08ai implementation
AI implementation is not a project. It is work that has to finish.
Most AI plans die between the pilot and the payroll. The demo works, the deck lands, and six months later nobody uses it. Implementation is the part where the software starts doing the work, inside the systems you already have, with someone accountable for the output.
What does an AI implementation partner actually do?
Four things, in order. Find the work. Build the thing that does it. Put controls around it. Stay on it after go-live.
The first is the one that gets skipped. Before anything is built we map a workflow end to end: who touches it, which systems it crosses, where it stalls, and what a good outcome looks like. That map is what makes the build cheap. Without it you are automating a process nobody has described.
Then we build a digital employee: one role, one job, working inside your existing tools. Not a platform you have to migrate to. See the seven steps for what each stage delivers.
What does an AI implementation cost?
A first digital employee starts at 3,000 euro setup and 850 euro per month. That covers the scan, the build, the integrations and the control layer, then the running cost of keeping it correct.
The honest comparison is not against other software. It is against the salary of the hours it takes over. Half a day a week across a team of ten is a quarter of a person.
The full breakdown, including where it goes up, is on the pricing page.
How long before something actually works?
Weeks, not quarters. A first employee typically runs in a controlled way inside four to eight weeks of the scan.
What makes it slower is never the model. It is access. Waiting for a system owner to approve a connection is the single most common delay, and it is worth arranging before the build starts.
The go-live is deliberately dull. It runs alongside the human it helps, it proposes rather than acts, and the autonomy dial moves only after you have read its work.
Is this about the EU AI Act?
No. This page is about implementing AI in a company, not about complying with the regulation.
The two do meet. Everything we build logs what it did, keeps a human on critical actions, and works inside scoped permissions, which is the same discipline the regulation asks for. But we are not a compliance advisor and we do not sell an AI Act readiness scan.
If control is the question you came with, how you stay in control is the page you want.
When is an AI implementation a bad idea?
When you cannot describe the work out loud. If every case turns out to be an exception, it was never a process, and no amount of implementation will make it one.
When nobody is named to read the output. An employee whose work nobody checks is abandoned within a month.
When the reason is that the board wants to be doing AI. That produces a pilot, and a pilot is not an implementation. We would rather say so on the first call than bill you for finding out.
Where to go next
AI automation · AI for smaller companies · What it connects to · Start with an AI Scan
Written from AImplement implementations at Dutch companies of 10 to 200 people.