"AI strategy" is one of the most overused terms of the 2020s. Most documents companies call "AI strategy" are in fact long lists of technologies that could in theory be exploited. A real strategy says which decisions in the company will be made differently in two years and what role AI plays in that change.
A good AI strategy answers three questions: in which business drivers is AI materially significant, which decisions move from humans to machines (or vice versa), and how does this change cost structure, customer experience or competitive position. If your strategy paper mentions "generative AI" but does not mention a single decision that machines will be making in two years, it is not a strategy — it is a technology summary.
Start by listing 3–7 business drivers that leadership has accepted as strategically material: e.g. customer churn reduction, price optimization, warranty exposure management, processing unit cost, workplace safety. Each driver already has a euro target — use it. AI strategy is built on top of these drivers. If a driver is not strategically material, an AI use case built for it will never receive executive attention or funding.
For each driver, identify 3–10 candidate AI use cases scored on a common scale: business value (euros per year), feasibility (data availability, model maturity), complexity (integrations, change management) and risk (EU AI Act, GDPR, ethics). The portfolio is not a spreadsheet — it is a living list re-scored quarterly as learnings come in from delivered initiatives. This is the continuously updated operationalization of the strategy.
A roadmap is not a project timeline — it is a sensible sequence where structural investments (semantic layer, governance, MLOps spine) precede the use cases that need them. Example: if the first half of the year aims to deploy four generative use cases, a shared agent control plane, prompt versioning and monitoring must be built by Q1 at the latest. Otherwise in Q3 every use case builds these on its own, and in Q4 the company has to refactor everything into one.
Strategy has to show up in board reports quarterly. A good AI metric stack contains both outcome metrics (business value, retention, unit cost) and stage metrics (use cases in production, MLOps maturity index, governance readiness as a percentage). "AI investment per year" alone is not a metric — it is a budget line. A metric must describe what is happening in the business, not what the business is buying.