How do you compare to similar companies?

"Are we behind?" is one of the two questions that comes up in every first conversation. It is also a question most vendors answer with a statistic from a global survey that has nothing to do with a Dutch company of eighty people.

Here are the actual Dutch numbers, from CBS, for 2025.

AI use by company size

A few things worth reading carefully here.

The size gradient is steep. A company of 250-plus is well over twice as likely to be using AI as one of 10 to 50. That is not because large companies are smarter about it; it is mostly because they have people whose job it is to try things, which a company of thirty does not.

If you have between 50 and 250 people and you use AI somewhere, you are in the majority, but only just. If you do not, you are in a large minority rather than an outlier. Either way, "everyone is already doing this" is not an accurate description of the Dutch market in 2025, and neither is "nobody is doing this yet."

A row of blocks of differing height. The gap between size classes is larger than the gap between sectors.

The number that matters more than the headline

CBS also asked companies not using AI why not. The answers, as a percentage of all companies in that size class, are where the useful information sits.

Read the first and last rows together, because that is the finding.

Lack of in-house expertise is the most cited barrier, at every company size, by a wide margin. It is cited three to four times more often than cost. The thing standing between Dutch companies and AI is not budget, and it is not that the technology is too expensive. It is that there is nobody inside who knows how to do it.

Almost nobody says it is not useful. One percent, identically across all three size classes. Whatever companies are struggling with, it is not scepticism about whether AI could do something for them.

Those two facts together describe a market where the demand is real, the budget is available, and the constraint is capability. Which is, admittedly, a convenient finding for a company that builds AI capability, so check the source rather than taking our word for it; the table number is at the bottom of this page and the CBS data is public.

What this does not tell you

Be careful with what "used AI" means in a national statistic. It covers a company where one department uses a text generator and a company that has a model in production making daily decisions. Those are not the same thing, and the 46 percent contains a great many of the former.

So the honest reading is not "46 percent of my peers have AI capability." It is closer to "46 percent have started, most of them shallowly." If you have not started, you are later than average. If you have started but nothing runs without a specific enthusiast present, you are exactly average, which is a less comfortable finding than being behind because it feels like progress.

The comparison that actually helps

Benchmarks against a national average are useful for one thing: settling the question of whether you are unusual. They are close to useless for deciding what to do next, because your next step depends on your operation and not on the sector mean.

A more useful comparison, and one you can do yourself this week: take the three processes where you spend the most expensive hours, and ask whether a competitor could plausibly do them faster than you today. That question has a specific answer, and the answer changes what you build. "We are at 46 percent like everyone else" does not.

The six questions in Je AI-positie bepalen zonder consultant do that job better than any benchmark can.

About this page

All figures come from CBS StatLine table 86119NED, "ICT-gebruik bij bedrijven; bedrijfstak en bedrijfsgrootte, 2025", covering companies with ten or more employees, retrieved directly from the CBS open data API and filtered to the whole-economy total rather than a single sector. The reason percentages are shares of all companies in that size class, not of the non-users only, which is how CBS reports them. We note explicitly that the expertise-gap finding is convenient for us; the table is public and can be checked. Written by Radical's own team; no client data was used.