«AI uncovered: The leaders driving the AI revolution», the new programme from BFM Business & Sopra Steria Next

Each month, in 13 minutes, a representative of Sopra Steria Next will talk to a spokesperson from a major group to gain a better understanding of the impact of AI on our economy and on specific areas of activity. The aim is to decipher the business, technological and human elements that have made AI projects a success.

[AI Uncovered – Episode 1] Generative AI and industry – escaping the curse of the POC

See the detailed description

Special edition. BFM Business Files. AI uncovered: the leaders driving the AI revolution With Frédéric Simottel.

Welcome to BFM Business in our Special Edition program "AI uncovered" in partnership with Sopra Steria Next. Today, we’re going to talk about Digital Manufacturing, that is the application of digital technologies in industrial production. We’ve been talking about industry 4.0 , robotics, Supply Chain improvement, smart and connected machines for a while. But now, we'll take a closer look at this. We’re going to look at data, AI and their impact in factories. To talk about it, today, we have two guests with us, Fabrice Asvazadourian, hello!

Hello Frédéric

Thank you for being with us. You are the CEO Of the consulting firm Sopra Steria Next. It represents more than 4000 consultants. This is the Consulting Division of the well-known group Sopra Steria. We have also with us Yves Caseau, hello!

Hello

Thank you for being with us. You are the Group Chief Digital and Information Officer of Michelin To remind you, Michelin is a group with a turnover of over 28 billion euros and that employs more than 132,000 people who need to be trained in AI, We’ll come back later to this topic. Michelin also owns 121 production sites, 9 R&D centers, ... Yves, let’s start with you like all of us, you have been surprised by the evolution, the speed of AI democratisation, especially generative AI. So very concretely, AI, what does it look like in factories?

If you visit a factory, you'll see screens that have appeared everywhere, in addition to machine tool displays. These screens allow operators to make their own applications Data-driven visualisation and optimisation. To what purpose? To optimise their process, to see if there is something that needs to be settled and to reduce material losses. So, it brings a lot of value. According to the German academy, Digital helps us to see better, to understand better, to predict better, and then to adapt better. And, that’s what Michelin is looking for. These machine tools are fully automated with sensors and data flows. So, AI helps to optimize. Let me give you a second example, about the optimisation of electricity consumption. We have a partnership with Microsoft and we use all these flows to improve the way our machines operate. It works very well. Let me give you a third quick example: in a factory, everything is constantly changing. We change dimensions, we change material, we are constantly adapting. For us, AI is not artificial intelligence, it is more like augmented intelligence. And we perceive this intelligence as tools that help us to change and make a transition from a model to another more easily. So, it is a progress that can be shared. One of the benefits of AI is to have tools to adapt better and then, you can share with others when you see something that works well in a factory.

To describe that, I'm using your term “toolbox”. It means that the work is done a lot upstream, and it represents a change in processes, a transformation of professions. How does Michelin deal with it ? As I said, Michelin represents 132,000 employees. There are a lot of people in the factories, and in all of Michelin's activities in general. All of these people have to move forward, and be onboarded.

AI, here, is the accomplishment of Digital transformation, of the digitalisation of all flows, of Software-based control. During the Covid pandemic, Michelin's Director of Manufacturing said we were able to restart factories much faster thanks to the ten years investment in digitalisation. Well, for the moment I've only shared the big advantages with you, but it's important to know that it's an in-depth work. And there is no AI without data engineering and software engineering. They form the basic framework. And then, as you also said, it's not just machines and systems, they are also the people behind the screens, and they need to make it their own. But if I took the example of screens which are used by operators to steer. This is in line with the trend of citizen development. It means that a significant portion of the value is generated when operators adopt the tools and invent local solutions for their specific business problems.

Yes, that's often what we hear in your profession, Obviously, in the industrial profession, this phenomenon is very present. When we talk about training, we tend to think of managers who must have the ideas. But ideas also come from the field, and it is these people who need to be listened to the most, in order to get the right ideas, to automate, to understand the operational tasks and processes. That's the idea.

That's right.

I'd now like to turn to Fabrice. So, AI is used at different levels, in the supply chain efficiency, to facilitate the work of employees. But above all, it has been quoted by Yves just before, it’s data. You need a rigorous architecture of data, a data hygiene, I like to use this term.

