AI adoption in the Dutch SME stalls around month 3 because the first pilot works, but the rest of the organisation never joined in. Lasting adoption follows a fixed pattern: MT buy-in first, then broad training, then internal champions, then department-level implementation. The key is people, not tooling, plus the autonomy jump from 50% to 97%.
It almost always goes the same way. A small team builds an AI solution, runs a demo that works, everyone is excited. Three months later the same three people use the tool and the rest just do the work themselves again. That is not a technology problem. MIT research shows that 95% of GenAI pilots deliver no measurable impact on the profit and loss statement. The CBS AI Monitor 2024 names lack of experience (75%) as the biggest barrier in the Netherlands. SPAIK, founded in 2024 in Amsterdam by Jochem van Laren, Thijs Bongertman and Jan Bolle, has since done 35+ implementations and trained 700+ people, with an average positive ROI within 4 months. The difference between the 5% who make it work and the 95% who stall is not in the model or the tool, but in how you bring your organisation along. This article shows why adoption stalls around month 3 and which pattern does lead to organisation-wide, lasting adoption.
The month-3 pattern: why the pilot works and the rest does not follow
A pilot with a handful of enthusiasts is a fine start. Scaling means something else: dozens of people change how they work, provide input, check output, hold the same quality standard. If the tooling is already in place before you ask your people to join, they get lost in processes you impose on them. They feel no ownership and fall back on old habits.
On top of that comes the problem of edge cases. Many pilots run on the nice input: standard PDFs, predictable requests, roughly 80% of the work. The difficulty sits in the remaining 20%: unusual invoices, odd file formats, unique requests. If you do not test those during the pilot, you only meet them at scaling, and AI does not understand them. We worked this pattern out earlier in from AI pilot to organisation-wide adoption and in why 95% of AI implementations deliver no return.
People are essential: ownership, the enthusiast and the expert
Adoption does not scale on tooling, but on people. Scaling needs two roles. The enthusiast keeps the larger group engaged: builds support, answers questions, removes concerns. The expert is technically skilled and decides whether the output is good enough. With only an enthusiast you get no quality. With only an expert you lack adoption. You need both.
At Movir, a disability insurer and part of Nationale-Nederlanden with 350 to 400 employees, you see why that works. There, 350+ people were trained and 10 internal champions were developed. Sales processing went from an hour to a minute, 60 times faster, and content production went up tenfold. Not because the tool was special, but because there were owners who brought the department along. The fact that the same people both do the work and improve it is exactly why you do not toss it over the fence to a tool. We worked out that principle in your best people do work a computer can do.
The pattern that works: MT, broad training, champions, department implementation
At organisations where adoption does stick, the same pattern repeats in four steps.
1. MT buy-in first
Adoption starts at the top. Without a mandate from management, AI stays with a few enthusiasts. At Movir the journey started with the MT, with CEO Maurick Schellekens. At the fraud department of Nationale-Nederlanden, part of NN Group, it ran in the same order: commitment first, broad action only after. If you do not want to overhaul the whole organisation at once, you start small without your own IT department, as we describe in AI for SMEs: you do not need an IT department to start.
2. Broad training
Then you train broadly, not just the core team. At Movir that means 350+ people; at the NN fraud department it was 60 people who learned the AI fundamentals. SPAIK's AI Fundamentals training scores 9.1/10 across 700+ evaluations, runs 6 weeks with coaching and delivers 4+ hours of time savings per participant per week. That fits the AI literacy obligation from Article 4 of the EU AI Act, in force since February 2025. How you turn that literacy obligation into a concrete training programme is on our page about AI training for SMEs under the EU AI Act.
3. Appoint champions
From the broad group the champions emerge. At Movir there are 10. At NN, the fundamentals were followed by a Champions training, after which that group drew up guardrails and requirements for AI use. Champions are the enthusiasts and experts who carry adoption within their department once we are gone.
4. Department implementation
Only then do you implement per department, with real processes and real tools. At Movir that ran partly via Adobe Journey Optimizer. SPAIK works tool-agnostic and modular, with a stack like n8n, Supabase and Claude, so you build no vendor lock-in. "Movir is the model": training, capability building and then a long-term partnership, the land-and-expand pattern. The full approach to such an implementation is on our page about AI implementation in the SME.
The autonomy slider: why the jump from 50% to 97% breaks pilots
Think of AI autonomy as a slider from 0% (humans do everything) to 100% (AI does everything). During a pilot you sit around 50%: AI does the work, humans check everything. To scale to dozens of users you need to reach around 97%, the point where you trust AI to handle the vast majority correctly itself.
That jump from 50% to 97% is where pilots stall. You do not earn that trust with a good demo, but by testing edge cases and making quality provable. Without that evidence no one dares move the slider further, everyone keeps double-checking everything, and the time saved evaporates. Reditus, a marketing company of 10+ employees, did get that slider up: finding affiliates went from 16 hours to 30 minutes per client, 36 times faster and 95% saved, with ROI within 3 months. At Euphoria Mobility, a mobility SaaS with 60+ employees, a hackathon with 25 engineers produced 6 working prototypes in 2 days, of which 3 went into production, with 12 times ROI in 1 month. That only works if you earn the autonomy step by step.
Measurability: start with a baseline
You cannot steer adoption you do not measure. Start with a baseline: how long does a task take now, how much output do you deliver, what is the quality? Only then can you prove what AI delivers. At Movir sales processing went from an hour to a minute; at Reditus from 16 hours to 30 minutes per client. Those are not estimates after the fact, but differences against a starting point. Without a baseline you hear "it feels faster", and no MT steers on that. SPAIK has by now trained 700+ people across 50+ organisations and 10+ sectors, with an average satisfaction of 9.1/10; you only get numbers like that if you measure from day one.
Do it yourself or choose an AI implementation partner?
Many organisations doubt whether to run this themselves or hire an agency. Forrester shows that 83% of buyers are dissatisfied about something with their agency, often because the pitch team turns out not to be the day-to-day team. So watch for that. In our buyer guide you can read how to choose the right AI implementation partner for the SME, with the questions you ask up front.
Whatever the route: the pattern stays the same. MT buy-in, broad training, champions, department implementation, and an autonomy slider you earn step by step with a baseline in hand. To read more about the broader principles of AI adoption in the SME, you will find the full approach there.
This pillar belongs with two others: see also AI training for teams and AI strategy for management.
Curious how we make this article and which sources we use? Read our editorial policy.
