Most guides on this subject open with "80% of AI projects fail" and never say where that number comes from. This page uses the actual figures — CBS and Eurostat — and publishes what each phase costs and how long it runs. Three phases, three fixed prices, and you can stop after any one of them.
Last updated: 11 June 2026
AI implementation in the Netherlands runs in three phases: an audit, a proof of concept, and production. At Crux Digits those are fixed at €2,500 (1–2 weeks), €20,000 (4–6 weeks) and from €50,000 (from 8–12 weeks), excluding VAT. A realistic first project therefore takes three to five months end to end. Per Eurostat, 33.2% of Dutch enterprises used AI in 2025 against an EU-27 average of 20.0% — and CBS found the single biggest barrier for firms that considered AI and did not adopt it was lack of experience, at 74.6%.
Almost every article on this topic recycles an unsourced failure statistic. The measured picture is more useful, and more encouraging.
Eurostat put Dutch enterprise AI use at 33.2% in 2025, well above the EU-27 average of 20.0%, but behind Denmark (42.0%), Finland (37.8%), Sweden (35.0%) and Belgium (34.5%). CBS measured 22.7% of Dutch firms with 10 or more staff using AI in 2024, up nearly nine percentage points in a single year. Note that Eurostat recorded a break in the Dutch time series in 2025, so the two figures are not directly comparable year on year.
The size gap is the part that matters for an SME. CBS found 59.2% of firms with 500+ employees using AI against 17.8% of firms with 10–19 employees. By sector, information and communication leads at 58.0% while construction (8.9%) and hospitality (9.0%) trail.
| Measure | Figure | Source |
|---|---|---|
| Dutch enterprises using AI (2025) | 33.2% | Eurostat |
| EU-27 average (2025) | 20.0% | Eurostat |
| Belgium (2025) | 34.5% — ahead of NL | Eurostat |
| Dutch firms 10+ staff using AI (2024) | 22.7% | CBS |
| Firms 500+ staff vs 10–19 staff | 59.2% vs 17.8% | CBS |
| Top barrier: lack of experience | 74.6% | CBS |
That 74.6% figure deserves attention, because it reframes the whole problem. Among Dutch firms that considered AI and decided against it, the dominant reason was not cost, not privacy and not technology — it was not knowing how. CBS found this held across every size class and every sector.
This is why a first project should be scoped to produce knowledge as well as software. An audit that tells you your data is not ready is a successful audit. A proof of concept that proves a use case does not pay back has done its job for €20,000 instead of failing at €200,000.
Most Dutch implementation frameworks have five phases. Ours has three, because two of the usual five are sales stages rather than work. Each phase has a fixed price, a real duration, and a deliverable you keep whether or not you continue.
| Phase | What it produces | Fixed price | Duration |
|---|---|---|---|
| 1 · Audit & strategy | Ranked use cases, data-readiness verdict, roadmap you own | €2,500 | 1–2 weeks |
| 2 · Proof of concept | A working system on your real data, plus a go/no-go | €20,000 | 4–6 weeks |
| 3 · Production | Integrated, monitored, documented — and yours | from €50,000 | from 8–12 weeks |
Three to five months from first conversation to a system running in production, assuming the audit finds a viable use case and your data is reachable. The variable that moves this most is not model complexity — it is how many of your systems the AI has to touch. One integration is quick; five is a different project.
Comparable Dutch providers publish 12–24 weeks for a first phase, which is consistent. Where estimates stretch to nine months or more, it is usually an enterprise with procurement and security review in the path rather than a technical constraint.
Four failure modes account for most of it, and none are exotic. Starting too broad — an AI strategy covering nine processes instead of one working system. Data that cannot be reached — it exists, but it is locked in a system with no export. No owner — nobody inside the business is accountable for the outcome, so it drifts. And no baseline — without a nulmeting before you start, you cannot prove the thing worked, so it quietly gets defunded.
The fix for all four is the same: pick one process, measure it before you touch it, and give one person ownership.
The technical work is the same either side of the border; the constraints are not. Dutch employers introducing a system that assesses or monitors staff generally need works council consent under article 27 of the Works Councils Act. Belgian employers face CAO nr. 39, which requires informing and consulting the works council before new technology with collective impact is introduced.
Funding differs too: the Dutch WBSO scheme can offset part of the R&D wage cost on qualifying custom AI development, and has no Belgian equivalent. And on the regulatory side the EU AI Act binds identically in both countries — though the high-risk deadline moved to December 2027 under Regulation (EU) 2026/1744.
Three fixed prices, all excluding VAT. An audit is €2,500, a proof of concept €20,000, and production starts at €50,000. Across the wider Dutch market, AI consultant rates run roughly €125–€250 per hour, and most agencies publish no price at all — so comparing usually means a sales call first.
Three to five months end to end for a first project. The audit takes 1–2 weeks, a proof of concept 4–6 weeks, and production from 8–12 weeks. The biggest variable is how many of your systems the AI must connect to — one integration is fast, five is a different project entirely.
Eurostat put it at 33.2% in 2025. That is against an EU-27 average of 20.0%. CBS measured 22.7% of firms with 10 or more staff in 2024. The Netherlands sits above the European average but behind Denmark, Finland, Sweden and Belgium.
Usually for organisational reasons, not technical ones. The four common causes are starting too broad, data that exists but cannot be reached, nobody inside the business owning the outcome, and no baseline measurement so success cannot be proven. CBS found the top barrier to adoption overall was simply lack of experience, at 74.6%.
With one process that is high-volume, rule-heavy and currently manual. Invoice processing, quote follow-up, customer-service triage and document handling are the usual first candidates for Dutch SMEs. Start with an audit that ranks options by payback rather than picking a tool first.
Start with the free 60-second scan, or book a consultation and we will tell you honestly whether a first project is worth running yet.
Book a free consultation →