That’s right Frédéric. Maybe, you're familiar with the saying “garbage in garbage out". Data is at the heart and is essential to use AI solutions. Over the past years, the amount of data stored by companies has doubled every years. It’s huge. In order to cope with this rise, all the major groups have set up governance, management models, tools, solutions, to be able to manage that. What's new, with generative AI, is that text, image, and other unstructured information have become data and must become Data for Businesses. This change requires us to review everything we've built and talked about up to now. We need to invest in new solutions to handle and give meaning to data.

Data Lake is no longer enough?

There are different kinds of data. Storing 0, 3, 7 is different from storing sentences and images. Last year, 90% of the new data stored was unstructured, it’s a lot. For companies, it's imperative to be able to fully exploit new data to take advantage of generative AI.

And for you, Yves Caseau, from Michelin, is this a worry for your company? Indeed, data from machines was already a challenge. You also stated that you’ve been investing for ten years in all of these systems. But I imagine that all this unstructured data, all these emails that are exchanged, these technical documents, …

So no, it’s not a worry, we have data lakes at different levels in our factories and even beyond. We have data lakes in the cloud, special or private. We have a data architecture that enables us to absorb what you're talking about Fabrice, the complexity and the heterogeneity of data. But then why isn't it a worry? The purpose of all this transformation is to be more efficient, more adaptable in an ever-changing world, and to have a more pleasant job. I can give you another example. We've always done control quality with sophisticated machines. Now, we have put new generation robots with Vision machines and neural networks. Operators remain the experts. It's not the machine that decides, but the machine does 90% work, including the routine work. The operator's job of quality control becomes much more interesting. That's why it's so valuable. And we need to do this in order to recruit, in our factories, the talents of tomorrow. Indeed, today, work in factories is not necessarily perceived as very attractive. It is therefore important to invest in these solutions to make these jobs more exciting, to attract new talents, to make the tires of tomorrow.

Yes, that's what you said, the transformation of professions is important. (engineers, technicians,) Yves, AI is also used to invent the decarbonised products of tomorrow. And at Michelin, do you work with this AI too? As I said earlier, Michelin has 9 R&D centers. So, at Michelin, are you also using AI for it?

Absolutely. I will give you three examples. First, we are moving towards a low-carbon world, which means that inputs will come from circular economy whether recycled or bio-based. These new inputs have stronger dispersions than petrochemical products. We need to adapt processes and to do it, AI is fundamental. AI makes it much easier to manage these new products. Then, we need to invent in these new materials and for this, generative AI is very useful, as Fabrice said. It enables us to improve and develop our knowledge engineering, our ability to mix and match loads of data sources. But generative AI isn't used for everything at Michelin. We don't plan, we don't do forecasting with Generative AI. However, to manage knowledge, share or refine it, Generative AI is very useful.

Yes, it allows you to accelerate.

At Michelin, our favourite slogan is that the tire is a high-performance composite. In a composite, there are both structure and materials. And you know it, we've already shown off our Vision tire 3D printed. We also made Uptis which is a tubeless tire with a structure. So, innovation and structure are fundamental. To do it, we use hybrid AI where we mix classic methods, with AI and machine learning. So why is it important? It's because we're convinced that in order to invent tomorrow's low-carbon solutions, the possibilities are endless. We need to explore with Digital Twins . Because, the world of tomorrow will be realised in a concrete way. But it invents itself with simulation. It invents itself with digital twins.

Fabrice Asvazadourian from Sopra Steria Next, AI is at the service of sustainability. However, AI is criticised for consuming a lot of energy. What do you think about it? Yves has just shared some striking examples.

So absolutely. First, I think all our customers know that many AI solutions are not sustainable today. And all AI leaders, like cloud providers, solution providers are working to invent new ways of doing sustainable AI, including generative AI. And that's in front of us. It's progressing fast but it's in front of us. On the other hand, as you mentioned, AI allows us to improve our ability to be frugal, to optimise energy consumption, to optimise the use of inputs, to avoid waste. All of this is made possible by AI. What I find exciting about the arrival of AI in factories is our ability to digitalise reality in order to simulate infinite scenarios and enable faster learning. For example, Digital twins, Industrial Meta Verse,… We've reached a point where we have millions of sensors in our factories, enabling us to measure, but also tools enabling us to create scenarios, at levels never reached before.

Exactly, we have minutes left to talk about this subject. How do we prioritise all these projects? When you have, people with ideas. These projects need to be prioritised; these costs must also be managed. Even if you’ve invested for years in digital and everything is ready, you still need to know how to manage and pilot this set of parameters.

