Eight days. €10,000. A written answer.

This is not an invented price. The format — 8 days of work, €10,000 excl. VAT (about US$11,500) — is taken directly from Diag Data IA, a public assessment framework operated by Bpifrance, France's public investment bank, open to companies of 10 to 2,000 FTE. One precision, for honesty's sake: the subsidy that framework provides applies only when the assessment is carried out by an expert referenced by Bpifrance — which we are not. We modeled our format on that public benchmark for a simple reason: it gives a verifiable frame and an established market price. It is not a promise of a subsidy — there is no subsidy here, and the price is the price.

Day 1 is always spent on auditing the quality and accessibility of your data. That is a deliberate choice: feasibility is decided there, not in the choice of model — an AI system built on unusable data fails before it is even deployed. If the audit reveals that the data is not ready, the sprint that would follow is postponed and documented, never launched on a fragile base.

The rest of the engagement maps one precise business process — not “your whole digital transformation” — puts a number on its bottlenecks, and ranks the identified use cases by their real added value to your revenue. The process in question often touches cash flow, document reconciliation, or the triage of incoming requests. The final deliverable is an operational action plan: the process mapped, what it costs today, a forecast of the expected gain, and the roadmap of the system to build — a workable project-management document, not a brochure.

Many decision-makers approach AI the way they were taught to approach the digital revolution: a full digital-maturity audit, a three-year digital strategy, a change-management programme, a digital culture to spread, business intelligence and new use cases to explore — the offer from the big integrators and consultancies is missing nothing, except a narrow starting point. That is the reflex the large firms encourage — we set it out in detail on our approach page. Our conviction runs the other way: to succeed at a transformation, you first have to succeed at one business process, measurably, before generalising anything.

So the assessment looks at a narrow perimeter, and looks at it thoroughly. Where does this process's data live — an on-premise ERP, cloud computing, shared spreadsheets, a cabinet of paper files? Who touches it, how often, with what tools? What actually costs hours every week? A company director walks away with costed answers on that perimeter — not with a generic maturity score nobody knows what to do with on Monday morning.

Many firms offer a bespoke AI audit with no external reference. Our conviction: anchoring the assessment on a recognised public framework — rather than inventing our own grid — gives the client a verifiable reference and, for companies that go through a Bpifrance-referenced expert, real funding — which, again, is not us. Bpifrance is not the only institutional actor in this field: France Num also publishes its own guides on adopting generative AI for French SMEs and mid-market companies; we build on that public corpus rather than reinvent it.

In practice, the engagement is run by one person with a dual data-and-AI skill set, not a generalist consultant. It is that dual skill that makes it possible to assess a real integration potential, not a theoretical one — and to tell a genuinely prioritisable use case from one that stays a good idea on paper.

Two possible outcomes, never improvised after the fact. If this step confirms a sufficiently costed process and usable data: a fixed-price sprint builds the identified system, with the same team that ran the assessment — never a hand-off to another outfit. If the assessment does not confirm it: the written deliverable (mapped process, bottlenecks, roadmap) remains useful in itself, with no obligation to continue.

Once the system is in production, a monthly retainer covers maintenance and evolution — an agent-based AI system is more temperamental than classic software; it is priced accordingly. To place this step within our general approach, or to see who it is aimed at first (mid-market companies, regulated professions).

A consulting deliverable is worth what you can do with it without its author. Ours comes in three parts, written to be used by your project lead — with or without us. First the mapping: the process as it actually runs, who touches what, where the hours go. Then the costing: what this process costs today, in hours and in errors, and the forecast gain if it is systematised. Finally the roadmap: what to build, in what order, with what data prerequisites — precise enough for another provider to pick it up.

That last point is not a nicety. An assessment that only exists to sell the next step is a pitch in disguise; ours is designed to stay useful even if the answer is no.

We turn down engagements, and it is not a pose: it is a written checklist, applied to every incoming request. One: you are a high-margin regulated profession or a mid-market company — not a micro-business with no budget, not a large group with an endless buying cycle. Two: the request is about one precise process, not “our transformation”. Three: the assessment can be sold and delivered before any heavier commitment. Four: nothing in the engagement puts us head-to-head with an established SaaS that already does the job well. Five: we can cite proof directly relevant to your case.

A box that does not tick is not an automatic refusal — it is a stop signal to ask the question honestly, rather than say yes out of commercial reflex. It is, as far as we know, the most effective protection there is against engagements that end badly.