In this article
- Why is now the time to choose?
- What does an organisation that uses AI strategically look like?
- Why is hype a poor compass?
- How do you filter hype with four questions?
- What does that look like in examples?
- What goes into an AI roadmap?
- How do you start with an AI strategy this month?
- When do you not need a strategy?
- What is your first step?
- Sources
In short
- An AI strategy is a decision, not a list of tools: where you use AI over the next ninety days, where you deliberately do not, and who owns it.
- We believe the difference a year from now will come down to who chose early and learned from it. Every quarter without a choice is a quarter without experience built up.
- Hype makes a poor compass. Test every idea against four questions and record the decision in an AI roadmap, owned by someone with the mandate to say no.
An AI strategy is a decision about three things: where you put AI to work over the next ninety days, where you deliberately do not, and who drives it. We think a board or management team should make that decision now. Every quarter without a choice is a quarter in which your organisation learns nothing about what AI can do in its own work. The main condition is that someone with a mandate is willing to say no.
Why is now the time to choose?
Every process you set up seriously with AI teaches your team something: what works, where it goes wrong, how much checking it needs and which agreements you need. That experience accumulates. This is our conviction, not a measurement. We think the difference between organisations a year from now will lie in who set up a process well first and learned from it.
The figures suggest there is room for that. According to Exact's MKB Barometer 2026 (1,719 interviews in the first quarter of 2026, companies with 2 to 250 employees), 70 per cent of the companies surveyed use AI. At the same time, 47 per cent recognise themselves in the statement that they do not really know how to make use of it (Exact, in Dutch). Statistics Netherlands (CBS) reports lower figures: in 2025, 27 per cent of companies with 10 to 50 employees used AI, and 45 per cent of those with 50 to 250 (CBS, in Dutch, provisional figures). One explanation is that the surveys measure different things: Exact asks whether a company uses AI at all, while CBS counts specific AI technologies, for a different year and a different group. Whichever count you pick, usage does not tell you that it delivers anything.
What does an organisation that uses AI strategically look like?
A fictional picture, a year from now. A wholesaler with forty employees chose one process last autumn: preparing quote requests, with an owner, a goal and a fixed moment when the management team looks back. Four other ideas are parked on purpose, each with a reason. When something new appears, the team sets it beside the same list and makes a decision.
That is what we mean by strategy: an organisation that knows what it is doing, why, and when it will look again.
Why is hype a poor compass?
Something new appears almost every day: a new model, a new agent, a story about what someone built over a weekend. What those stories rarely show is what it takes to make something like that work in a real business process, with your own data, your own systems and people who have to use it. That is our reading, not a measurement. Much of what is new is genuinely interesting. It gets difficult once hype sets the agenda: the news of this week then decides what you do next month.

Image: Hello Growth, schematic illustration of our view, without measured data.
The diagram shows our reading, not a measurement. The dashed line is what the news promises, and it rises steeply. The dark line is an organisation running ad hoc experiments that nothing connects. The blue line is an organisation with an AI roadmap: it rises in steps, because every decision point records what has been learned. You will not catch up with the hype with either of them. What is useful to you is the blue line.
How do you filter hype with four questions?
Put every idea next to the same four questions. If it fails the first one, you do not need to go on.
- Which problem does it solve? Can you express it in hours, errors, lead time or revenue? "Competitors are doing it" is not a problem. Sometimes a simpler process improvement beats AI, and that is a perfectly good outcome.
- How certain is it? Does it work at organisations like yours, or only in a demo? Look for a source that is not the seller. An agent that decides on its own is often less certain than a fixed workflow with AI in a few places. See also What is an AI agent, and when do you really need one. Which tool suits which task is covered in AI tools for your business: which one for which job?.
- Who owns it and who checks it? Without an owner who knows what good enough looks like, an idea stays an experiment. How to organise that knowledge is covered in AI implementation: the bottleneck is people, not the model.
- What happens to your data? What information may go in, which licence goes with it and who keeps an eye on that? You need to know this before you start, even if the agreements are not finished. How to put that on paper is covered in Using AI safely: how to write an AI policy for your team.
Rate what is left on impact, certainty and feasibility, each from 1 to 5. The exact score matters less than the order it gives you, which you can explain to someone who was not in the room.
What does that look like in examples?
These are fictional examples. Suppose the wholesaler above has collected five AI ideas. After the four questions and a score from 1 to 5, this is what is left:
| Idea | Impact / certainty / feasibility | Decision |
|---|---|---|
| Preparing quote requests | 4 / 4 / 4 | Start, ninety days |
| Writing meeting minutes | 3 / 4 / 5 | Include, after agreeing data rules |
| Autonomous agent for customer queries | 5 / 2 / 2 | Later, first fix processes and data |
| Chatbot on the website | 2 / 3 / 4 | Not now, no clear problem |
| Avatar for social media | 1 / 2 / 3 | No |
The agent has the highest impact but does not come first, because certainty and feasibility are low. Two ideas drop out on purpose. That is what separates a strategy from a wish list.
Suppose three other organisations make the same weighing and reach a different result:
- A building firm chooses to record its site meetings, because hours go into that every week.
- An advisory firm starts with agreements on client data, because files may not go into every tool.
- An online shop puts the customer-query agent on the later list, because its returns data is not yet in order.
What goes into an AI roadmap?
A usable roadmap fits on one or two pages and contains:
- The goals AI must contribute to, in time, quality or revenue.
- Three to five candidates with their score and the reason for the order.
- One process that starts now, with an owner, freed-up time, a measurable goal and a fixed moment to look back.
- The conditions: which tools and licences, what may and may not go in, who monitors that.
- What you deliberately do not do, with the reason. That stops the same discussion coming back every quarter.
- Decision points: when you look at the list again, and on what results you continue, adjust or stop.
The decision point is where the roadmap earns its value: something new will be released next quarter as well, and you weigh it up instead of chasing the news.
We work with ninety days: short enough to stay sharp and long enough to learn something. Another period works too, as long as you attach a moment to look back.
How do you start with an AI strategy this month?
- Collect the ideas that already exist, including what people quietly do with free accounts.
- Put the four questions and the score next to each idea, together with the people who know the process.
- Pick one for ninety days, with an owner, a goal and a moment to look back.
- Write down what you are not doing and when you will look at it again.
- Take the decision in the management team, so that it does not rest on one enthusiast.
One management conversation in which you choose together is enough to begin.
When do you not need a strategy?
If your team mainly uses AI to write text, summarise or look things up, solid basic training and a clear agreement on what may go into a tool is often enough. Start with Getting started with AI. A strategy becomes useful when you have to choose between several processes, free up money or people, or have seen an earlier initiative fizzle out.
What a strategy costs or delivers depends heavily on your processes. There is no general figure for that, and a source that gives one deserves suspicion.
What is your first step?
Anyone who knows in advance where they want to go can weigh every new development without running after it. We think organisations that settle this now will have something many others lack a year from now: experience of what works in their own business, and a management team with one line.
Schedule the conversation this month, with the ideas that already exist, the four questions and someone who is allowed to say no.
Sources
- Exact, MKB Barometer 2026 (pdf, in Dutch), survey of 1,719 owners and managers of companies with 2 to 250 employees, first quarter of 2026
- CBS, Bedrijven gebruiken AI vaakst voor marketing of verkoop (in Dutch), 12 December 2025