So yes, AI is not a destination, it's a journey. It started years ago at Michelin. Every generation, we learn from failures and previous successes and therefore in terms of investment and piloting, we are working at a loss acceptable. That is to say, we’re trying things, but fortunately We've had successes that made us want to go for it. As we are big, we have a hybrid structure with a center and locations called “Hub and spoke”. It allows us to try to find the right compromise between innovation close to the field, as we said, widely distributed across continents, with the capitalisation through a hub. And also, there are some difficult areas such as machine learning with reinforcement, for processes, for things where we're going to create very specialised teams. And we're not going to create that everywhere.

And as a consulting firm, Sopra Steria Next's aim is to help companies prioritise their projects. And one of the most important criteria for making projects a success is controlling costs. That's right. So, first, the costs of IT will continue to increase. No one believes that they will decline in the coming years. We've seen this with the Cloud. Secondly, what we recommend to our customers is to keep in mind projects where AI is mature enough to be deployed at scale, under 12, 18, 24 months and to be able to deploy. With this criterion of speed and getting out of these POCs failures: these POCs that fail and are never deployed. And besides, as Yves said, you must have a budget envelope to make real innovation on things we don't know if the value is going to be there. Often, we are asked to make business cases very early on. But sometimes there are situations where to make a business case, so everyone in Excel can do a lot of things, but, at the end, it's not necessarily what's needed. But what is very important for our customers is to separate well where they are in logic of time to market, obsession with time to market and where they are in a logic of exploration. It's important to separate them.

I hope you've been able to find some tips for your AI projects. Thank you to both of you, Fabrice Asvazadourian, CEO of the consulting firm Sopra Steria Next and Yves Caseau, Group Chief Digital and Information Officer at Michelin. Thank you for giving us the opportunity to hear your testimony. See you soon for a new special edition. On this topic "AI uncovered”

Special edition, the BFM Business files.

[AI Uncovered – Episode 2] How to deploy generative AI at scale and efficiently?

See the detailed description

Special Edition, BFM Business Files. "AI uncovered, the leaders driving the AI revolution", with Frédéric Simottel.

- Welcome to our programme "AI Uncovered", in partnership with Sopra Steria Next, where we'll be talking about the implementation of generative AI in companies, the transformation of organisations, cultural transformation, the transformation of businesses... and also, new forms of collaboration or how to avoid bias. We're going to talk about all this with our two expert guests. Fabrice Asvazadourian

- Hello

- Hello Fabrice, CEO of the consulting firm Sopra Steria Next. It currently represents 4000 consultants and is a subsidiary of Sopra Steria. And also with us is Jean-Paul Mazoyer. Hello Jean-Paul

- Hello

- Thank you for joining us. You are the Deputy Chief Executive Officer of Crédit Agricole, with particular responsibility for digital technologies and payments. You are also Chairman of the bankcard group. So, we have enough distance today to talk about AI, because it's true that it's been a while, this generative AI in business. And my first question, Jean-Paul, is how, at Crédit Agricole, when you're a manager and you see new projects coming up all the time, how do you prioritise them? Then the one we're going to scale up.

- I used to say that, to quote Bill Gates, we tend to overestimate the impact of technologies in two or three years' time, and underestimate them in ten years' time. But that's not why we shouldn't jump on the bandwagon, we shouldn't be inactive. I think that for AI and generative AI, we're going to have to face up to this. So there's a craze out there today, and it's important to follow it, but obviously remain measured in what we do,

- Yes, and to encourage

- Yes, to encourage. So I think we're witnessing a change of period. There are a lot of initiatives in companies today, but there are also a lot of POCs, which are arriving all over the place, but there aren't many experiments that have been thought through on a large scale and that have already been deployed with a significant impact, either on the company's NLP, or on the number of employees affected. In my opinion, it's important that today's generative AI projects are driven and supported by the business. It's not just about technology, it's not just about choosing the right LLM, it's about thinking "OK, I'm the boss of a business, what's it going to change in my business? We need this business-driven approach. Obviously, we need to have technological control over what we do, and we need human support. Which means that these generative AI projects need to be managed at the highest level of the company. And so that's at Comex, it's by the boss. You know, we in the Crédit Agricole Group are natively decentralised, so it's important that every manager in each of our entities feels fully invested in the need to understand these technologies and to support them.

- So you set these criteria, you said to each of the business line managers "Here, give me one or two projects, then we'll see how we get on".

