How I price: the four formats, their duration, and what drives the number
How I price an AI project, without a magic rate card: the four formats, their size in days, and what really drives the number. You get the ballpark from the first exchange, for free. The firm price is set in the quote, once the scope is written down.
First exchange is free and no-obligation, reply within 24 hours.
Where I stand on pricing
I don't publish a rate card, because it would be false.
Two projects of the same type don't carry the same budget: it depends on the state of your data, your integrations and your reliability bar. A single published price would either inflate the simple cases or understate the complex ones.
What I can give you instead: the formats I work in, what pushes the number up or down, and a ballpark from the first exchange, for free. And if AI doesn't pay for itself at your place, I say so before invoicing you anything.
The formats
Four ways to work together.
Durations are in days and they are firm. The default entry point is the audit: we only go further if the numbers justify it.
AI Audit
Start here4days
One square = one day
An honest lay of the land: where AI would pay off, where it would just burn money, and in what order to tackle it.
4 daysPOC Sprint
12days
One square = one day
A prototype that proves the value on your real data, with success criteria set up front.
12 daysProduction Build
20+days
One square = one dayat minimum
A reliable, monitored, documented system in production that your teams keep using.
20 days (at minimum)Embedded AI architect
10+days · renewable
One square = one dayat minimum
The AI expertise your team is missing, on an agreed volume with an end date set in the contract.
10 days (at minimum)
After delivery your teams take over, which is the point. If you still need me, it's on time spent, under a framework agreement signed at delivery. I don't sell subscriptions.
What sets the price
Same format, two very different budgets.
The five levers, by order of impact. You can place your own case before the first exchange even happens: check which side of each line you are on.
- 01
The state of your data
What brings it down
Clean, current and in one place: the project moves fast.
What pushes it up
Scattered or in need of cleaning: it stretches everything. This is the first hidden cost of an AI project.
- 02
The scope
What brings it down
One targeted use case, one team, one tool. Start small and useful, widen afterwards.
What pushes it up
A platform touching ten tools and five teams from day one.
- 03
The integrations
What brings it down
A tool with a clean API. With the MCP standard, the connection is written once and reused.
What pushes it up
An in-house piece of software with no documentation, to be reverse-engineered before it can be wired at all.
- 04
The reliability bar
What brings it down
Internal use, where a mistake is caught by hand with no consequence.
What pushes it up
A system in direct contact with your customers: evals, guardrails and red-teaming. The security level has a price, and it is a choice.
- 05
The hosting
What brings it down
A standard EU cloud, already running at your place.
What pushes it up
An on-prem deployment with private models. Sovereignty has a real cost, worth paying when your data demands it.
Always included
What is never an option.
Whatever the format. This is what separates a demo from a system you actually run.
Evals and guardrails
Quality is measured before every release. Nothing ships blind.
Observability
You see what the system does, what it costs, and you're warned when it drifts.
Documentation and handover
The system is yours, your teams can take it over.
No dependency
Documented architecture, open standards, no black box and no lock-in.
One person
From scoping to handover. I work solo, and that's a choice: no agency layer to bill.
How I bill
Fixed price or time and materials, by how sharp the scope is.
Two modes, one default. Either way, I warn you before, not after.
| The criterion | Fixed priceMy default | Time and materials |
|---|---|---|
| When | The scope is sharp: an audit, a scoped POC, a well-defined build. | The scope moves: embedded work, exploration, continuous evolution. |
| Who carries the overrun | Me. You know the price up front and it does not move. | You, but bounded and visible: you set the priorities week by week. |
| What protects you | Any scope change goes through an amendment signed before the spend, never a surprise invoice at the end. | A committed volume and an end date written into the contract, extended by amendment if the work runs over. |
Public funding
Depending on your size and your project, public schemes may cut the bill. Bpifrance's Diag Data IA co-funds data and AI diagnostics for small and mid-sized companies; other regional or sector schemes exist case by case.
I guarantee no eligibility: these schemes have their own criteria and calendars, and the decision isn't mine. What I can do is provide the technical scoping an application needs. Check with Bpifrance or your region for your own situation.
A ballpark, in thirty seconds
How much AI could save you, before we even talk.
Three numbers are enough for an estimate. You get a ballpark of the yearly gains, something to put on the table internally to make the case. It's an estimate, not a quote: the price of a system depends on your case, and we scope it on the specifics.
data entry, document search, email triage, reporting
around 5,850 EUR of time freed per year
that's about 130 hours a year handed back to your team for higher-value work.
A deliberately conservative figure. It counts time only, not the errors avoided or the delays cut, which often weigh more. It also doesn't subtract the setup cost: that's exactly what we scope together.
Straight answers
What people ask before we talk budget.
How much does an AI audit cost?
It depends on the size of the scope and the state of your data, but it's the most predictable format: a few days of work, a clear report and a costed roadmap. I give you a firm ballpark from the first exchange. And the audit stands on its own: it commits you to nothing further.
How much does an AI agent in production cost?
The price mostly comes down to the range of actions, the number of tools to wire up and the reliability you need. An agent focused on one task costs a fraction of a multi-agent orchestration. I size the ballpark from the first exchange, then architect it so the token bill stays under control.
Why is there no public rate card?
Because it would be false. Two projects of the same type don't carry the same budget, depending on your data, your integrations and your reliability bar. A fixed grid would force me to either inflate the simple cases or underestimate the complex ones. I'd rather quote honestly on your real case.
Fixed price or time and materials, which should I pick?
Fixed price when the scope is clear: you know the number up front, I carry the overrun risk. Time and materials when the scope moves, like embedding with your team or continuous enhancements: you keep control of priorities, with an agreed volume and an end date in the contract. I default to fixed price whenever the scoping allows.
What varies the price the most?
The state of your data and the reliability bar, well ahead of everything else. Clean data plus an internal tool that tolerates the odd error is fast and cheap. Data to clean up plus a secure customer-facing system is a different budget. We talk it through honestly from the first exchange.
Is the first exchange paid?
No. The first exchange is free and no-obligation, and I reply within 24 hours. We scope your need and I give you a ballpark. If AI doesn't make sense for you, I tell you right there, before any billing.
Contact
Ready to go from demo to production?
Reply within 24 hours · first conversation free, no strings attached.