94% of French companies launch AI, but 50% fail the test due to lack of data governance. A difficult move to scale.
According to a SalesForce and OpinionWay study, 94% of French companies have launched AI projects. However, one in two projects does not pass the testing phase. This paradox sums up the reality of AI transformation in France: little scaling up despite a real desire. And an underestimated cause: the lack of data governance.
What the barometer doesn’t say
Behind the adoption figures lies a more contrasting reality. As Jonathan Bodin, strategy director of Seenovate, points out, almost all of the companies that declare “using AI” are in reality practicing Shadow AI, that is to say isolated, uncoordinated uses, outside the radar of the IT department. Only 20% actually integrate AI into their processes.
AI as a revealer, not a solution
For several years, we have been supporting companies of very different sizes and sectors in their AI projects. Often, the same observation emerges: where the project stumbles, there was already a dysfunction. Blurred responsibilities regarding data. Teams in silos. Decision-making processes that do not allow for the necessary arbitrations. AI reveals these weaknesses.
Without a data strategy, AI remains theoretical or even risky
This is not a consultant’s opinion. This is confirmed by Élise Tissier, director of Bpifrance Le Lab: “if there is no Data strategy, with collection and structuring, it is impossible to take full advantage of AI.” Olivier Marcheteau, CEO of Freelance.com, goes in the same direction: “The exploitation of AI is based on a well-established Data base, technological skills specific to AI, and business skills allowing us to completely rethink processes.” Data governance is the reality condition for any serious AI project.
There’s a lot of talk about tool costs and talent matching. But the most underestimated cost of AI without governance is a concrete operational risk. As Olivier Marcheteau puts it: “It’s extraordinary to say that we obtain 97% accuracy with AI. But in a business, the 3% inaccuracy and hallucinations can be very expensive.”
New professions that cannot be improvised
The response to this fragility does not only involve technical recruitment and the identification of skills. It involves profiles that companies still have difficulty identifying and promoting.
Data Governance Experts, Data Stewards, AI Ethics Officers, project managers trained in AI: these hybrid functions, at the intersection of technology, business and compliance, are becoming strategic. They are not luxuries reserved for large groups, they are the condition for sustainable AI deployment, in an increasingly demanding regulatory context with the IA Act.
The HR challenge is real: according to our barometer, 62% of companies believe they do not have the internal skills necessary for their AI projects. The market is responding, job offers linked to AI increased by 273% in France between 2019 and 2024, but the match between needs and available profiles remains insufficient.
What leaders need to hear
AI is often presented in CODIR as a performance opportunity. It’s true. But on condition that organizations are ready to accept that their data is imperfect, their processes can be improved, their chains of responsibilities sometimes vague. This “weakness” is the starting point for any serious transformation. The companies that move the fastest are those whose leaders initiated the conversations that technology made inevitable.
Basically, the message is simple. No good AI project without data governance. There is no governance without people who have responsibility, legitimacy and the means to act. And there can be no lasting transformation without general management that treats data as a strategic asset and not as a technical subject delegated to the IT department. The algorithm is ready. The question is: is your organization?
