AI audit in retail: €3,000 direct, or from €2,500 with Bpifrance funding

By Tony Infantino · Published 4 August 2026

Why audit before launching anything

AI rarely arrives top-down in a retail chain. It arrives because a buyer tested an assistant on their range file, because a regional director saw a demo at a trade show, because IT received three vendor proposals in one quarter. The common outcome: four parallel initiatives, no measurement of gain, and an executive committee unable to arbitrate.

The published figures on gains do not settle that arbitration, they feed it. The Accenture France “Responsible AI” 2025 barometer puts the retail productivity gain at +17 %. McKinsey points to 20 to 30 % additional process efficiency in retail by 2026. These are sector averages: they establish that there is potential, without locating the gain inside your organisation.

That is what an audit is for: establishing which tasks, in which department, with which data, produce a measurable gain within 90 days.

What an AI audit must produce

An audit that ends with a trends presentation has not done its job. Four deliverables separate a report from a decision.

The task map by role. The survey covers time actually spent, gathered in interviews with the people doing the work, rather than processes as documented. In a retail chain, the gaps between the official process and actual practice sit exactly where AI has the most to gain.

The data and tools inventory. Which ERP, which POS, which PIM, and which spreadsheets keep the company running alongside the official systems. An appealing use case whose data lives in four unreconciled manual exports will cost three times the announced budget.

Ranked use cases. Each case positioned on two axes: estimated gain in hours or euros, and real technical difficulty. Prioritisation is only worth something if it rules cases out and says why.

The 90-day roadmap. A first workstream ready to launch, with its internal owner, its success criterion and its measurement point. The rest can stay as an indicative sequence.

Prices charged in France in 2026

The market has settled into three readable segments.

The flash diagnostic ranges from free to about €1,500. It is usually a two-day loss leader aimed at quick wins, often attached to an implementation proposal. Useful to start a conversation, insufficient to arbitrate a budget.

The full audit for an SME sits between €2,000 and €8,000, clustering around €3,000 to €5,000 for two to four calendar weeks and three to five days of actual work. This is where a three-day engagement scoped on one business function sits, at €3,000 excluding VAT.

The mid-cap or multi-site audit runs from €10,000 to €30,000, because it multiplies interviews, reconciles several systems and sometimes several countries.

These ranges apply to work paid for directly by the company. As soon as a public scheme enters the picture, the logic changes.

The three Bpifrance schemes that apply

Bpifrance subsidises diagnostics carried out by an expert the company names in its application. On an AI subject in a retail environment, three schemes apply, and the right one depends entirely on the company profile.

Diag Data IA: €6,000 net cost

Launched under the Osez l’IA plan, Diag Data IA covers 8 days of work by an approved expert, spread over a maximum of 3 months. The engagement is invoiced at €10,000 excluding VAT, Bpifrance covers 40 % of it since 17 June 2026, leaving €6,000 to the company.

The scope matches point for point what an audit must produce: technical and operational assessment, identification of concrete and applicable use cases, prioritisation by added value.

Eligibility: independent SME or mid-cap with 10 to 2,000 employees, revenue of at least €1M, more than one year of existence, not in financial difficulty, all sectors. The 10-employee floor rules out very light head office structures, but stays within reach for most chains operating their own stores. Legal form is irrelevant: only the thresholds filter.

Diag Axes d’Innovation: €6,500 net cost

Diag Axes d’Innovation, from the Innovation family (diaginno.bpifrance.fr), is invoiced at €13,000 excluding VAT and subsidised at 50 %, leaving €6,500 net. Its scope covers ideation and, above all, technical and techno-economic feasibility studies on a specific project.

It is the right vehicle when the question is no longer “where can AI help” but “is this specific project feasible, at what cost, with what architecture”. Typically: automating the flow of a supplier file into the ERP, or generating a catalogue’s product sheets from heterogeneous sources.

Conditions: independent company at least 5 years old, revenue of €1.25M to €50M for an SME or above €50M for a mid-cap, and no national innovation aid under the RDI regime in the past two years. That last criterion rules out companies that have recently consumed an innovation subsidy, which is worth checking before anything else.

Diag Scale-Up: €2,500 net cost

Diag Scale-Up is invoiced at €5,000 excluding VAT and subsidised at 50 %, leaving €2,500 net for 5 days over 4 months. It is the cheapest of the four routes, cheaper even than the directly paid audit.

It is reserved for independent startups under 8 years old, with under 250 FTE and under €10M in revenue. For a young chain, a growing pure player or a brand rolling out retail, this is the scheme to look at first.

One clarification on scope. Diag Scale-Up is not officially an AI audit: it covers scaling up, across three areas, market and positioning, offer and go-to-market, resources and scalability. Automation belongs there as a scalability lever, which is precisely the reasoning when a team of fifteen must absorb a tripling of volume without tripling headcount. Growth hacking, marketing strategy, market research and fundraising, however, are explicitly ineligible: those words have no place in the scoping note.

One further point: a company can benefit from Diag Scale-Up, or the former Diag Croissance, only once in its lifetime.

What these schemes imply

Three points are worth knowing before committing.

There is no retroactivity. The application is filed on the Bpifrance platform, the company names the expert, the contract is signed, and only then does the engagement start. Work already begun cannot be regularised.

The application is filed by the client company, not by the consultant. The consultant supplies the scoping note in the required format, with their registration number, the context, the phased plan and the financial table. The company files it, attaches its social and tax compliance certificates less than three months old, and tracks the case.

The expert must be approved, or obtain a waiver. When the expert brings the client themselves and that client names them, the review of the application is accelerated and a single-engagement waiver remains possible even when the expert pool is full.

Direct or funded: how to decide

Three criteria are enough.

Eligibility first. It takes five minutes to check and eliminates half the options. Under 8 years of existence points to Scale-Up. Over 10 employees and €1M in revenue opens Data IA. An already identified technical project and 5 years of existence point to Axes d’Innovation. A company that fits none of the grids should not spend three weeks exploring a scheme that will not go through.

The calendar next. A Bpifrance scheme adds several weeks between the decision and the first day of work: pre-qualification, filing, review, contract. If the subject must be settled before the next committee, the €3,000 direct audit is the only one that holds the deadline.

Scope last. On a single function and a single system, three days are enough and 8 diagnostic days would be oversized. Across several departments, several countries or a fragmented information system, the longer funded format is better calibrated, and costs the company twice the price of the short audit for nearly three times the volume of work.

Where training fits in

An audit produces a list of ranked workstreams. Executing them assumes the teams concerned know how to use the tool on their own files. In chains where the subject moves fast, the two engagements combine, in an order that depends on maturity.

When teams have never produced anything with an AI assistant, training first surfaces use cases nobody on the executive committee would have named. A buyer who has spent four hours running Claude Code on their own range file comes back with a far more precise list of automatable tasks than any audit interview produces.

When the scope is already set and the budget is tight, auditing first avoids training fifteen people on subjects that will not be retained.

On format: a four-hour session for 4 to 10 people, with the tool installed on their machines, three guided cases and one hour on a real case submitted in advance, costs €500 excluding VAT per participant. The tool subscription, in the order of €80 per month per user, is on top and remains the company’s responsibility.

Sources

• Diag Data IA, Bpifrance
• Diag Scale-Up, Bpifrance
• Innovation diagnostics, Bpifrance
• Becoming an approved Bpifrance diagnostics expert
• Accenture France “Responsible AI” 2025 barometer, retail productivity gain
• McKinsey, retail process efficiency by 2026