Introducing AI into your company — a sober guide.

The worst first question is "where can we use AI?" — it leads to a solution in search of a problem. You already know the better question: which work eats hours every week without anyone deciding anything while doing it? Retyping, summarising, triaging, looking things up.

A single such spot is enough to start. Whoever has carried the first case through cleanly can judge the second — before that, every AI roadmap is decoration.

Before anyone creates an account, you need a line that fits on one page: which data may go into a tool that runs outside the company, and which absolutely may not? Without that line, one of two things happens — nobody uses the tools for fear of getting it wrong, or everybody uses them and nobody thinks about it.

Since February 2025 this also includes the EU AI Act: companies that use AI must make sure their people have a sufficient level of AI literacy. No reason to panic — but a good occasion to introduce the topic properly instead of on the side.

Nearly every AI adoption is a mix of three things. First: using ready-made tools like ChatGPT, Copilot or Claude directly. The fastest route, sensible for almost anything made of text — what it mainly requires is that people can handle the tools, which is exactly what training is for.

Second: integrating AI into the software you already run. The route with the largest lasting effect, because the help arrives where the work happens — in the CRM, the inbox, the ERP. And third: automating workflows. Where a task follows clear rules, it often needs no AI at all — just software that does it reliably.

An adoption that arrives looking like a cost-cutting programme gets quietly undermined by exactly the people whose cooperation it needs. So the rule is: whoever does the work today still makes the decisions afterwards — what disappears is the retyping, not the person.

In practice that means involving the people affected early, letting them practise on their real tasks, and taking scepticism seriously. A sceptical clerk who has experienced a real benefit convinces the rest of the team better than any presentation.

One initiative, a few weeks, a checkable result: does the task get done measurably faster or better? If yes, the next initiative follows. If no, the loss was small — which is also a usable result.

What actually makes adoptions fail is rarely the technology: it is missing rules, missing everyday skill, and projects that impress as prototypes and then belong to nobody. All three are solvable — but only if they are planned for from the start.