AI strategy: how to write one your organisation will use
BY PRAXES··1 MIN READ
TL;DR
Most leadership teams already have something they call an AI strategy. Usually it is a slide with a vision statement, a list of tools the company has bought, and a pilot or two. We keep seeing the same thing happen next: six months later the slide is unchanged, and so is the way work gets done.
This guide covers what an AI strategy is, how fast to move, what a strategy needs to contain, how to write one in seven steps, and how to spot the problems where a better process would do more than any model. I have written it for the managing director or functional lead who has been handed the brief and wants something the leadership team will use on Monday.
An AI strategy is a decision document. It states where artificial intelligence will create value in your specific business, what has to be in place for that value to show up, and how you will know it has. Which platform to buy, which team to train first, and which vendor to call all follow from those answers.
It sits alongside two other documents. An AI policy people will follow sets the rules for how staff may use AI tools. A roadmap sequences the work. You will need all three eventually, and the strategy comes first because it tells the other two what they are for.
Access to AI is no longer the constraint. 88% of organisations report regular AI use in at least one business function. Yet only 37% say AI has contributed positively to EBIT, a share that has barely moved in a year, and 74% of companies struggle to achieve and scale value from AI.
Three patterns sit behind most of those numbers. Tools arrive and nobody builds them into the work, so the work carries on as before. Training happens and Monday looks exactly like Friday. And pilots run in pockets that the bottom line never notices. None of these is a technology failure. Each one is what happens when a strategy starts at the tool and works forwards.
We have had a run of conversations with executives who feel overwhelmed by how fast AI is moving. They all ask some version of the same question: should we move at the same speed?
Our short answer comes from the fable of the tortoise and the hare. Some companies want to be the hare. That is a legitimate choice, and it carries more risk. Fast experiments and fast rollouts can buy you an edge, and they can just as easily buy you a pile of half-integrated tools nobody uses.
For a company that has not started yet, we prefer the tortoise. In practice, being the tortoise means four things:
The hare looks impressive on launch day. The tortoise looks impressive a year later.
The tortoise only wins because it entered the race, though. In the story, nobody wins from the grandstand. You do not have to match the fastest company's pace, but you do have to start moving. Waiting for things to calm down is a decision too, and we do not expect the pace to slow.
A useful question for this quarter: what is the smallest pilot you could launch that is built to scale? In our view, built to scale means it has a business owner, runs on data you already trust, and replaces a step in a workflow people use every day. A pilot like that can grow into daily work. A pilot missing any of them usually stays a demo.
A strategy that holds up answers one question in each of five dimensions. AI readiness fails at whichever of these is weakest, which is why we look at all five before recommending anything.
| Dimension | The question it answers | What a good answer looks like |
|---|---|---|
| Strategy | Where does AI create value in this business? | Two to four named problems, each with an owner and a measurable outcome |
| Data and systems | Can our systems give a model what it needs? | A plain list of the data each use case depends on, where it lives, and how clean it is |
| People | Will the people doing the work change how they work? | A sponsor per use case and a plan for every team whose workflow changes |
| Process | Where in the workflow does AI sit? | A redrawn process map with the human decision points marked |
| Tooling | What do we build, buy, or leave alone? | Platform choices tied to specific use cases, each with an exit route |
Think of it the way an operations director thinks about a production line. Buying a faster machine does nothing if the bottleneck is two stations downstream. The five dimensions are a way of finding your bottleneck before you spend on the machine, and a structured AI readiness assessment is how you find it.
Short enough that the leadership team can hold it in their heads. A few pages for the main document and a one-page appendix per use case is plenty. I would rather see a four-page strategy that moves three budget lines than a 40-page one that moves none.
Length is a symptom. A long strategy usually means the hard choices were not made, so every option stayed in. The value of the document is in what it leaves out.
More often than the market suggests. AI multiplies whatever you give it, so an inconsistent process run through a model becomes a faster inconsistent process. If two teams disagree on how a handover works, fixing the handover will do more than any tool.
Saying this out loud is part of a good strategy. It protects budget for the problems where AI does fit, and it builds credibility with a team that has seen technology projects come and go. An AI operating model then gives the strategy a home, so decisions like these have an owner.
An AI strategy is a short document that states where AI will create value in a specific business, what must be in place for that value to arrive, and how progress will be measured. It should change how money and time get spent, otherwise it is a vision statement.
A useful AI strategy covers five dimensions: strategy, data and systems, people, process, and tooling. Each needs a clear answer, because the weakest one limits what the others can deliver.
In our view, a first version needs a few weeks of focused leadership time rather than months of analysis. The bigger commitment is the quarterly review that keeps it current as the technology and the business change.
The chief executive or managing director should own it, with each use case owned by the business leader whose numbers it affects. IT and data teams are essential contributors, but ownership by a technical function alone tends to produce a technology plan.
Yes, and it can be lighter than a large enterprise's. A company of 50 to 500 people needs clarity on a handful of use cases, the data they depend on, and who is accountable, which fits on a few pages.
There is no need to match the fastest company in your sector. Fast rollouts can buy an edge, and they can also leave you with tools nobody uses. For a company that has not started, we recommend a steadier pace: AI-ready data, early governance, a fixed workflow, and a pilot built to scale. Waiting for the pace to slow is also a decision, and we do not expect it to slow.
If you want a scored view of where you stand across the five dimensions before you start writing, the AI Readiness Diagnostic is free and produces a report you can take to your exec team.
Ready to move from reading about AI to getting your organisation ready for it? Start with the diagnostic.
Take the AI Readiness Diagnostic