The discovery is already done
On site, one real piece of work traced end to end, the numbers built with your finance team. The proposal rests on what we found, not on what we assume.
AI agents in production · Paris
AI agents that are still running a year later.
Every operations team has seen an AI demo. Very few have one running inside their operation a year later, on their own data, under their own rules, with someone accountable for it. We start with one real piece of work, a priced first step, and a gate that can say stop.
01 · What it costs you
What does it cost you today?
Nothing we build will add revenue. It gives you back money you already lose or tie up: what gets written off, the costs you absorb because nobody caught them in time, the cash that sits a month longer than it should, the purchasing you pay too much for because the knowledge of what is acceptable lives in three people's heads. We build the case from your finance team's own numbers, line by line, and we separate what they already record from what we estimate. The estimate is what the first step measures.
02 · What it is
What is it?
Not a new system. A layer that reads your systems of record and writes back into the data platform you already have. Rules where rules work, a model where judgement over a lot of information is needed, and a person on every exception, with the reasoning attached. Every decision is explained, every override is attributed to someone, and the audit trail exists from the first day, not after the first finding. The technology costs little to run. The cost is people's time, and we size it before you commit to it.
03 · The first step
What is the first step, and what is the whole bill?
The first step is a mapping: one to two weeks, at a fixed fee, inside your environment. We trace one real piece of work end to end with the people who do it, from the request to the invoice, and we come back with the jobs ranked by return, readiness and effort, the success criteria we agree to be held to, a reference set of real cases with the right answers, and a costed design. It is yours whether or not you continue. Everything after it is a separate decision at a gate, priced on what the mapping finds, and the full estimate for the whole programme is on the table from the start. Then two or three jobs go live in four to six weeks. Then we run it, with the numbers every month.
04 · Your teams' time
How much of your teams' time?
Days, not weeks. Over the mapping, a few days in total from the team that does the work, one session with IT, about an hour with each system owner, and one contact in finance for the baseline. Nothing changes for the operators until something is live, and when it is, they are trained on the screen they use every morning.
05 · Ownership
Who owns it?
Everything built for you is yours: the code deployed in your estate, the data, the catalogue, the model, the evaluation set, the run log, the regression suite. It runs in your environment, with the model you have approved. If any of our own tooling ends up inside the system, you get a perpetual, royalty-free licence to it, with the runbooks. Nothing you run depends on us. If we ever step away, your team, or any integrator, runs it. We put that in writing.
06 · Why us
Why us, and not an integrator?
Three things an integrator does not bring.
A working system to look at
There is a system to look at before you commit, running on real cases from the mapping. Not a deck, not a pilot that postpones the decision.
Accountable at a fixed price
We are accountable for the result at a fixed price, with a gate that tells you to stop if the money is not there.
An integrator bills you for the first two and never offers the third.
07 · Year three
Who is still here in year three?
We would like to be, running the system at a fixed fee and bringing the next jobs onto the same foundation. Month after month, new cases, new rules, a model version that moved: that is the hard part, and it is the part we built the firm for. And you are never stuck if we are not.