where dutch smbs really stand

In 2025, 33 percent of Dutch companies with ten or more people used at least one AI technology, against 23 percent in 2024 and 14 percent in 2023, according to Statistics Netherlands (CBS, 2026). Yet every measurement that asks about depth comes out far lower: in research by software company Exact, only 6 percent of small business owners say AI is an integral part of how they run the company (Exact, 2025). “Using AI” has become an ever lower bar; structural embedding remains rare. Anyone who wants to know where Dutch small business really stands has to keep those two curves apart.

how many dutch companies use ai?

A third. In the CBS report Digitalisering en kenniseconomie 2025, its annual survey of digitalization and the knowledge economy, 33 percent of companies with ten or more people used at least one of eight AI technologies surveyed (CBS, 2026). A year earlier that was 22.7 percent, the year before that 14.0 percent (CBS, 2025). A tripling in two years, in official statistics. That is a pace you rarely see.

Behind that single percentage sit three different worlds. In early 2026, CBS published a separate report on micro companies with two to nine people for the first time: there, AI use in 2025 stood at 13.8 percent. SMBs with ten to 249 people sat at 29.8 percent, large companies from 250 people at 66.2 percent (CBS, 2026). Within the ten-plus class the spread is wide too: in 2024, 17.8 percent of companies with ten to nineteen people used AI, against 59.2 percent of companies with five hundred or more (CBS, 2025). So with every adoption figure, look at the size class first, or you end up comparing a five-man installation firm with a bank.

Internationally the Netherlands does well. Eurostat measured an EU average of 20.0 percent for 2025 among companies with ten or more people, against 13.5 percent in 2024 and 7.7 percent in 2021; among large European companies it was 55 percent (Eurostat, 2025). In the measurement over 2024 the Netherlands ranked sixth in the EU, behind Denmark among others; for 2025 that ranking has not been confirmed yet. And one more figure that tilts the picture: of all Dutch companies using AI in 2024, 65 percent were small companies with ten to fifty people (CBS, 2025). In absolute numbers, most AI users are small companies, even though the percentage within the class of large companies is much higher. That same CBS article, for the record, adds its own warning about the correlation between AI use and company size: whether AI users are bigger and stronger in revenue because of AI cannot be established from these figures.

the bar to count as an ai user drops every year

Behind the CBS figure sits a survey question: does your company use at least one of eight AI technologies, such as text generation, image recognition or speech processing? One application is enough to count. A company where someone drafts quotes with a chatbot lands in the same category as a company that runs its entire planning on AI. The statistic measures whether AI has entered the building. It says little about what it does there.

Ask about depth and you get a different number. Software company Exact uses a ladder in its MKB Barometer, its recurring survey of Dutch small business: 43 percent of small businesses are actively exploring the possibilities, 23 percent are working on implementation and 6 percent have made AI an integral part of the business (Exact, 2025). That is commercial market research from a software vendor, with its own definition and sample, so treat it as an indication and never as official statistics. But the direction matches what the government finds qualitatively: research agency Dialogic, commissioned by the Dutch Ministry of Economic Affairs, concluded that AI in small business is rarely structurally embedded and usually gets stuck at loose applications such as text and image generation (Dialogic, 2025).

This is also where the common framing that “small business is lagging” needs correcting. On experimentation, Dutch small business barely lags at all; a third of the ten-plus companies now does something with AI and the growth curve is steeper than the EU average. The problem sits one layer deeper, at the step from loose use to process, and that step is hard everywhere, including at organizations with their own innovation department. Internationally the same echo sounds: MIT researchers reported in 2025 that 95 percent of generative AI pilots deliver no measurable business result, though that research covers mostly large, international organizations and says nothing specific about Dutch small business (Fortune, 2025). Why that figure needs careful reading is covered in the 95 percent myth. For this essay the lesson is simpler: the difference between small and large sits in the adoption figures, far less in the embedding.

why do the ai numbers contradict each other?

