What does an actual AI strategy document contain?

You have been told to produce an AI strategy, the document is open, the cursor is blinking at the top of a blank page, and you have no real idea what is meant to fill it. You know what a tool list looks like. You know what a budget request looks like. This is neither, and the people who asked for it could not define it either.

So here is the answer, plainly. An AI strategy document is a short, defensible argument about how AI changes the way your business makes money, followed by the choices that argument forces. It has six working parts: a thesis on your economics, a value map, an operating model, a workflow-redesign plan, a capability plan, and a measurement framework. Everything else is appendix. If a section connects to where value is created or destroyed in your model, it belongs in the document; if it does not, it belongs elsewhere.

Why does every AI strategy I see read like a shopping list?

Because the easy thing to write down is the technology, and the technology is the smallest part of the problem. BCG's 10-20-70 framing is the clearest correction here: in a successful AI deployment, roughly 10% of the value comes from the algorithms, 20% from technology and data, and 70% from people and processes, the roles, workflows, change management and governance. BCG names the primary obstacles to AI transformation as organisational rather than technical. A document that spends most of its pages on models and platforms has inverted its own weighting.

The evidence for what happens when you skip the 70% is now hard to argue away. S&P Global Market Intelligence, surveying over a thousand IT and business leaders across North America and Europe, found the share of organisations abandoning most of their AI initiatives jumped to 42% in 2025, up from 17% the year before, with the average organisation scrapping 46% of its proof-of-concepts before they ever reached production. The IBM Institute for Business Value, surveying 2,000 chief executives across 33 nations, found that only 25% of AI initiatives had delivered the return their leaders expected, and only 16% had scaled across the enterprise. The pattern is consistent. Value lives in the work, so the document has to describe the work.

What are the sections an AI strategy document should actually contain?

Start with the thesis. One page. How does AI change the economics of your specific business: your cost to serve, your pricing power, your speed to a decision, the defensibility of what you sell? This is the part everyone skips, and it is the part that makes the rest coherent. A retailer's thesis is not a software company's thesis. Write yours.

Then the value map. List the places in your model where AI moves a number, rank them by value and feasibility, and be honest that most of the list will wait. With the average organisation already scrapping nearly half of its proof-of-concepts, scaling everywhere at once is how you join the 42% that walked away. Choose the few where you can win, and protect them.

The operating model section names who owns this. Gartner found 70% of chief data and analytics officers (the executive who owns an organisation's data and the way it is used) now hold primary responsibility for the AI strategy and operating model, and the share of those officers reporting directly to the CEO rose to 36% in 2025, up from 21% the year before. Ownership is being formalised at executive level for a reason: an AI strategy with no named owner and no decision rights is a wish. This section also carries your governance and risk posture, in plain language a board can hold.

How do I make this a strategy rather than another roadmap to nowhere?

By forcing every section to commit. Dataiku's 2026 Harris Poll of global chief executives is uncomfortable on this point: those leaders estimated that about 35% of their own AI projects were more about optics than outcomes, AI washing to signal innovation rather than deliver value, and 78% worried AI could cost them their job and their company's future. The blank page you are looking at is the honest version of that anxiety. The way through it is specificity. A strategy commits to numbers that will move and workflows that will change; a roadmap to nowhere lists initiatives and hopes.

There is a quieter section most documents omit, and it belongs to you, the author. The pressure to produce this under a board that is rushing you degrades the very judgement the document needs. The decision-fatigue research is stark: studying over 1,000 parole rulings, researchers found favourable decisions fell from around 65% at the start of a session toward nearly zero by its end, then returned to about 65% after a break (Danziger and colleagues, PNAS, 2011). And higher resting heart rate variability, the autonomic marker linked through the vagus nerve to the prefrontal cortex, tracks with better executive function, though that work is still early and exploratory. The point is plain. The clarity to write a good strategy is a physiological state, not only an analytical one. Upgrade yourself first, then the page fills faster.

1. Write the one-page economic thesis first. Name the levers AI moves in your model: cost to serve, pricing power, speed, defensibility. The rest of the document serves this page.
2. Build the value map and rank it. Pick the few opportunities where you can win, and say out loud which ones wait.
3. Assign ownership and governance. Name the executive who owns the operating model and the decision rights that come with the role.
4. Specify the workflow redesigns. For each chosen opportunity, describe how the work itself is rebuilt, because that is where the return is found.
5. Define the measurement. Tie each initiative to a P&L line and a date, so the document can be held to account.

About the author

Thomas Green

British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.