What funding is available for an AI project in a French SME?

What types of public support exist?

A tax credit reimburses money already spent: declarative, but auditable for years. Grants and subsidised support must be requested before work begins. Loans and repayable advances protect your cash flow without reducing the cost. Regional and European schemes run as calls for projects.

So qualify your project first — genuine research, a new product, or adoption of existing tools? — then pick the family. Most funding mistakes come from confusing the three.

Does the research tax credit (CIR) apply to AI?

The CIR covers R&D (fundamental, applied, experimental development): 30% of eligible expenses up to €100 million, 5% beyond, in mainland France. The base: research staff salaries, certain depreciation charges, plus a flat-rate operating allowance of 40% of staff costs. SMEs in the EU sense can request immediate repayment of the credit — a real cash-flow lever.

But is your project really R&D? Integrating an existing model, configuring an off-the-shelf tool or connecting an API is not research in the CIR sense. You need scientific or technical obstacles the state of the art cannot resolve, and an experimental approach documented as you go. Outsourced work is only eligible if the provider holds a CIR accreditation from the French research ministry — the list of accredited bodies is public. A poorly qualified claim earns nothing; it loses money in a tax audit.

Is the innovation tax credit (CII) more accessible?

The CII is reserved for SMEs. It covers the design of prototypes or pilot installations of new products: 20% of eligible expenses in mainland France, capped at €400,000 a year, for expenses incurred until 31 December 2027.

The decisive nuance: the product must be new to the market, outperforming what exists — not merely new to you. Improving your internal operations, however useful, does not qualify. If AI differentiates a product or service you sell, review the option with your tax adviser. The CIR/CII line is settled case by case, not by intuition.

What do the State and Bpifrance fund under “Osez l’IA”?

“Osez l’IA” is the national plan for AI adoption, run by the Directorate General for Enterprises and Bpifrance (France 2030) since July 2025. For an SME it is the most realistic entry point: no R&D required.

The pathway: a free online self-assessment; co-funded data and AI diagnostics delivered by referenced experts, with part of the cost left with you; then implementation support, in line with IA Booster France 2030. Bpifrance also lends — Prêt Boost IA, Prêt Boost Transformation numérique — with no guarantee or security over the company’s or the director’s assets. Conditions (company age, headcount, bank co-financing), rates and ceilings vary by variant and by year: check the current terms on bpifrance.fr or francenum.gouv.fr before you apply, not in a dated article.

What can you expect from the regions and the EU?

Each French region helps its SMEs its own way: digital vouchers, calls for projects, consulting aid. Criteria and calendars change every year. The right reflex: your region’s website, your chamber of commerce, and the business portals — les-aides.fr (chambers of commerce network), aides-entreprises.fr, mesaidespubliques.infogreffe.fr, or Place des Entreprises to reach an adviser.

The European Digital Innovation Hubs (EDIH) offer SMEs free or co-funded support: maturity assessments, test-before-invest, connections to expertise. Accessible, and under-used. The EIC Accelerator funds deeptech breakthroughs — grants can reach several million euros — with extreme selectivity. For adopting AI, it is the wrong door.

Which schemes are not for you?

Let us be blunt. Without genuine R&D, the CIR is not your lever: claiming it invites a tax reassessment, not a gain. Without a new product aimed at the market, neither is the CII. The young innovative company status (JEI), which reduces social charges, requires being under eight years old and devoting a minimum share of costs to R&D.

For the typical SME — making a process reliable, exploiting its data, equipping its teams — the realistic levers are subsidised support, regional aid and loans. Less spectacular, but free of tax risk. A well-scoped project on these instruments beats a fragile CIR claim.

What makes a funding application credible?

Three things.

A defined need. Funders read dozens of “we want to integrate AI” applications. What stands out: a precise use case, the data involved, a verifiable success criterion. Exactly what a successful AI project demands anyway.

Timing. A grant must be requested before work starts: expenses incurred before filing generally make the project ineligible. Tax credits are declared afterwards, but documented during the project, not three years later under audit pressure.

Traceability. Time records, invoices, recorded technical choices: proof the declared work happened. One caveat: the tax qualification (CIR or CII) commits your company. Secure it with your accountant or a tax lawyer, not with an article — this one included.