In this article
An AI strategy is not a list of tools. It is not a vision slide. And it is not a policy, though it needs one. If it cannot be read in ten minutes and acted on next week, it is not finished.
This guide is for the people who set direction: owners, directors, trustees, senior managers. For a first task to try this month, start with AI for business: where to start. This is the step above it.
Start from your goals, not from AI
Write down what the organisation is trying to achieve in the next year or two. Three goals is plenty. More customers in a new region. A shorter wait for a service. Better-fit enquiries. More time for supporters.
Then, for each goal, ask what work stands in the way. Where does time go that should go elsewhere? Which tasks repeat every week? Where is information read, sorted and retyped by hand?
That is where AI might help. AI is good at legwork and unreliable at judgement, so the question is not "where could we use AI?" It is "which of our goals has legwork in the way?"
Starting here also gives you the business case. In the government's 2026 survey of UK businesses, 65% of those not using AI cited a lack of a clear business case or proven return. For each proposed use, set out the expected benefit, the cost and how you will test it.
Decide what to aim AI at
We use three kinds of aim to organise the options.
| Kind of aim | What it looks like | What it needs first |
|---|---|---|
| Inside the organisation | Notes, drafts, summaries, research, reporting | Rules on tools and data, and a person checking the output |
| For the people you serve | Faster replies, clearer answers, help finding the right person | Evidence of what those people need, and a check before anything reaches them |
| In what you offer | A new service, or an old one delivered differently | Proof from the first two, and a decision at board level |
Aim first at the work inside the organisation, on tasks where errors can be caught and corrected before the output is used. Move outward only when the evidence says you are ready.
The strongest aims for the people you serve come from finding out how they choose. For HCML, we started with a behavioural study into how intermediaries choose a corporate health provider. On the rehabilitation side, insurers and solicitors choose by location and specialism first, so we built a case manager finder that filters by location, profession, specialism and region. The aim came from research, not from the technology. The same holds for AI: ask the people you serve before you decide what to build for them. There is more on that in what good discovery actually looks like.
Decide what to leave alone
A strategy is as much about what you will not do. Positioning works the same way: choosing a position means giving something up.
Write a short not-doing list and put it in the strategy. Four questions help you build it.
- Is the relationship the product? If people choose you for the person they deal with, keep a person there.
- Can someone check the output? If nobody has the time or knowledge to check it, AI should not produce it.
- Can the data go into a tool? If the information is too sensitive for any tool you have approved, the use waits.
- Is the process sound? AI makes a messy process faster at being messy. Fix the steps first.
Decisions about people belong on the list too: hiring, money, care, access. AI may inform them. A person decides.
Sequence it
Order matters more than ambition. Here is the sequence we would start from, over a year. Adjust the timing to your organisation.
- 01First quarter, find out
Ask people what they use now and where time goes. List the tools in use, official or not. Agree the rules on tools, data and checking.
- 02Second quarter, prove
Run two or three uses on real work, each with an owner and a before measure.
- 03Third quarter, decide and spread
Keep, change or stop each use on the evidence. Train the teams who will use what stayed, on their own tasks.
- 04Fourth quarter, look outward
Only now weigh the bigger aims: uses for the people you serve, work that needs better data, tasks worth automating.
Each stage earns the next. A strategy that starts with a large build skips the stages that tell you whether it will work.
The rules come in the first quarter for a reason. How to write an AI policy covers them in detail. For questions about AI and personal data, the ICO's Guidance on AI and data protection is the reference. It is under review after the Data (Use and Access) Act, so read the current version, and take legal advice where the stakes are high.
Govern it without a committee
A policy is not governance. A document nobody reads changes nothing. Governance is who decides, and how often.
Start with these three arrangements, then assess what else each use needs.
One senior owner. Someone at the top who is answerable for the strategy and has time set aside for it. Not the most enthusiastic user. The person who can say no.
An owner for every use. One named person per use, who checks the output, keeps the measure and calls the decision.
A register. One page listing each use, its owner, the data it touches, who checks it and its review date. The leadership team or board reads it each quarter.
We run our own weekly work this way. Each Monday, figures for the sites we look after are pulled from Google's own tools, a person ticks the week's plan, and checks follow, each with a plain verdict. A second AI lab reviews our code and client work before it goes out. It works because each job has an owner, a rule and a record.
Measure what changed
Measure at two levels.
For each use: time, quality and take-up, set against the before. If you did not measure the before, you cannot claim the after. Then keep it, change it or stop it.
For the strategy: is the goal moving? Time saved on reporting means little unless it went somewhere useful, such as faster replies or more time with supporters.
Watch take-up as closely as results. In the ONS survey of June 2026, only 10% of businesses with 10 or more employees that use AI said they use it extensively. Licences bought are not the measure. Work done differently, week after week, is.
Stopping a use is a good result. It means the measure did its job.
What to write down
The finished strategy fits on two pages. This outline is ours, not a standard.
| Heading | What it says |
|---|---|
| Goals | The two or three things the organisation is trying to achieve |
| Aims | The uses of AI that serve those goals, and which kind each is |
| Not doing | What AI will not be used for, and why |
| Sequence | What happens each quarter, and what has to be true before the next |
| Rules | Which tools, what data may go in, who checks the output |
| Owners | The senior owner, and an owner for each use |
| Measures | The before for each use, and how the goals will be tracked |
| Review | When the leadership team looks again |
Review it at least twice a year. Tools change, and so does what your people can do with them.
Where to start
This week, write your three goals at the top of a page. Under each, list the work that stands in the way. Then ask the people who do that work what they already use, and what worries them.
If you would like help turning that into a sequence your team can run, that is what our AI consultancy does. Or get in touch and talk it through.
Common questions
What is an AI strategy?
A short set of decisions about how an organisation will use AI: the goals it serves, the uses worth doing and in what order, what is left out, who owns it and how it is measured. It sits under the organisation's own strategy, not beside it.
Does the UK have an AI strategy?
Yes. The government published its National AI Strategy in September 2021, and the AI Opportunities Action Plan in January 2025, both on GOV.UK. They set national direction. An organisation still needs its own, built on its own goals and work.
Is there an AI strategy template?
The outline above is a starting point: goals, aims, not doing, sequence, rules, owners, measures and review. No template can know your goals or your work, so fill it in from conversations with the people who do it.