- Exactly.

- So, we have to change our way of thinking, our software, if we stay in the IT world. Fabrice, do less experimentation, fewer POCs, and scale up too. And it has to be good. As we've just said, it's your job to manage projects. We keep saying it, we say it for other IT projects. But in the case of generative AI, it's even more important.

- Absolutely. I think we're seeing it clearly with generative AI, but also with other technologies such as 'low code, no code'. We're moving towards decentralisation, as close as possible to the users, of computing. In a way, natural language is almost becoming the programming language, and this gives business people much greater capacity to use technology to solve their business problems. Technology is only useful if it solves problems. That's what we see, and what we recommend to our customers. With regard to your question about mastering the time dimension, perhaps it's a question of thinking in three time horizons. It seems to us that there are areas in which we now have sufficient evidence to say that we can move into deployment mode. I'm thinking of virtual assistants, I'm thinking of digital marketing, I'm even thinking of tools for computer engineers. We know today that some use cases, not all, but some are mature and so it's important for a company to move quickly.

- Yes, with productivity gains from automation...

- That's it. We need to prepare for wave two, which will focus on, probably, more cross-functional uses in HR, finance, etc., but which will require companies to mobilise their proprietary data even more in order to refine the AI models that will come from Big Tech or start-ups. And then there will be the third phase, but we're probably three years or more away from really applying AI to companies' core businesses. Before we get there, we still need to have levels of confidence and reliability that are a certain number of steps ahead.

- And then levels of acculturation. Jean-Paul Mazoyer, we can see that generative AI raises a lot of questions about its uses. With its prompts, it is accessible to everyone. So we really need to look at whether it's accessible to anyone in the banking world, whether it's someone in a branch or someone who's a telephone operator like the top executive. So that's what's changing today with this technology. For you, it's information, it's acculturation, it's all quite technological, it's very important. And then relearning how to work, or relearning how to work collaboratively, everyone already works together a lot, but there are a lot of things we need to relearn.

- I share this opinion. I believe that generative AI is going to have an impact on all business lines for a company like a bank, with a very large majority of employees who are going to have to learn to live and work with AI on a daily basis. It's not AI that will replace employees, it's employees who will know how to use AI, who will replace employees, whatever their level, whatever their job. So all the men and women need to be trained, to be acculturated, to understand the uses, to understand what it can enable them to do. So a lot of support is needed, both for employees and for managers. A lot of support, a lot of training, a lot of deployment. A few years ago, we deployed collaborative tools like Teams, and when you ask people how they use it, they say it's a video tool. And few have really transformed the way they work to move towards collaboration. It shouldn't be the same with generative AI tools. You have to ask yourself, "What's this going to change?" "How can I reinvent my way of working, my way of connecting with others using these tools?

- I imagine that you've launched policies and training programmes for each profession. How are things going?

- Yes, that's exactly it. We're going to look at each business and each function, and the rate at which tools are deployed, because it depends on the tools you're going to deploy. Let's take some very concrete examples, like when you deploy Copilot for office automation tools. This can have a major impact very quickly on the transcription of a meeting, a report, a translation, whatever you want. People still need to be able to use it as easily as possible. When you deploy GitHub Copilot for IT teams, it can also have a very significant impact. But you still need to provide support, because there's always the necessary critical eye on the part of the employee about any potential hallucinations. So it's important that we have employees who are able to use it, who understand it, but at the same time are able to take a critical look at what's being produced.

- What's more, we know that we're going, that no matter what happens, we'll be moving towards these technologies. Fabrice, we need to think about a certain industrialisation of AI, because there are lots of things we need to push on. You have to be able to bounce back and forth between the proximity of each of the business divisions, the use cases, the effects of scale, the technologies too, because we haven't gone into detail here, but obviously there's Mistral, OpenAI and Google Gemini, and then there's a whole bunch of others coming along. After that, you sometimes have to combine that with the processors behind it. There are a lot of things that need to be taken into account.

- You're right, Frédéric, the industrialisation of AI is the theme of 2024. And with us, all the requests from our customers are "Help us to industrialise" "How do we go about taking POCs, which have demonstrated in POC mode that they bring benefits, and deploying them at scale?" Four questions generally come up? The first, as we've already discussed, is "how do I prioritise?" I can't do everything at once. So how do I prioritise? The second is "But for there to be AI, my data platform has to be up to scratch". The level of requirement for AI is much higher than it was before. So how do I modernise my data platform? There's the subject to talk about AI industrialisation again. "I have a POC, how can I deploy it while maintaining the quality level of the results?" Because we realise that there can be a gap between the protected mode of the POC and the deployment mode. "How do I integrate these AI solutions into all my industrial IT?"