Because definitions, populations and years get lost along the way. A concrete example: Exact’s 6 percent “integrally embedded” was read by ictmagazine.nl as general AI use in 2024, after which a growth leap “from 6 percent to 70 percent now” could be claimed. Two incomparable measurements, glued together into a spectacular growth story, which then gets cited by others in turn. That is how numbers start echoing. Claims like “67 percent SMB adoption” and “95 percent of Dutch organizations run AI programs” circulate as well: vendor or content-marketing surveys with their own definitions, every one of them, and you can never put them next to CBS figures as if they measured the same thing.

Even within CBS you have to watch which table you pick up. In the autumn of 2025 two publications on AI use per sector appeared that seem to clash at first sight. The September news release, on companies with ten or more people, puts construction at 8.9 percent in 2024 and hospitality at 9.0 percent (CBS, 2025). A December publication with a broader business population and a different question arrives at around 4 and 3 percent for the same year (CBS, 2025). Both are correct, within their own definition. The ranking, though, is robust: information and communication far ahead, construction, hospitality and transport at the back. For the niche trades where many of my readers work, installation, cleaning, events, no official adoption figure exists at all; the figures by SBI sector, the Dutch standard industry classification, are too coarse for that, and so I am not going to extrapolate them here either.

Three control questions prevent most accidents: which companies does it cover (from which size class up), what exactly counts as AI use, and who paid for the research.

what holds smbs back on ai?

Knowledge and experience, far more often than money or distrust. Of the Dutch companies that considered AI but decided against it, 74.6 percent named lack of experience as the reason, across all size classes and sectors (CBS, 2025). Among micro companies that is 71.6 percent (CBS, 2026), and EU-wide Eurostat finds nearly the same: 71 percent name lack of relevant expertise as the main barrier (Eurostat, 2025). Three independent measurements landing this close together makes this one of the best-substantiated facts in the whole dossier.

The OECD sees the same pattern outside Europe: in six G7 countries, more than half of small and medium-sized companies name lack of knowledge about generative AI as the biggest barrier, and of the SMBs already using generative AI, fewer than 30 percent offer training for it (OECD, 2025). For the Netherlands, Dialogic adds costs, the organization’s level of digitalization and the availability of data (Dialogic, 2025). The picture is consistent: the barrier is learnable. That is good news, because a learnable barrier is something you can work on; “too expensive” or “our customers don’t want it” far less so.

what should a small business do with these numbers?

One thing: pick one process and embed AI in it completely, with everything that belongs to that. The adoption statistic is not a race you are losing. As a micro company, any AI use at all already puts you in a minority of 13.8 percent, and structural embedding is rare everywhere, by every available measurement.

What you definitely do not need to do: write a thirty-page AI strategy, try ten tools at once, or panic over headlines about exploding adoption. Those headlines measure the low bar, and two thirds of the companies in your size class still stand at zero.

What does work, in six steps:

1. Pick one process that recurs often and is measurable: drafting quotes, answering customer questions, planning, following up on invoices.
2. Measure the baseline for two weeks. How much time does it take, where does it go wrong, what gets left undone.
3. Deploy one tool and describe the process around it: who checks the output, what happens on an error, where does the customer data live and is it allowed to live there.
4. Train the people who work with it. Lack of experience is the biggest barrier in every measurement, and training is exactly what happens at fewer than 30 percent of users (OECD, 2025); that is where your head start lives.
5. Evaluate after six to eight weeks against your baseline, in hours and euros.
6. Only when this runs without you standing next to it do you take on the next process.

Finish these six steps and you are further along than nearly every “AI user” in the statistics. That is not a big promise; it is what the numbers on the gap between use and embedding literally say.

My position: the adoption percentage is the least interesting statistic in this whole dossier. It measures whether you cleared the lowest bar, and that bar drops further every year. The question that counts is whether there is one process in your company that demonstrably runs better since AI became part of it, with someone who feels ownership of it. In my own work for small service companies I see exactly what Dialogic describes: the chatbot is already there, the process around it is missing.