From first call to an agent in production.
Many AI projects stall at the prototype stage. We plan yours backward from the day the agent handles real cases for your team, then we track the hours it saves. Every step is below, with what you receive at each one.
01
The five rules every project follows
- Scoped for production from day one. Each project starts from a task your team does every week, and we agree with you on the indicators that will judge the agent once it runs.
- Scope and price fixed before we start. After the mapping week you get one page with the task, the expected gain, the date and the price. Nothing is billed by the hour.
- Your team reviews the agent every week. The people who do the task today try it on their own cases, and they decide when it is ready to go live.
- You own everything. The code, the prompts, the configuration, the documentation and the test results belong to you. You can take them in-house or to another provider whenever you want.
- You see the gain every week. Time saved, error rate and volume handled are tracked from the first day, on a dashboard anyone on your team can read.
02
The six weeks of a first agent
A typical first agent goes live in four to eight weeks, depending on the systems involved. This is the six-week version.
| When | What happens | What you receive |
|---|---|---|
| Week 1 | We watch the work, count it and pick the task with the clearest gain. | A one-page scope, the expected gain, a fixed price and a date. |
| Weeks 2 and 3 | The agent is connected to your tools with the least access it needs. | A working agent in a test environment, used by your team. |
| Weeks 4 and 5 | We replay a set of your real past cases and measure the results. | A test report listing the cases it got right and those it passed to a person, with the reason for each. |
| Week 6 | It goes live with approval on sensitive actions. | The agent in production, with its logs, its dashboard and a hand-over session for your team. |
| After | We monitor it, fix what drifts and choose the next task with you. | A monthly report of hours saved and cases handled. |
03
What we refuse to build
We turn these projects down even when the budget is there, because they end badly for the client and for us.
- A project with no measurable gain. If we cannot say what it gives back, we do not start.
- An agent that takes sensitive actions with no human approval.
- Anything that sends your data to be used for training someone else’s model.
- Cold-calling or mass messaging robots.
- A system you could not run or change without us.
04
What you have at the end
- The agent, in production, connected to your tools and used by your team.
- The code and the configuration, in your repository or ours, handed over on request.
- The documentation, written for the people who supervise the agent day to day, with a technical part for your developers.
- The test report, on your own cases, and the dashboard of time saved.
- A trained team, able to supervise it and to ask for the next one.
It all starts with a free 30-minute assessment of one task you want to hand over.
Book the free assessment