Why French Companies Skip AI, and What It Costs

Ask a French company why it does not use AI and cost comes sixth on the list. Not first. Sixth, at 32 percent. What comes first is "no use for the business", at 71 percent, followed by "lack of expertise" at 54 percent. That is not a survey of opinions about AI. It is the national statistics office asking companies to explain themselves, and the answer reorders everything about how this technology gets sold and bought.

Most of the conversation about AI in business is still an argument about price. Vendors discount, buyers hesitate, and everyone assumes the blocker is the invoice. The data says the blocker is that nobody can point at a first task worth doing, and that the people who could do it are not in the building. Those are different problems with different fixes, and one of them cannot be solved by a better quote.

What the French numbers actually say

Three institutional sources published within thirteen months of each other, all free to read, all more useful than any vendor blog.

Insee, the national statistics office, surveyed 11,000 companies of ten or more employees and published in July 2026. Eighteen percent use at least one AI technology. That number is heavily size-dependent: 15 percent at companies of 10 to 49 people, 31 percent at 50 to 249, and 58 percent at 250 and above. When non-adopters explain why, the ranking is the one above, with cost sixth of eight reasons. One detail matters more than the headline: at companies of 250 or more, "lack of expertise" jumps to 73 percent. Large companies can see the task and cannot staff it. Small ones cannot see the task at all. Source: Insee Première n° 2120.

France Num, the state programme for small business digitalisation, ran a parallel survey through CREDOC covering 11,021 smaller companies and published in September 2025. Twenty-six percent use AI tools, double the 13 percent of 2024 and five times the 5 percent of 2023. Usage splits hard by sector: 56 percent in IT and telecoms, 41 percent in specialised technical services, 30 percent in finance, down to 15 percent in construction, logistics and food, and 9 percent in agriculture. It also splits by who runs the company: 36 percent where the leader holds a Bac+3 or above, 8 percent where the leader holds less than a Bac. Source: Baromètre France Num 2025.

The two headline figures differ, 18 and 26 percent, and both are correct. Insee asks about AI technologies at companies with staff. France Num includes the very smallest firms and asks whether AI tools are used for professional purposes, which counts a subscription to a chat assistant. Use whichever matches the company in front of you and never blend them into one number.

Bpifrance Conseil and Siparex, the public investment bank's advisory arm and a private equity group, published twenty real projects from French mid-sized companies in April 2026, with costs attached. Seventy percent of those projects cost between 15,000 and 100,000 euros. Twenty percent came in under 15,000. Eighty percent went live in under six months, and 45 percent of them in under three. Half the cases were rated intermediate complexity, 30 percent simple. Source: Les audacieux de l'IA.

Those are French figures. In the rooms I have worked in across Europe, the shape holds: the stated blocker is budget, the real blocker is that no one has named the first task, and the price of naming it correctly is far below what people brace for.

The three things I actually see stopping small companies

The surveys measure what companies report. Here is what it looks like from inside the room, and none of it is exotic.

Nobody owns the first task. Everyone can describe the annoyance in general terms. Almost nobody can name one repeated, bounded piece of work, done by a specific person, on a specific day, that would be better with help. Until that sentence exists, every conversation about AI is a conversation about vendors.

The pilot is designed so it cannot pay off. The chosen use case is either too large to finish or too trivial to matter. It runs for a quarter, produces a demo, and dies without anyone being able to say whether it worked, because no one wrote down what "worked" would mean before starting. I have written about the specific ways this happens in six ways AI pilots quietly die.

The knowledge leaves when the consultant does. The system runs, the team cannot change it, and the next adjustment is another invoice. That is the failure the expertise number is really describing. Fifty-four percent of companies say they lack expertise, and buying a system without absorbing how it works leaves that number exactly where it was.

Ready to name the first task for your own team? Book a 30-minute discovery call and we will do it on the call.

Put the objection ranking next to the price list

This is where the two sources become useful together, and it is a genuinely uncomfortable pairing for anyone selling AI on price.

The market's stated fear is cost. The measured ranking puts cost sixth. The published price of real projects at real French companies starts under 15,000 euros, with most between 15,000 and 100,000, and most of them shipping within six months. So the fear is aimed at a number that is both lower than expected and not what is actually stopping anyone.

What that means in practice: if you are hesitating on budget, you are probably hesitating on the wrong variable. The question that will determine whether this works is not what it costs. It is whether you can name one task, measure it before and after, and have someone on your team able to run the result without help. Get those three right and the cost question answers itself. Get them wrong and any budget is wasted, including a small one.

Four starters worth running this month

Every one of these is a bounded task, not a transformation programme. Each one is written as a guide you can follow without me.

• Where to start with AI. One task, one measure, one week. Start here if you cannot yet name the first task, which is the 71 percent problem.
• The Monday brief. Your week assembled automatically from the places your information already lives.
• The meeting machine. Walk into every meeting prepared, without the hour of prep.
• The research automator. Market and competitor watching that runs whether or not you remember to run it.

If you run an accounting firm, start instead with AI for accounting firms, which maps where the hours actually go before it touches any tooling. The full set lives in the guides.

Pick one. Run it this week. If it works, you have your answer about the second one, and you will have learned more than any vendor demo can teach you.

How I work, and why it ends

Everyone wants AI. No one knows where to start. It starts with one task.

I work with a team on one real task at a time, hands on the keyboard, with the person who owns that task in the room doing the actual work. Every session ends with something built. We measure the task before and after, so the result is a number you can defend rather than a feeling that things are faster. For one person, that is a four to eight week arc. For a team, two to three months, because company context slows everything down and pretending otherwise helps nobody.

The point is the exit. Don't just rent an AI expert. Use one to become your own: my job is to bring the expertise, then get out of the way as fast as I can. A system only you can run is not an asset, it is a subscription with extra steps. If your team cannot change what we built after I leave, the engagement failed, whatever the demo looked like.

You can read the full method on how it works, or skip ahead and book a discovery call. Bring one task you repeat every week. That is the whole prerequisite.