Budget is no longer what keeps most organisations from using AI. Know-how is. In our analysis of Eurostat data, 70.3% of EU firms that considered AI and did not use it in 2025 named a lack of expertise, while cost fell from 45.8% to 38.4% in two years. The fix is to borrow the know-how for the first build while your own people learn it, and to treat the legal question one use case at a time.
This post is for the IT manager, CIO, informatiemanager or security officer at a Dutch or Belgian organisation of 250 to 5,000 people. Crux Digits builds AI systems alongside your ICT partner or IT team and hands over code you own; we are not an ICT partner.
Is cost still what stops AI in large organisations?
No: in the Netherlands only 21.6% of firms that considered AI named cost, 20th of 21 EU countries, against 78.4% naming expertise.
That is good news. The barriers that need capital or a new system are shrinking across the EU. Between 2023 and 2025, incompatibility with existing systems fell 6.8 points and data availability or quality fell 4.3 points. Over the same period the share of EU firms using AI rose from 8.1% to 19.9%, according to the same study.
Large firms are further along. Among EU firms with 250 or more staff, only 9.8% of those that considered AI called it not useful, half the 19.2% of firms with 10 to 49 staff. Your board is not asking whether AI is worth it. It is asking who in the building knows how to do it, and that is a question you can answer within a budget year.
Should we hire AI specialists or work with a partner?
Do both, in order: a partner builds the first system with two of your people, then you hire for what you will run.
Our decision rule is simple. Hire for work you will do every week for years: owning the product, keeping the data pipelines from your ERP, CRM or document management system healthy, and judging output quality in your own domain. Borrow for work you do once or rarely: the architecture of the first system, setting up evaluation, and the method for classifying use cases under the AI Act.
Hiring first has a known failure mode. A new AI specialist arrives before anyone has decided which process to change, and spends the first months writing a strategy. A first build answers the hiring question for you, because it shows which skills you lacked and which your ICT partner already covers.
Is the legal uncertainty around AI as big as it feels?
It is smaller: the AI Act sorts systems into four risk levels, and the high-risk rules apply to eight named areas in Annex III.
The European Commission's AI Act overview lists the levels as unacceptable, high, transparency and minimal risk, with spam filters as a minimal-risk example. Annex III names the high-risk areas: biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration, and justice and democratic processes. Recruitment screening is high-risk under point 4(a). A tool that drafts replies from your own manuals is not on that list; it is a transparency question at most, under Article 50.
The dates are fixed too. The Digital Omnibus on AI entered into force on 27 July 2026 and moved the Annex III high-risk rules to 2 December 2027. Our Omnibus explainer covers what else changed.
The data says the doubt came before the rules. Among large EU firms, unclear legal consequences rose from 37.0% in 2023 to 51.4% in the 2024 survey, answered before the AI Act entered into force, and held at 51.2% in 2025. In the Netherlands the national figure of 37.9% hides a size effect: it is 50.0% among firms with 250 or more staff. A run through our EU AI Act risk checker per use case turns a general worry into a short written answer.
Can the AI Act's literacy duty close part of the expertise gap?
Yes: Article 4 already obliges deployers to take measures that build AI literacy among staff, so the training budget has a legal basis.
The duty has applied since 2 February 2025, per the Commission. The Omnibus version requires measures that take into account staff's technical knowledge, experience and the context of use, and says it does not require a guaranteed level of AI literacy for any individual. The Autoriteit Persoonsgegevens published a practical starting guide, Aan de slag met AI-geletterdheid, on 30 January 2025.
Here is the information most guides miss: the same measure can serve compliance and capability. A generic e-learning for 2,000 staff meets the letter of Article 4 and closes none of the expertise gap. Spend the deeper part of the budget on the six to ten people who will own, test and explain your first AI system, and give everyone else a short module on what the system does and where its limits are.
Which five moves close the gap in the next twelve months?
Five moves, each tied to a finding in the study: fund know-how, name owners, classify early, aim training, and brief your ICT partner.
1. Fund one process, not a platform. Cost is the least cited barrier in the Netherlands, so the constraint is attention. Pick one process with a clear owner and a measurable outcome, such as answering technical questions from your own documentation, and fund the know-how to build it.
2. Name two internal owners from day one: a business owner from the process and a technical owner from IT. Write the transfer into the contract: code in your own repository, architecture and data-flow documentation, the evaluation set with its results, and a runbook. Expertise is the top barrier in 20 of 21 EU countries; this is how it moves in-house.
3. Classify each use case before building. Record the AI Act risk level and run the DPIA screening against the Autoriteit Persoonsgegevens list or, in Belgium, the criteria in the EDPB guide. Put your privacy officer in the first meeting: large Dutch firms rank third in the EU on privacy concerns, at 61.4%.
4. Aim the Article 4 measures at the owners first, then at users, as described above.
5. Brief your ICT partner before the build, not after. Incompatibility with existing systems is cited by only 33.7% of Dutch firms, 20th of 21, but integration still happens in their landscape. Agree who runs identity, network access and monitoring, and who changes the AI system itself.
What does 2027 look like if the trend holds?
If the 2023 to 2025 direction holds, know-how, not budget, will separate organisations in 2027.
Two dates frame the year. The Eurostat survey for 2026 is the first one answered after the Omnibus, and the Annex III high-risk rules apply from 2 December 2027. Organisations that classify their use cases now will meet that date with paperwork already written.
The opening is real. In 2025, 16.0% of large EU firms had considered AI without using it yet; in the Netherlands the figure was 12.8%. Large Belgian firms are the most engaged: 85.1% either use AI or have considered it, fourth in the EU, against 80.4% in the Netherlands. Belgian legal uncertainty also jumped from 36.9% to 57.9% in one year, so a Flemish IT director should expect the legal review to be the longer conversation internally.
Where does Crux Digits fit?
We supply the know-how for the first build and leave it behind. We design and build one AI system at a time next to your ICT partner or IT team, work with the two owners you name, document the AI Act classification and the input for your DPIA, and hand over the code, the evaluation set and the runbook. We do not run your infrastructure, helpdesk or wider IT estate. The full figures are in our study of AI barriers.
How we work with organisations in the Netherlands: AI consulting in the Netherlands.
Last updated: 28 October 2026.
A consultant tells you where AI pays off; Crux Digits also builds it. A fixed price per step, one named expert, from Utrecht.
AI consultancy in the Netherlands →
