AI Strategy & Adoption
Enterprise AI adoption has advanced quickly. Organisations have acquired models, platforms, development tools, and increasingly capable infrastructure. They have established governance functions, trained specialists, and launched growing numbers of experiments. What remains much less consistent is the ability to determine where AI should be applied, which opportunities deserve investment, and how those opportunities should progress towards production. This is why I increasingly think about AI adoption as an organisational capability rather than a collection of projects.
An organisation that depends on individual experts to recognise valuable AI opportunities can produce excellent results, but it will struggle to reproduce them consistently. The quality of the opportunity pipeline depends on who happens to be in the room, how well they understand the business, and whether they can connect that understanding with what AI can realistically deliver. Making AI adoption systematic means creating a repeatable way to bring business context, domain knowledge, economics, data science, and production thinking into the same process.
The starting point is not the model, platform, or form of AI. It is the business. Every organisation can be understood through the domains in which it operates, the journeys through which it creates value, and the decisions that shape the outcomes of those journeys. Some of those decisions work extremely well. Others produce unnecessary losses, false positives, delays, rework, customer friction, or operational cost. These are the places where an AI conversation should begin because they connect the technology directly to something the organisation already wants to improve.
This changes discovery from asking where AI could be used to asking where business performance could be improved. Generic use-case discovery tends to produce generic opportunities such as fraud detection, customer service, document processing, forecasting, and employee productivity. Domain knowledge allows the conversation to go deeper. A mortgage discussion may begin with refinancing churn but expose more consequential decisions around underwriting or delinquency. A payments conversation may begin with fraud and reveal opportunities around merchant pricing, settlement, retention, or liquidity. Understanding the surrounding business journey makes it possible to discover opportunities adjacent to the original problem rather than accepting the first plausible use case.
AI discovery is often measured by activity. Workshops generate dozens of ideas, which become use-case inventories and eventually backlogs of potential proofs of concept. I do not think the objective should be to maximise that number. A credible opportunity should have a material business outcome, an identifiable decision or constraint contributing to it, sufficient data and intelligence to improve that decision, and a practical point at which the organisation can intervene. The expected improvement must also justify the effort, cost, and operational change required to capture it. Applying those tests should eliminate opportunities, leaving a smaller portfolio with a much stronger basis for investment.
Production intent also needs to exist from the beginning. A proof of concept can demonstrate that a model works, but it does not establish that the organisation can create value from it. The data available during development may not exist at the point of decision, required features may be difficult to reproduce within the required time window, or a model may produce more exceptions than an operational team can handle. The organisation may lack authority to act on the prediction, or the intelligence may arrive after the outcome has become difficult to change. These are not implementation details to consider after a successful experiment. They help determine whether the opportunity was viable in the first place.
The technology required to build AI has become increasingly accessible. Models are easier to consume, infrastructure is more capable, and development tools continue to improve. The ability to make good AI investment decisions has not become equally accessible. Organisations still need repeatable ways to understand the business, identify consequential decisions, compare opportunities, bring the right expertise into the conversation, and establish production intent before scarce engineering resources are committed.
That is what I mean by making AI adoption systematic. The objective is not simply to produce more AI. It is to create an organisational capability for repeatedly identifying where AI belongs, determining why it matters, establishing whether it can be delivered, and moving the strongest opportunities towards production.
