Our method
An AI agent does a task in place of your teams: the data entry, the chasing, the formatting. Three stages to get one to work: 10 days to put a figure on it, 40 to take it live, 10 per team for adoption. You may take only one. The point is not that you need us for a long time, it is that you stop needing us.
Three situations that come back in almost every mid-sized company we meet. The third is the most expensive.
A spreadsheet, copy and paste, and the follow-up living in two people's heads. Messages go out when someone finds an hour. The channel stops the moment the workload rises.
Our answer
We frame the target, we build the sourcing and campaign chain, and every send stays yours to approve.
A bid takes three days. A deliverable, a week. Paperwork is checked by eye and data is retyped from one tool into another. You handle no more volume than you manage to hire for.
Our answer
We measure what the process really costs, we put a figure on the gain, and we only build the cases that stand up to the arithmetic.
A demo went down well in a meeting, then nothing. No real data, no screen, nobody to use it, and nobody whose job it was.
Our answer
We pick the pilot up where it stopped: the collection, the integration, the screen, the monitoring, and a first scope genuinely opened up.
Prioritization
Every task is scored on two axes. We build the ones in the top left, and we rule the rest out in writing, with the reason.
How often the task repeats, per day, per week, per month.
The time spent and the money at stake each time it runs. A rare but costly task counts as much as a frequent one.
Access to the data, the state of the tools, the number of rules to write. This is what sets the timeline and the price.
A case whose gain does not cover the cost of building and maintaining it is ruled out, and the reason goes in the report.
What we measure
Without a starting figure, a gain cannot be proven. Four indicators, recorded before and after.
What the process really takes, measured with the team.
Between a file arriving and leaving, rush jobs included.
How many files go all the way through without a human stepping in.
How many results get redone by hand. The least watched, and the most expensive.
Share of time actually billed
The same calculation at your company, with your figures. If the gain does not cover the cost of building and maintaining it, the case is dropped and the reason goes in the report. The last line is the only one that matters: recovering the whole gap does not happen, recovering a third of it is already a good result. This calculation promises no gain. It tells you whether a case is worth building.
The method
Three stages, and you may take only one. The scope and the dates are set at framing, and they do not move afterwards without written agreement.
01 — Temps 1
10jours
We look at how the work really gets done, and we calculate what it costs.
What we do
What you receive
02 — Temps 2
40jours
We go and get the data, we build the AI agent and its screen, then we open it up in stages.
What we do
What you receive
03 — Temps 3
10jours
We train your teams, then we leave. The point is that you no longer need us.
What we do
What you receive
This pace holds for a first use case whose data is accessible. That is the most common situation, not the only one.
What can stretch it, and we say so at framing
The first week is spent at your own desk. What we write afterwards, we have seen.
On siteThe interviews are held with the people who do the work, not only with their managers. We describe the process as it actually runs, workarounds included.
At the costingVolume handled, time per file, rework rate, regulatory constraints. A use case without those four figures does not go into the roadmap.
LiveA first scope opened, watched, corrected, then widened. You can roll back at every step, and the rollback is documented before anything opens.
De la donnée jusqu’à l’écran
Link 3 is the one everyone sells: the AI agent. On its own it stays a script nobody opens. That is the most common reason a pilot never goes into service.
Go and get the data where it lives.
Your tools, your documents, your shared mailboxes, and public sources where they exist. Data that has to be retyped by hand to feed an AI agent does not last three weeks.
Clean it and reconcile it before using it.
Deduplication, one common format, matching by identifier. The rule we never work around: we match on an identifier or a domain, never on a name, because namesakes manufacture errors that make no noise.
The work itself, with a closed scope.
Every AI agent says what it does and what it does not do, and points to the next one for the rest. An agent that does everything is an agent you can neither test nor correct.
The screen your teams use it from.
An AI agent with no interface stays a script only its author knows how to launch. Your teams have to be able to trigger it, see what it proposes, correct it and approve it, from a tool they already open.
Know what is running, what it costs, and how to stop it.
A log per run, the cost per AI agent, and a documented rollback before go-live. That is the difference between a service and a demo.
Our commitments
And it does not change along the way without an amendment signed by you.
Nothing is written into your tools or sent outside without your approval. The stopping point is in the specification.
Documents, code, documentation. You can carry on with another provider.
A diagnostic that spares you a pointless project has paid for itself.
Price
We do not publish a rate because there is not one single rate: every scope is quoted once it is written down, and the quote is firm. Here are the three variables that move it, so you know which side you fall on.
A fixed price, agreed at signature. No variable share.
For you if you want a fixed budget and no surprises.
A reduced fixed share, and a share paid only if the agreed indicators move.
Solid: due whatever happens. Hatched: paid on results.
For you if you want us committed to the result, not to the delivery.
These are two standard schedules. Yours is set in the quote, at the same time as the scope, and it does not move afterwards without a signed amendment.
One use case or five. This is the variable that weighs most, and you set it at framing.
An AI agent that reads a mailbox does not cost what an agent that writes into your ERP costs.
If the data is already clean and accessible, collecting it is quick. If not, it is half the work.
The quote is firm once the scope is written. No time and materials, no daily rate billing, no overrun without a signed amendment.
Once it is live
Exceptions, duplicates, missing documents, month-end peaks: none of that shows up in a demo. All of it shows up the moment you open it. That is why we open it department by department.
Contact
Describe a situation that costs you time. We will tell you honestly whether AI is the right answer.
Tell us about your project