--- title: "AI Stages of Adoption Explained" date: 2026-08-15 description: The five stages from Experimenting to Scaling, why the model plots investment rather than time, and the multi-speed reality: your organisation is not at one stage, it is at several at once. author: Mario Thomas canonical: https://mariothomas.com/videos/ai-stages-of-adoption-explainer/ --- This is the AI Stages of Adoption, AISA, the model I built to answer a question every Board is now asking. Where are we, honestly, on the AI journey? Before I walk you through AISA, I want to tell you where the model came from, because its history explains why it looks the way it does. As organisations began moving their on-premises compute estates to the cloud in earnest, my world was migration and modernisation, and we needed a simple way of describing where an organisation truly was on that journey. There were two kinds of organisations at the time, digital natives, think companies like Netflix or Monzo, who were born in the cloud, and enterprises carrying enormous on-premises investments approaching an inflection point, where refresh cycles or technical debt forced the question. One airline I worked with had its cloud moment decided for it, when the airport compulsorily purchased its data centre to make room for a runway. So we plotted value against time, digital natives on one curve, enterprises on another, and described four stages of cloud adoption. Project, the first hands-on experience, discrete initiatives, learning by doing. Foundation, governance, security, landing zones, the Cloud Centre of Excellence. Migration, moving the estate at scale to the cloud. And reinvention, going cloud native, modernising, building what the cloud made possible. It worked because organisations could self-identify. Naming their stage drove investment and sponsorship decisions. The Cloud Stages of Adoption, or SOFA, became common language that we used in all of our customer conversations. But when AI arrived, I knew that model wouldn't carry over, because it was built for a technology journey that IT led. AI adoption is different in three fundamental ways. First, there are multiple parallel journeys. Cloud adoption was one linear path driven by IT. AI initiatives emerged simultaneously all over the business, each at its own pace, with its own objectives. Second, it's business led, not IT led. AI is a matter for the Board. It transcends the IT boundary. Business units adopt it directly, chasing immediate value, often without asking anyone. And third, value realisation varies wildly. Some AI initiatives are quick wins. Some need serious investment before any value shows. Cloud value followed a predictable pattern. AI value doesn't. So the AI Stages of Adoption had to be a different kind of model. Every journey starts with Experimenting. This is where organisations and individuals explore and test AI, often without any formal oversight. It's ad hoc. Initiatives are driven by single departments. Budgets are small. And the tools are the consumer ones. ChatGPT, Claude, Amazon Q, Grok. This is where shadow AI is born. People reaching for tools the organisation hasn't yet sanctioned to solve problems the organisation hasn't prioritised. For example, a marketing team quietly using a chatbot to draft campaign copy, while customer service pilots automated triage on common queries. Nobody signed it off, but both are delivering value. The signs you're here are widespread unofficial use of consumer tools, small, isolated pilots, no dedicated team, limited executive attention. And that's exactly what moves you forward. The moment successful pilots create demand across departments, and the realisation lands that ungoverned adoption is a business risk. Executive awareness turns into sponsorship. And Experimenting becomes Adopting. Adopting is where AI gets formalised. The technology is integrated into existing processes. People are trained. Dedicated budget appears. And the first governance frameworks emerge. This is the stage where the AI Centre of Excellence takes shape, and where it reports matters enormously. My advice is that it anchors to the CFO, or to the risk committee, but not IT. Delegate AI governance to the technology function, and the Board loses exactly the line of sight it needs. For example, document processing in operations, natural language models extracting and routing what used to be manual keying, running under real policies with a real budget line. The signs you're here are a forming Centre of Excellence, initial AI policies, dedicated budget, multiple successful pilots, and executive sponsorship that's real rather than polite. You move on when the focus shifts from getting it working to making it better. When production use cases start demanding standardisation, and your growing internal expertise starts hitting its limits. Optimising is where the effort turns to maximising value from what you've adopted. Models are tuned to your domain and your data. Feedback loops sharpen them. Governance and risk management operate as frameworks, not aspirations, and ROI is measured rather than asserted. For example, a fraud detection