Back to the blog

What Dutch companies use AI for: the CBS figures for 2025

· By Bee Flow · 6 min read

One in six Dutch companies used AI in 2025, according to Statistics Netherlands. Marketing and sales come first, large companies are far ahead, and lack of experience is the main hurdle. Three sensible first uses for a company of 10 to 50 people.

Four office buildings in a row from small to tall, with more windows lit in amber in each larger building

On 12 December 2025 Statistics Netherlands (CBS) published new figures on how Dutch companies use AI (CBS: Companies using AI above all for marketing or sales). One in six companies with two or more employed persons used AI in 2025. That is 17 percent, against 13 percent in 2024 and 8 percent in 2023, so the share doubled in two years. The 2025 figures are preliminary.

For a company of 10 to 50 people that total says little. What matters more is what companies use AI for, and how far small and large companies are apart. This piece looks at both and ends with three first uses we would suggest at that size.

The bigger the company, the more AI

CBS breaks the figures down by the number of employed persons. Share of companies using AI:

Employed persons 2023 2024 2025
2 to 9 7% 11% 14%
10 to 49 11% 19% 27%
50 to 249 20% 31% 45%
250 or more 42% 53% 66%

So a little over a quarter of companies with 10 to 49 people use AI, and two in three of the largest ones. According to CBS the fastest growth is among companies with 50 to 249 people, from 20 to 45 percent in two years.

Sectors differ widely too. In information and communication 54 percent use AI. In hospitality, construction and transport and storage it is 6 to 7 percent.

Marketing and sales come first

Of the companies that use AI, 35 percent use it for marketing or sales. Business administration follows at 32 percent, and CBS gives risk assessment, selecting job applicants, basic translations and planning as examples. Then R&D or innovation at 25 percent, production or service processes at 18 percent, accountancy or financial administration at 17 percent, ICT security at 11 percent and logistics at 4 percent.

Among the technologies, text mining grew the most: from 4 percent of all companies in 2023 to 12 percent in 2025. Speech recognition tripled, from 2 to 6 percent. Natural language generation, having AI write text, stands at 8 percent.

What small companies do differently

The StatLine table behind the release (CBS, table 86116NED, in Dutch) shows per size class what AI is used for, as a share of all companies in that class. CBS notes with the table that its weighting method was improved this year and that figures may therefore differ slightly from earlier years. In a few places the table comes out one point higher than the news release, for example 28 instead of 27 percent for 10 to 49 employed persons. For comparisons across years we follow the news release, for the breakdown the table.

At companies with 10 to 49 employed persons, 11 percent use AI for marketing or sales and the same 11 percent for administrative processes. R&D follows at 7 percent, accounting at 6 percent and ICT security at 4 percent.

At 250 or more the weight sits elsewhere. There 37 percent use AI for administrative processes, 30 percent for ICT security and 24 percent for marketing or sales. Speech recognition is at 38 percent in those large companies and at 9 percent at 10 to 49 people.

How they get their AI differs as well. Of companies with 10 to 49 people, 15 percent use commercial AI software as supplied, and 5 percent use AI developed by their own staff. At 250 or more those figures are 50 and 19 percent. The smaller company mostly buys something ready-made.

Three quarters of all companies use no AI and have not considered it. Of those that considered it and did not start, 73 percent give lack of experience as the reason. 49 percent mention privacy, and 42 percent legal consequences such as liability for damage. Only 15 percent say AI is not useful to them.

That last figure tells us the most. Companies that looked at AI and stopped rarely did so because they expected nothing from it. The hurdle is how to start and what happens to their data.

Three objects side by side: a fountain pen on a sheet of paper, a stack of papers with a paper clip and a magnifying glass on an open binder

Three first uses for a company of 10 to 50 people

The choice below is ours. It follows where the CBS figures show the most use, and where a mistake stays small. This is not legal advice.

1. Text for marketing and sales

This is where the largest group already starts: a first draft of a newsletter, a product description, the covering text for a quote or a LinkedIn post. It saves time quickly, and you usually spot a mistake before anything goes out.

Watch two things. Do not paste customer details into a chat service when you do not know where the text ends up. A name, an address or a phone number is personal data, which comes with rules (Business.gov.nl: protection of personal data). The second point: AI sometimes makes things up, such as a delivery time, a discount or a certification you do not have. Have someone who knows the product read every text before it is sent.

2. Administrative work that does not decide about people

Summarising a long document, writing up the notes of a meeting, translating a letter, making a first schedule. It is the kind of administrative task that, according to CBS, 32 percent of AI users already do.

In the same category CBS also lists selecting job applicants. We would stay well away from that for a first step. The EU AI Act counts AI for recruitment and selection, such as CV-sorting software, among the high-risk uses (European Commission: regulatory framework for AI), with heavier requirements than for ordinary tools. Our rule of thumb: let AI help with texts and schedules, and leave decisions about people to people.

3. Asking questions of your own documents

Text mining is the technology that grew the most in the figures. In a company of thirty people it can look like this: a colleague asks what the warranty terms for a product are, or how a complaint is handled, and gets an answer from the manuals, contracts and work instructions you already have. That also touches the largest obstacle in the survey, lack of experience. New staff find their own way through what the company already knows.

Watch three things here. An answer should name its source, so someone can check which document it came from. Old versions of documents give old answers, so tidy up first. And decide beforehand who may see what: a salary overview does not belong in the same collection as the product catalogue.

What Bee Flow does about it

Bee Flow is an AI workspace you run on your own server or in a European cloud. You choose the language model yourself: a cloud model, a model hosted in the EU, or one on your own hardware. Privacy Shield looks for personal data such as names and IBANs in a message before the text reaches the model, and by default holds the message back. A knowledge base answers questions from your own documents with a reference to the source document, and a colleague only gets answers from the knowledge bases they are allowed to see. How that works is on our privacy and security page. If you would like to see whether it fits your company, get in touch.

#ai-at-work #privacy