Everyone says have an AI strategy. What does that actually mean in practice?

Everyone keeps telling you the business needs an AI strategy, and you nod along in the meeting, and then if anyone pinned you down you could not actually say what having one even means. You picture a deck. You picture a roadmap with coloured swimlanes. Some part of you suspects that whoever asked for it could not define it either, which is the quiet thing nobody says out loud around the table.

So here is the plain answer, before anything else. Having an AI strategy is not owning a document. It is living by a small set of decisions: where AI does and does not touch the business, who owns each bet, and how you will know it worked. That is the whole thing. The deck is just where those decisions get written down, and a deck with no decisions inside it is the most common thing in the building. If you can say those three things out loud in a lift, you have a strategy. If you cannot, no amount of formatting will rescue the slides.

Is an AI strategy a document or a set of decisions?

It is a set of decisions you behave by, and the document is downstream of those. The reason this matters is that a document can be produced in a weekend and still mean nothing, while three real choices, held consistently, will reshape a year. Think of it the way you would a household budget. The spreadsheet is not the discipline; the discipline is the small number of decisions you keep returning to, week after week, when the easy thing would be to drift.

And the drift is the default. Gartner's December 2025 survey of 197 senior business leaders found just 27% said they had a comprehensive AI strategy. So the gap you are feeling is the normal condition, not a private failing. The interesting part is what sits inside that gap. It is rarely a shortage of tools. It is the absence of a decision: nobody has said, in a sentence the whole team could repeat, what AI is for here and what it is allowed to leave alone.

What are the actual decisions inside a strategy?

Three, and they are smaller than the deck wants them to be.

The first is a boundary. Where does AI touch the business, and just as importantly, where do you keep it out? A strategy is mostly a list of what you have chosen not to do, so that attention concentrates on the few places AI moves a real number. The second is ownership. Who owns each bet, by name, with the authority to change how the work flows and the accountability when it does not pay? The third is measurement. How will you know value arrived, in a profit-and-loss line you would defend to a sceptic, by a date you fixed in advance? Boundary, owner, measure. Everything else in the document serves those three.

Look at what happens when the second one goes missing. Deloitte's October 2025 State of AI work, drawn from a survey of senior leaders, found that 51% named leadership and strategic misalignment as a barrier to scaling AI value, and that the function most often left holding work redesign was IT, at 39%, against just 12% for the people side in HR. Ownership is not a box on an org chart; it is the behaviour that decides whether a pilot ever escapes the pilot. And without the third decision, the measurement, projects quietly die. Gartner predicted that at least 30% of generative AI projects would be abandoned after the proof-of-concept stage by the end of 2025, citing unclear business value as a leading cause. Unclear value is what you get when nobody decided, up front, what success would look like.

Why does a tool list feel like a strategy when it is the opposite?

Because the tools are the visible, purchasable, demonstrable part, and decisions stay invisible until you live by them. It is easier to buy a platform than to choose a boundary. So the deck fills with logos, and everyone feels productive, and the year passes without a single number moving. Deloitte's October 2025 research put the imbalance plainly: only 16% of organisations said they had fully designed the roles, processes and operating models needed to fit AI into the actual work, and the 59% who took a technology-first approach were 1.6 times more likely to report that AI was not exceeding expectations. The models were fine. The decisions about how the work should change were never made.

This is the spine of the whole thing, and it is worth saying flatly: the technology works. The approach is what fails. BCG's 10-20-70 framing (the Boston Consulting Group rule of thumb that splits where AI value actually comes from) makes it concrete: in a successful deployment, around 10% of the value comes from the algorithms, 20% from technology and data, and 70% from people and process, from the roles, workflows and behaviours your three decisions actually govern. A tool list spends its energy on the 10%. A strategy spends it on the 70%. That is the difference, and it is the entire difference.

There is a quieter layer underneath these three decisions, and it sits with you. The clarity to draw a boundary, to assign an owner, to commit to a measure you might miss, comes from a settled head, not an anxious one. Most leaders are trying to install new software on broken hardware: a clever tool bolted onto the same overloaded judgement. The bottleneck is no longer the technology. It is the steadiness of the person deciding what the work serves. The leaders who win the next decade are the ones who upgrade themselves first, and then the three decisions get easier to make and far easier to hold.

1. Draw the boundary. Write one sentence naming where AI touches the business this year, and one naming what it deliberately leaves alone. Saying no is the work.
2. Name an owner per bet. One accountable person, by name, with authority over how the work flows. An unowned bet drifts back into the old way.
3. Commit a measure in advance. Tie each bet to a profit-and-loss line and a date before you start, so value is decided up front, not argued about later.
4. Live by them in public. Repeat the three decisions in every review, because the direction the room keeps hearing is the direction the organisation keeps steering.

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.