I'm Virginie Mathivet. I'm French. I'm the CEO of ML Ops. So the name is ML Ops, if you want to say it correctly.

The AI Adoption Gap in France

And I will start with a number. Forty -nine percent. Do you you know what it is? An idea? No? Yeah? Oh, it may be that. It's not that, but it's a good idea. Thank you.

Now, it's 99 % of French companies are not using AI in production. They have not deployed AI in the company. So, half of them have no AI.

And right now, maybe for you because you're here today it may seem something interesting or strange or I don't know maybe you think France is very late in the AI thing and the AI race but it's the situation nowadays and these numbers where does it come from from the national dsi So in English it is that, but in French it is the Baromètre National des Cidatas et IA.

I think it's a good thing for you to go on Google to search it and to download it because it's full of very interesting numbers. And if you want me to tell you after that where you can find it, I will tell you again during the network part.

so it's not a problem if you don't have the name right now but this barometer has been made late 2025 with more than 700 answers and they are all CTO or CEO of companies C -levels so it's a very interesting barometer and I think it's one of the best one we have in France right now So I will speak about that and I will compare it with what I have seen in companies.

Perspective from an AI Practitioner

So who am I and why am I here today? I started AI 24 years ago, I did a PhD in Liris, like you, yes. So it was a long time ago and screens were big, really big at that moment. and computers too were big and not very powerful and I was working on neural networks and at that time neural networks was not a thing and everyone was thinking that symbolic AI was the only thing that will work one day but not neural networks but we are here today speaking about generative AI LLMs, and so on. So maybe, in fact, neural networks work. They work. So it's a good thing. But at that time, it was not really working in real cases.

So after my PhD, I was a professor in school, but there were not a lot of AI courses. So I was also a pony instructor. So, I had to work with ponies because I couldn't work with AI only at that time, so it was a long time ago. And now I'm very happy to work in AI only, ponies only for the weekends.

And so, right now I have some awards, I'm an expert for BPI France, I'm an ambassador for Auxilia, it is a French thing in France to help companies and I work with companies since I think 10 years ago I started working with companies, 12 something like that and so I've seen a lot of things in France and I will compare what is in the barometer and what I have seen and why do we have half of French companies with no AI and so just to be sure I have a small

survival bias I think you have to be aware of that because if companies are working with me it's because they want to do AI as they will not pay me to do AI and if they have finished everything and they think they can manage it themselves, they will not call me. So what I see is only companies that need me. So they are starting things but do not know really how to do things. So this is my bias. And

Why AI Stays Stuck in Testing

if we go back to the 99 % of French companies that are still stuck at the testing stage, age, what the numbers are saying is that 12 % of companies have no initiative around AI. Not at all, nothing, no AI, nothing at all, officially.

Shadow AI Cannot Be Ignored

Because if we have no AI officially, what will we do? Shadow AI. And in fact, half of companies have shadow AI and are not ready for that. And what I see is that every time I go in a company, they have shadow AI. I have never seen a company with no shadow AI.

And so some companies are trying to just block it and say we will just forbid everything that is not official. We will ban some AI tools, we will block the access, the URLs, you cannot go there, and so on. And so we will not have any shadow AI. It's false. If you block things, people will find another way to go there. They all have a phone. And you can use your smartphone to use ChatGPT, Gemini, Cloud, or whatever you want. So it's more dangerous.

use to close the doors than just to look at what happens at the doors. So the first thing is give teams tools, of course, so that they can use them and train them about the risks not using your tools and look at what happens near the door. If you close the door, there are windows, so be careful about that.

Governance Drives Adoption

The second point is that you need governance. You need a lot of governance, in fact, and in this graph, you see the adoption score from zero to three for companies having none, one, two, three, four, five items of governance. And what you see is what? The more tools you have for your governance, the better you are in AI adoption.

So you have to create a climate, something about AI, with things that are working, that are defined, that are told to people. And what you can do, there are a lot of them, frameworks, assessment, use case, use cases registry, executive sponsors, written policy, see the famous Chartier that we find in a lot of French companies and AI committees. Be clear about how you will manage AI in the company.

It's not just one tool that you just buy one day and say, okay, I bought one tool and it's finished. No, it's not finished. It's not such as something you buy and you can go on vacation.

I've seen that in one company, in fact, once. They have given the AI thing to one guy, and say, okay, you have six months to put AI in the company. And after the six months, they said, okay, AI is done. Now you will work on something else, blockchain. So now the next six months is about blockchain.

It's not finished, it's changing every day. And everything is changing, the tools, the needs, the use cases, the people. So if you just say it's six months and then it's done, it will not work.

The Priorities for AI Success

Another very important thing is free list that you can find in the barometer. The top barriers, the top risks, and the top priorities of CTO for the next year. So it was in late 2025, so the next year is now, in fact. So there are the 2026 things that they have told the people behind the barometer.

Data, Skills, and Governance

And what you can see is that data is everywhere. Data leaks, data quality, data governance. So data are important. You can do AI right now with nearly no data, but AI is better with data, so do something about your data.

The second one is mainly the skills, and there are some training priorities and maybe some bias and ethics also is something linked to the training part, so you have to create skills in your company. It's not something everyone can just learn like that. You just give someone Microsoft copilot license and, oh, I'm very good at prompting right now because I have the tool. No, you have to learn how to prompt.

Maybe it's very easy for a lot of people in this room, but in the world outside, it's very difficult for a lot of people. So you have to train them to explain to them how they can prompt, how they can ask something to this kind of tools.

And the last one is the governance, but I have said things about that.

Assessing Your Company's AI Maturity

The next one is what is your priority right now? It depends. It depends which type of company you are.

And so in the barometer you can find four types of companies depending on the governance level and the adoption level you have. The best thing is to be the mature one because you have a high governance, high adoption, very good, good for you. You can just show the value of AI to other people.

The very dangerous thing is to be the unaware. You have a low governance so you don't really know what is happening right now in your company but you have a high adoption so people are mainly doing shadow AI and this kind of things and you cannot say exactly what they are doing and where are your data and where they are sent to and so on so be careful about this one.

The cautious is the type I'm not doing nothing I'm doing nothing so I had no problem. No, I will not use AI. It's not fair. It's very dangerous because in this case, in the future, you will be late regarding other companies doing the same thing as you're doing because you will not have maybe the same, I don't know, productivity, time to market, budget, things like that. So So you're in danger. Maybe not right now today, but you will be in danger in the future. So be careful if you are in the cautious one.

The inactive one is people that have made a lot of policies and things like that, but no one is trained, so no one is using AI. So, okay, you have a very good framework with no one inside. Maybe you will have to do a bit of training to have a better adoption.

Four Takeaways for Moving Forward

And so just for the wrap -up, four things, and it's really what I see every day in companies, big or small, from TPE, very small enterprises to very large groups. The first one is shadow AI is there. You cannot do without shadow AI. So just be careful about how you manage shadow AI.

The second one is the governance. It's very important to do it, and it will pay at the end. Maybe not right now when you're doing the first step. I don't know if you remember, but the curve is doing that before going up after that.

Don't forget the training. Skills are not magic. It's not because you have a tool that you know how to use the tool.

And the last one is know where you stand to know what you will do next. Thank you. Thank you and if you have any questions...