Bpifrance Diag Data IA: a subsidised AI audit for companies in France

By Jérôme Denis, AI & data governance consultant · 8 min read · 12 August 2026

Bpifrance is the French public investment bank, the state-backed body that lends to and co-funds French companies. Among its instruments is a family of subsidised short consulting engagements called Diags, and one of them, Diag Data IA, pays part of the cost of having someone audit your data and tell you what artificial intelligence could realistically do with it. If you run a French subsidiary and have been told the group will not fund an exploratory study, this is the mechanism that gets it funded anyway.

What Bpifrance is and why this exists

The reasoning behind the scheme is that most companies sit on years of accumulated operational data they have never examined, and that jumping straight to buying an AI tool without examining it produces expensive disappointment. The subsidy exists to make the boring diagnostic step cheap enough that companies actually do it. Whether public money should be spent this way is a fair question. That it is available, and that most eligible companies never claim it, is the practical point.

The audit is delivered by an independent consultant listed by Bpifrance, not by Bpifrance itself. You choose the consultant from its directory, which means the quality of the outcome depends on that choice more than on the scheme.

Who qualifies

Four conditions, and the first one is the one international readers usually get wrong.

Establishment in France. The support attaches to the French entity. A German or American parent does not disqualify a French subsidiary. What it does affect is the size calculation, because EU state aid rules require thresholds to be assessed at group level where ownership links exist. A twenty-person French subsidiary of a ten-thousand-person group is not an SME for this purpose.

Size. Under 250 employees to count as an SME (PME), under 5,000 for a mid-cap (ETI). Turnover under 50 million euros or balance sheet total under 43 million for the SME category. These are the standard EU definitions, so if you have qualified for an EU SME scheme before, the arithmetic is familiar.

Financial health. Companies in insolvency, administration or judicial reorganisation are excluded outright.

Enough data to audit. A listed consultant confirms this in a scoping call. This is not a formality: if there is nothing usable, the ten days get spent designing a data collection plan rather than an AI project, which is useful but is not what you thought you were buying.

Sector coverage is broad, spanning manufacturing, business services, wholesale and retail, subject to data protection rules being respected in whatever is proposed.

The money, precisely

The engagement is priced flat at 10,000 euros excluding VAT for ten days of consultant time. Bpifrance covers 40%, which is 4,000 euros. Your net cost is 6,000 euros excluding VAT.

The payment mechanism is worth understanding because it protects cash flow. It works on a tiers-payant basis: you pay the consultant only your 60% share, and Bpifrance settles its 40% directly with the consultant once the deliverables are validated. You never advance the full amount and never wait for a reimbursement.

VAT is charged on the full 10,000 euros, so 2,000 euros at the standard rate, which you recover through your normal returns. And the subsidy falls under the EU de minimis regime, which caps total public aid to one undertaking over a rolling three-year period. If your French entity has drawn on other public support recently, check the running total before you count on this.

What ten days actually covers

The engagement runs over four to six weeks in three phases.

1. Scoping and bringing the leadership up to speed

   The consultant works through the company's structure, its objectives over the next few years and its main operational processes, and gets the management team to a level where they can evaluate proposals rather than be sold to. This phase is worth more than it sounds in companies where the executive committee has never had to assess an AI claim.

2. Technical audit of the data

   Quality, regulatory position, accessibility and security of what actually exists across the systems. The output is a map of where usable data sits and where the gaps are. Most of the unpleasant surprises land here.

3. Use cases and a costed roadmap

   Concrete applications matched to the business, prioritised by estimated return against technical feasibility, with implementation costs, a schedule and a plan for bringing the team's skills up.

A worked example

A precision parts manufacturer supplying aerospace, 110 employees, sees quality varying on its main production line and wants to use machine sensor data to anticipate tool failures.

The consultant pulls twenty-four months of output from the industrial controllers. Temperature and vibration readings turn out to have been captured continuously and stored in incompatible formats, which is why nobody had ever correlated them. During the audit phase the consultant cleans a representative sample and builds a trial model showing that 18% of major line stoppages could have been predicted.

The roadmap proposes centralising the data on a single server, at an estimated 14,000 euros of hardware and software, with payback under ten months from reduced unplanned downtime and scrap. What the ten days produced was not the prediction system. It was the evidence that the prediction system is worth building and roughly what it will cost, which is the document that gets a capital request approved.

The honest limits

It is a study, not a build. You get a roadmap; every euro of implementation is yours. Companies that read "AI diagnostic" as "AI project" are consistently the unhappy ones.

It depends on your people being available. If the operations managers and the systems administrator cannot free time during the audit weeks, the consultant works from documentation and assumptions, and the conclusions degrade accordingly. Ten days of consultant time requires perhaps three days of internal time, and that needs to be booked in advance.

It cannot fix an empty cupboard. Where there is no usable history, the ten days go on designing collection rather than analysis. Still valuable, just not what was on the brochure.

And GDPR sets real boundaries on what can be done with customer data. The consultant will flag the constraints, but a genuine legal review is a separate exercise that this engagement does not include.

Applying

Start by picking a consultant from the Bpifrance directory and agreeing the scope with them. Then file the application on the Bpifrance platform with the consultant's detailed quote, your recent statutory accounts and a declaration of de minimis aid already received. Bpifrance reviews it over a few weeks and issues a formal approval, which is what authorises the engagement to start. At the end the consultant delivers the report to you and to Bpifrance, and you settle your 6,000 euro share directly.

One practical note: do not start the work before the approval arrives. Aid schemes almost universally refuse to fund expenditure committed before the decision, and this one is no exception.