insurance to improve compliance with plain-language underwriting.
Then comes risk tolerance. Financial firms may delay launches until security is bulletproof. Duolingo, by contrast, pushed GPT-4 features into the market quickly, taking advantage of lower regulatory risk in edtech.
The third decision point is speed versus stability. Snap Inc. launched AI features like My AI and AI Lenses at a rapid pace. General Motors has preferred a slower rollout, testing gen AI in legal and training functions before full deployment.
And the final aspect is “build in-house versus buy, or a bit of both”. Proprietary models can unlock unique competitive advantages, but off-the-shelf APIs often deliver faster, cheaper wins. JPMorgan Chase, has hired more than 1,500 AI experts to build proprietary models for fraud, compliance and customer services, or buy proven tools. Klarna, on the other hand, has bought an off-the-shelf OpenAI-powered assistant for most of its customer chats. However, many firms combine both – for instance, Enel integrates in-house gen AI with third-party APIs, and Maersk blends commercial tools for port modelling with custom logistics systems.
The moment of truth: execution and delivery
Once an organisation has selected its generative AI archetype, the real challenge lies in execution. Our research indicates that five foundational elements distinguish experiments that stall from strategies that succeed.
It begins with data readiness — clean, integrated, ethically sourced information. For instance, American Express rebuilt its data infrastructure from transactions, customer service and fraud monitoring for accuracy and privacy, while Roche created cross-functional governance to manage sensitive patient data under GDPR.
Next is technology architecture. Netflix operates a customised cloud system for rapid feature rollouts, while CVS Health combines cloud and on-premises systems to balance speed with security.
Governance is equally critical. Microsoft’s Responsible AI Standard enforces fairness checks and transparency, while Salesforce’s AI Ethics team embeds oversight into product development.
Organisational readiness can accelerate or block adoption. Airbnb uses cross-functional “tiger teams” to bypass bureaucracy; ING has embedded AI into its agile squad model, helping teams move smoothly from idea to production.
Finally, capability building promotes AI literacy throughout the workforce. PwC is investing $1 billion in Gen AI staff training, and Unilever trains marketing, finance, and HR teams via prompt engineering workshops.
In every case, these pillars must work as a system, aligned with the organisation’s archetype and ambitions — otherwise even the boldest strategy risks remaining just a plan on paper.
Our research shows that each archetype requires its own set of these pillars. Bold Innovators thrive on flexible cloud systems, rapid data integration, and lightweight governance that doesn’t hinder innovation. Disciplined Integrators rely on structured, privacy-compliant data, hybrid systems, and strict oversight. Fast Followers value speed, using cleaned but not perfect datasets, vendor APIs, and simple governance templates. Finally, Strategic Builders invest heavily in centralised data lakes, secure bespoke architectures, and deep capability building to sustain proprietary ecosystems.
Alignment is key
Ultimately, generative AI is not simply plug-and-play. To unlock its value, leaders must choose where to focus, how to execute — and be ready to adapt as they progress.
So, what’s the wrong move? Chasing every shiny new tool without a clear understanding of where it fits. And the right move? Making deliberate trade-offs, committing to a path, and ensuring that data, technology, governance, culture, and skills are all aligned to pull in the same direction.
Success depends on aligning clear priorities with disciplined execution. By choosing an archetype that reflects their goals, risk appetite, and resources, leaders can align the five pillars to build a focused roadmap. With the right framework in place, generative AI can become a catalyst for transformation — delivering tangible, lasting impact.
—–
About the Author
Yash Raj Shrestha is an Associate Professor at the Department of Information Systems, Faculty of Business and Economics (HEC) and Group Head of Applied AI Lab at the University of Lausanne. His research program aims to pioneer innovative and responsible advances in AI, fostering solutions that positively transform organizations and society.
Recently, Prof. Shrestha has been awarded “Future of Organizations Fellowship” by Organizational Design Community for 2024-2026. His research has appeared in leading outlets in both management and computer science, including Nature Machine Intelligence, Nature Computational Sciences, Organization Science, Strategic Management Journal, Strategic Entrepreneurship Journal, Harvard Business Review, MIT Sloan Management Review, EMNLP, IJCAI, ECAI, KDD, RecSys, MICCAI, CSCW, etc.
