Measured before we build, checked before it goes live
The same seven steps for every AI implementation. You can stop after any of them.
AI transformation is a problem of governance as much as technology. Our method puts the measurement, ownership and human approval in place first, so the AI that reaches production stays there.
- 01Intro callA short call, no charge. We learn how your finance and operations run and whether there is a fit.
- 02BaselineIn the Sprint (2 weeks) we measure how the work runs today in your own data: invoice lead time, cost per supplier invoice, days to close.
- 03Choose the workflowAI use cases are sized in euros, with effort, risk and EU AI Act class. You decide what to build first, or whether to build at all.
- 04Shadow modeOur engineers build the AI agent, and it runs alongside your current process. Nothing it produces is used until the numbers hold up.
- 05Go live, people approvingEvery invoice and posting that matters is approved by a person before it is final.
- 06Day-90 reviewResults against your baseline, in your ledger, with you in the room.
- 07Run and improveOptional. Monitoring, updates when systems or models change, and a monthly review.
What you can hold us to
- A baseline before we build
- Shadow mode before go-live
- People approving material outputs
- Runbooks and a one-page AI Act record per system
- A defined scope per package, agreed before we start
We do not publish client results until the client has measured and approved them. See our services.
Questions about the method
Why do so many AI implementations fail?
Most stall for reasons that have little to do with the model: no baseline to prove value, no owner after go-live, no governance, and pilots picked for novelty rather than return. Every step of our method closes one of those gaps.
Why start with a baseline?
Without a measurement of how the work runs today, nobody can say whether AI helped. The baseline is taken from your own ledger and systems, so the day-90 result is a comparison, not an opinion.
What is shadow mode?
The agent does the work alongside your team, but its output is not used. We compare it with what your people actually did and go live only when the numbers hold up.
Who approves what the agent produces?
A person at your company approves every invoice and posting that matters before it is final. Which outputs count as material is agreed with you before go-live.
How long does a build take?
A Pilot Build takes about 6 weeks per workflow. It starts after the 2 weeks Sprint.
Do you publish client results?
Only when the client has measured them and approved them. Until then we do not publish any.