You can stop after any block, and you keep everything.
Five blocks. Four fixed-fee decisions. Each one ends in something you can see working — not a slide, not a promise. If we are the wrong partner, you learn it early and cheaply.
What each block leaves you with
Every block is scoped and priced up front. At the end of it you decide whether to continue — and either way, everything produced is yours to keep.
Why this looks different from what you have seen before
You see it working, or it does not pass
Every fixed-fee decision is on a working system, on your data, in front of you. Slide decks are not a form of acceptance we accept.
Numbers you can defend to a board
30% is a hypothesis until the baseline says otherwise. We measure the current process first; only then does a target go into a contract.
The gap we are not dressing up
Where something rests on an assumption we have not verified, we say so — and name the block that closes it. You will not find surprises after signature.
Predictable running cost
Model usage is capped, reported and alerted, with the same discipline you already have around cloud spend. No end-of-quarter surprise on the invoice.
Everything ends up in your repository
Source, models, configuration and artefacts land in your version control from day one — not at the end. Changing supplier is a decision, not an ordeal.
Your team owns it at the end
Rollout is not accepted until your team makes four representative changes on their own, with us watching. Handover is the test, not the promise.
What ends up in your hands
By the end of Rollout, everything below is in your repository, under your account and available to your team — without us. No licence to renew. No hosted service to be cut off from.
| What | Why it matters to you |
|---|---|
| The working software | Source and configuration in your repository. Your team can build it, deploy it, change it — the same way they do with any other production system you own. |
| The dataset and the accuracy measure | The labelled data used to prove the system works, together with the test that produced the numbers. You can re-run it any time — and disagree with us on the result. |
| The runbooks and the training | What to do when a step fails, how to switch the process back to manual, who your team calls at 2am. Exercised during Rollout, not written after. |
| The compliance pack | The section of the Verfahrensdokumentation, the works-council annex, the AI Act classification, the DPIA fragments — produced during delivery, ready for your reviewers. |
Handover is a test, not a promise
Rollout is not accepted until your team performs four representative changes on their own, with us present as observers. If any one cannot be done, the block does not close and the gap is ours to fix.
Move a threshold, change how a case is routed, adjust a contract-class parameter — and put it live through your normal deployment path.
Add a step, change how a decision is made, adjust an SLA — and decide what happens to cases already in flight.
Carry a new field end-to-end — from the systems that produce it, through the process, into the systems that consume it.
Add a system the process did not talk to before, using the documented pattern and the tests we leave with your team.
The point of the test is not that we go away. It is that you can choose whether we stay — from strength, not because you have to.
Book an architecture review.
45 minutes. One process, one obligation you have to meet. We tell you honestly what it would take — and whether we are the right partner to do it.
Book a review