Why the licence is still being paid for somewhere
Last summer you bought a Copilot licence for two people. You expected to win back time. By October, only one of them used it now and then; by January, no one. The takeaway sits in your head: AI is nice but not for our kind of work. That is half true.
You have tried it twice. ChatGPT for one adviser. A marketing tool with built-in AI. Maybe a free month of Notion AI or Claude. And yet it does work for that mortgage adviser in your network. Not because he writes better prompts, but because he set up the pilot differently.
What the figures show, with context
The Searchlab report AI in the Netherlands Statistics 2026 shows that a large majority of Dutch SMBs have now deployed AI in some form. At the same time, nearly half of business owners say they do not know how to use AI effectively. Source context: this is a study on Dutch SMBs broadly, not specifically on professional services, but you will probably recognise the gap between "having" and "using effectively".
The summary of those figures is this: adoption says nothing about impact. Buying a licence is not running a project. Acquiring a tool is not improving a process.
Why pilots stall at the individual level
Three reasons we see structurally.
One, ownership. One person uses it, leaves or gets too busy, and it disappears. There is no one to guard or develop it. The pilot lives on the enthusiast's laptop, not in the company.
Two, place in the process. The pilot sits next to your work, not inside it. An adviser opens ChatGPT in a second tab when he thinks of it, forgets it half the time, and the saving never shows up in the lead time. What you do not measure, you do not feel.
Three, integrations. The tool does not know your customer data. For every prompt, the adviser has to paste context in by hand. That works three times, then he gives up. Not because he is lazy, but because it is a touch more work than it is worth.
In short, the pilot ends where it began, namely with one person, next to the process, without a data integration.
What the difference is between a pilot and a working system
A pilot answers questions ad hoc. A working system sits in your process.
A concrete example at a mortgage firm. The pilot is: I use ChatGPT to summarise my customer advice. The working system is: every financing request is automatically read, categorised and pre-filled with the right fields in our CRM, then goes to the right adviser based on complexity. The same process, but embedded.
For an estate agency with several branches. The pilot is: two agents use Claude for their copy. The working system is: social posts from all four branches run centrally through an approval flow where AI checks the house style and the boss approves in a single dashboard. No more four WhatsApp groups, no more different voices.
For a security firm. The pilot is: the planner uses AI for the roster. The working system is: incidents and patrols automatically produce a structured report that goes to clients without a team leader having to retype it twice afterwards.
The difference comes down to three words. Owner, process, integration. Who is responsible, where it sits in our workflow, how it reaches our data.
Why an audit makes the difference
The AI audit forces three questions for every opportunity we find.
Which three processes together cost 50 to 100 hours per month and lend themselves best to automation. Who becomes the owner of the first implementation. Which integrations are needed so this does not become another island.
If none of those three questions is clearly answered, every engagement stays a pilot. If all three are clear, the transition from pilot to working system is predictable. No wishful thinking, just work.
What you get back after fourteen days
An AI opportunity report with 8 to 15 opportunities, scored on impact and feasibility. A process inventory, as-is, where the time, frustration and repetition currently sit. A roadmap in which the top 3 to 5 opportunities are worked out with order and dependencies. A closing presentation with the owner-director and team. A follow-up proposal with clear choices for implementation or a retainer.
No 80 pages of PowerPoint. But concrete action points you can put on the management agenda tomorrow.
The order that does work
Audit first. Then implementation of your first use case in five to eight weeks, with a clear owner and integrations built in. Then an AIOS retainer for further development, so it does not stop at that one use case. No Big Bang, but three steps that reinforce each other.
The mortgage adviser next to you is no genius. He simply did not make the mistake of thinking a licence was enough.
The first step
If you recognise that your first pilot is sitting somewhere in a licence overview without anyone doing anything with it, that is no reason to conclude AI does not work. It is a reason to start over, with the order that does work.
Fourteen days later, you have something concrete in hand. Not a pilot.
