What the September 2026 report presents
In its report of September 4, 2026, the Ministry of the Economy announced the deployment of 1,000 AI Data Diagnostics. The presentation targets small businesss and ETIs with 10 to 2,000 employees, with a service of eight man-days covered at 40% by France 2030. These elements describe the official announcement; they do not constitute an individual eligibility or financing decision for your business.
The same press release describes other forms of support, including accelerators with their own duration and funding. Do not mix their conditions. Consult the review of the Dare AI plan, then ask the operator for the criteria in force, the cost, the remainder payable and the steps to be taken before any commitment.
Check suitability for your business
Your number is not enough to decide whether this diagnosis is useful. See if you have an identifiable problem, accessible data, and someone capable of participating in the work. A diagnosis can help choose between several use cases; it will be less actionable if no one can explain the current processes or locate the necessary documents. Preparing these elements improves the quality of the initial discussion.
If your company has fewer than ten employees, do not assume that this presentation gives you access to the same system. Ask for guidance towards appropriate support, without confusing a resource open to all with a service financed under conditions. You can already use our AI implementation plan for small businesss and small businesss to specify a first task and avoid subscriptions taken without objective.
Formulate three measurable business problems
Prepare three situations where the team wastes time, repeats entries or has difficulty finding information. Describe the frequency, volume and consequence. “We want AI” does not allow us to compare solutions. “We spend several hours classifying requests received by email” provides a starting point, to measure before promising a win.
For each situation, note the exceptions. Incomplete files, illegible documents or last minute changes often represent a significant part of the actual work. A demonstration on a perfect example masks these difficulties. Bring a few representative cases that are appropriately anonymized. The aim is to show the process as it works, including rework and invisible checks in a simplified diagram.
Prepare data without sharing everything
Take inventory of sources: shared folders, business software, messaging, tables and procedures. Indicate their owner, their format, their update frequency and their sensitivity. Two files with almost the same name may contain conflicting versions. This organizational problem must be visible from the diagnosis; an assistant connected to inconsistent data risks reproducing the inconsistencies.
Start with fictional examples when discussing a solution. Before transmitting real data, clarify confidentiality, access, retention and responsibilities. Don't give blanket access to all your tools to speed up a first meeting. Our article on documentary assistants explains how to prepare a working basis that allows the answers to be checked.
Request deliverables that can be used after the mission
A diagnosis should leave you with elements that support a decision. Ask precisely what will be delivered: inventory, compared use cases, constraints, estimate of means and possible steps. Clarify the prioritization method. You should be able to understand why one project is recommended and why another is rejected, even if you are not familiar with artificial intelligence models.
Also ask what is not included: tool configuration, development, training, maintenance or data migration. These boundaries avoid believing that a recommendation implies an already installed solution. Identify the documents that will belong to you and the delivery formats. A report that can only be viewed in the service provider's space can make it difficult to take over the project with another team.
Calculate the cost beyond financing
The remainder of the diagnosis is only part of the cost of your approach. Add internal time, data preparation and, if you continue, onboarding, licensing, training and controls. Ask for a written costing of each step. Do not deduct the final amount from a published percentage without knowing the basis, the conditions and any applicable ceilings.
Fictional example: a pilot saves thirty minutes on a daily task, but requires twenty minutes of additional control. The net gain is then ten minutes per occurrence, before the cost of the tool and maintenance. This simple arithmetic can change the priority. Check out our article for measure the gains of AI and define an honest comparison with current operation.
Choose a driver and stopping criteria
At the end of the diagnosis, choose a limited scope: a team, a task and a test period. Write down what the system can offer and what remains validated by a person. Determine acceptable errors, blocking errors and situations where manual processing is resumed. This preparation allows you to evaluate a tool without immediately entrusting it with the entire process.
A document sorting pilot can measure total time, correct classifications and files requiring rework. Plan a decision at the end: continue, adjust or stop. Stopping a poorly adapted test is not a diagnostic failure if it saves you from costly deployment. The value of the work lies in the quality of the decision obtained, not in the number of tools ultimately purchased.
Prepare the first exchange with the operator
Gather a page of background, your top three priority issues, data sources, and the name of the internal manager. Add the timetable constraints and the resources that you could mobilize afterwards. Ask for updated criteria, administrative steps, service provider selection procedures and when a commitment becomes firm. Keep written responses with your file.
Do not present a free private diagnosis as the equivalent of a publicly funded system. At ISS Agency, a first discussion on your project can help clarify the need; it does not constitute a financing agreement and does not assume any unestablished authorization. Your choice must remain guided by the task to be improved, the data constraints and the team's ability to adopt the solution.
Frequently asked questions
Is the AI Data Diagnostic free?
The ministerial announcement mentions partial coverage of 40%. Have the price and your remaining payment confirmed before committing.
Do eight man-days mean eight consecutive days?
Not necessarily. This is an announced volume of service. The schedule and distribution of interventions must be specified with the operator.
Does the diagnosis include a complete deployment?
Don’t assume so. Request deliverables and contractual exclusions, then separately cost the work required for implementation.
Can a business without well-organized data start?
Yes, she can start with an inventory and clarification of her problems. The diagnosis can reveal that putting data in order usefully precedes any AI project.
Official sources
Transform reading into action.
Describe the task you want to improve. We can help you define your needs; the operator of the system remains your contact for eligibility and financing.
Automation: organize the follow-up of your requests →
Request the audit