Bpifrance AI Funding in France: A Guide for Foreign SMEs

If your company operates in France, or has a French subsidiary, there's a public co-funding scheme worth knowing about before you commission an AI audit: Diag Data IA, run by Bpifrance, the French public investment bank. It's a France-specific mechanism with no direct equivalent abroad, so this guide starts by explaining what it is, who it's for, and what it actually costs, before walking through eligibility, real cost and how the process runs in 2026.

What is Diag Data IA, in one sentence?

It's a diagnostic co-funded by Bpifrance that helps an SME identify its priority AI use cases and their expected return on investment, delivered by an accredited expert over roughly 10 days spread across 3 months. The goal isn't a theoretical report, it's a roadmap leadership can actually execute.

Is my company eligible?

The criteria are deliberately broad. The essentials to check before filing:

• An SME or mid-sized company with 10 to 2,000 employees. The floor matters as much as the ceiling: a company with fewer than 10 employees doesn't qualify.
• At least €1m in revenue on a 12-month balance sheet.
• More than a year old, in any sector, anywhere in France including overseas territories.
• Registered in France with the French companies register (RCS).
• A diagnostic delivered by a Bpifrance-accredited expert, a non-negotiable condition for co-funding.

Companies in financial difficulty under EU rules are excluded.

One point often misunderstood: mid-sized companies (ETI) are eligible, contrary to what's sometimes assumed. The 2,000-employee ceiling does rule out the largest of them, since the mid-sized category in France runs up to 4,999 employees. Either way, the most common obstacle isn't eligibility, it's simply not knowing the scheme exists.

What makes a good candidate?

Meeting the administrative criteria isn't enough on its own. A diagnostic delivers its full value when the company has some usable data (in its ERP, CRM or files) and leadership ready to act on the findings. Conversely, a company that expects the diagnostic to decide everything for it, with no internal involvement, gets less out of it. The right posture is to arrive with a few concrete pain points already in mind, a time-consuming task, a decision made without good data, ready to put on the table.

How much does it really cost?

The reference cost is around €10,000 excl. VAT for the 10 days. After Bpifrance's co-funding, set at 40% since June 2026, the net cost to the company sits around €6,000 excl. VAT.

That rate has moved several times during 2026 (42%, then 25% in January, then 40% from 17 June). The safe approach is to confirm the net cost when you file, rather than relying on a fixed percentage.

That net cost compares to what an AI audit of similar scope would cost without any support. But the real question isn't the entry cost, it's the return. A well-run diagnostic identifies two or three initiatives whose expected gain clearly outweighs the spend, and above all it avoids investing at random in a poorly scoped AI project, which costs far more than the diagnostic itself.

How does the diagnostic actually run?

Over roughly 10 days spread across 3 months, in four clear stages.

• Scoping: goals, boundaries, stakeholders.
• Stocktake: available data, maturity level, constraints (including data control).
• Use cases and ROI: identifying and pricing the priority initiatives.
• Roadmap: a prioritised plan presented to leadership.

What concrete results come out of it?

A good diagnostic doesn't hand back a list of generic AI ideas, it hands back priced, prioritised initiatives specific to your business. Depending on the sector, recurring themes include:

• Manufacturing and trading: automated reading of supplier documents, demand and stock forecasting, pricing support.
• Services: qualifying inbound requests, generating quotes, drafting recurring documents.
• Support functions: automated reporting, natural-language search across management data, accounting pre-processing.

Each idea comes with an estimate of expected gain and effort, so leadership can decide with full information. That prioritisation is what separates a useful diagnostic from a catalogue of good ideas.

How do you prepare well for your diagnostic?

A few simple habits noticeably raise how much you get from the 10 days:

• Name an internal point of contact who's available, knows the processes, and can open up access.
• Gather your data sources (ERP, CRM, files) beforehand, even if imperfect.
• List your pain points: that's often where the best use cases are hiding.
• Involve leadership in the final presentation, since they're the ones who'll decide what happens next.

The more prepared the company arrives, the further the diagnostic goes into concrete territory instead of spending time rebuilding context from scratch.

Who delivers the diagnostic, and how do you start it?

The diagnostic is delivered by an accredited expert, and the company names that expert directly on the Bpifrance portal when filing its request.

If you've already identified your expert, you can name them directly. That's often what turns an intention into a project that actually launches.

That's precisely the role I play alongside companies: scoping the need, running the diagnostic, and delivering a roadmap you can act on, not a document that sits in a drawer.

What comes after the diagnostic?

The diagnostic is only an entry point. The initiatives it surfaces have their own funding paths: team training can run through your OPCO (a sector-based training funding body, see AI training for teams), and implementation can draw on regional funding schemes, covered in how France funds AI projects for SMEs. These schemes stack across a single roadmap, as detailed in how to fund an AI project. The point is to think about funding at the scoping stage, not after.