model, fine-tuned on your own transaction history, sitting inside a standardised deployment pipeline, with monitoring and retraining built in. The signs you're here are a documented AI strategy, a standing steering cadence, dedicated teams, multiple production systems, standardised development practices. The transition beyond this stage is the deepest one in the model because it's cultural. It happens when AI stops being a tool you apply and starts reshaping how the business operates. Then AI is embedded in core processes, and new AI-enabled products start emerging. That's when Optimising becomes Transforming. Transforming is where the business redesigns itself around the capability. Processes are rebuilt rather than assisted. Roles change. Business models change. An AI-first mindset takes hold. New products, services, and revenue streams appear that simply couldn't exist without the technology. For example, generative design in product development. The organisation isn't using AI to speed up the old design process. The design process itself now runs through the model, and the products it ships are ones the old process couldn't have produced. You're here when you see AI embedded in core products and services. Data-driven decisions become the norm, and revenue is attributable to AI, and you have cross-functional initiatives. AI is a standing item in the boardroom rather than occasional presentation. You move on when transformation has been proven in enough places that the question becomes scale, standardising what works enterprise-wide. And Scaling is the final stage. The organisation has fully embraced. AI and is extending it across the whole enterprise and its ecosystem. Platform infrastructure serves many use cases at once. Operations are automated. Governance is mature. The culture sustains innovation on its own. For example, an enterprise AI platform running supply chain optimisation, demand forecasting, and process automation off shared infrastructure, shared data foundations, and shared controls, with partners and suppliers drawn into the same capability. The signs that you're here are AI is driving significant revenue, you have automated operations, advanced model development is now happening in-house, and you have an ecosystem that renews itself. But a word of caution. Scaling is the final stage, not the mandatory destination. Not every function needs to reach it, which brings me to probably the most important property of the AI Stages of Adoption. If you note the axes, the stages are plotted against investment, not time like the stages of adoption for cloud, and that's deliberate. Investment here means five things at once. Money, obviously, but also compute, licences and services are in that money category. But we also include people, training them, hiring them, building teams. Data is an investment, so cleaning it, labelling it, governing it. Processes that get redesigned, workflows and change management, and of course, time itself. Plotting value against investment shows you the honest shape of AI initiatives. Some are quick wins, with minimal investment and outsized value, like foundation models applied to content generation. Others, custom models for complex operations, domain-specific systems, demand substantial investment before value appears, and then become strategically decisive. The slope of each curve is its value investment ratio, and the ratios vary. That's not a flaw in the portfolio. That's what an AI portfolio looks like. AI is, of course, multi-speed. And here's the reality that makes AISA different from every adoption model that came before it. Your organisation is not at one stage. It's at several at once. There is no single migration. There are pockets of adoption everywhere. A language model generating content in marketing. Machine learning doing predictive maintenance in production, each on its own path, at its own pace, with its own return. So never ask what stage are we at, and accept one answer. Read the model function by function. Marketing may be Optimising, while finance is Adopting, and operations are still Experimenting. And success in one area catalyses the others. For a Board, that means stage appropriate governance for each function. Don't over-govern the experiments, and don't under-govern the systems already in production. The averaged score hides the picture. The function by function reading is the picture. The AI Stages of Adoption is openly published as part of the Remake Library. The five stages, the indicators that tell you which one you're truly in for which function, and the questions a Board should ask at each are all described at mariothomas.com. If you take one thing from this video, take the multi-speed point. Your organisation isn't at one stage, it's at several at once. Marketing may be Optimising, while finance is still Adopting, and operations is Experimenting. So read the model function by function, and govern each function at the stage it actually reached, not at the average. Everything I write on AI and emerging technology for boards and senior leaders is at my website at mariothomas.com. Thanks for watching.