- Yes, especially as there are some that need to be kept within a perimeter and others that can be opened up a little.

- Yes, they have to feed off data and interact with other tools. And the IT of a large group like Crédit Agricole or others is the IT industry. So what works on a small scale has to work on a much larger scale. And then the fourth question. Is the question "Does performance not deteriorate over time, or does it improve? "Doesn't a new technology replace the choices I made a month ago? So it's this maintenance in operational condition. And that's it. And all this leads companies to ask themselves questions in terms of organisation. I don't know if it's a question of scale, but it's more a question of learning effects. Today, it's a question of "how can I learn faster? You're right, you have to choose technologies, partners...

- Who move every three months!

- Which change... And you have to define the frameworks within which you want your employees to be able to use AI with confidence and security. So there are a number of issues that need to be addressed, and the answers will vary depending on the group's culture. Some are natively more decentralised; Crédit Agricole is an obvious example, and so we see central teams defining standards, rules and choices, and then leaving a lot of freedom to the business lines to produce their solutions locally. Others are building AI factories, bringing together people from the tech world with specific skills, and then people from the business who, for a year or a month, will step outside their operational role to build solutions. There are these two models. And then all the shades of grey in between. Every company needs to know itself.

- What would terrify me if I were in the shoes of a manager like Jean-Paul, or advising managers like Fabrice, is the speed of innovation, saying "Wow, in three months we've already got a new version of ChatGPT out, etc.". Then there's another point, which is all about ethics, ethical AI, and I imagine that in banking, of course, you have it in every business. But there you have it, there's a new indicator that needs to be put in with this artificial intelligence and linked to this ethics.

- Yes, at Crédit Agricole we have two convictions in this area. The first is that we're going to have to make massive use of AI and generative AI tools, with all that Fabrice has just mentioned. So there's no question of missing out on this revolution. We need to give ourselves the means to consider it properly, to organise ourselves properly to be able to deploy it on a massive scale, because it has a massive impact on customer relations and internal organisation. The second conviction is that we are convinced that all of this must be at the service of people. Our logic is that these tools should enhance the human element, but they should also be enhanced by human responsibility. In other words, if we have this direct interaction with customers, we must always have the possibility that a woman or a man from Crédit Agricole can take over and complete the relationship with customers. So the first rule, in a way, is to say that responsibility must remain with the men and women of Crédit Agricole. We've been aware of these ethical issues for many years, and we already had scoring tools that we were obliged to be careful about. So we have these issues, but it's becoming much more complex with generative AI to be able to understand what data has been used, whether there are any biases, and that's what we're going to have to control. And we're obviously going to need to have a position on these biases or algorithms and therefore be able to work in confidence with certain LLMs to be sure that the data used are the right ones. There's one final conviction Yes, it's about electricity consumption. We know that AI and generative AI consume a lot of electricity.

- Yes, and that's the big issue that's going to come up in the next few months.

- Exactly. And we already know that the biggest AI companies are the biggest consumers of electricity. We're going to have to ask ourselves the question of, not just LLMs, but SLMs, or Small Language Models, so that we can be more economical, more frugal in our use of technology.

- We'll certainly have an opportunity to talk more about this with Fabrice at an upcoming “AI Uncovered “ program in partnership with Sopra Steria Next. Thank you both, Jean-Paul Mazoyer from Crédit Agricole and Groupement Cartes Bancaires and Fabrice Asvazadourian from Sopra Steria Next. Thank you for following us, and we look forward to a new programme, "In search of AI", very soon.

- Special Edition, BFM Business Files

[AI Uncovered - Episode 3] Generative AI in the insurance world: barrier to be broken down or opportunity?

See the detailed description

Voice over: Special edition. BFM Business Files. "AI uncovered: the leaders driving the AI Revolution" with Frédéric Simottel.

Frédéric Simottel: Welcome to our programme "AI uncovered" in partnership with Sopra Steria Next. We're going to talk about the implementation of AI, generative AI. We've already done two programmes with leaders in the digital word, from major companies and with you Fabrice Asvazadourian, hello!

Fabrice Asvazadourian: Hello!

Frédéric Simottel: Good morning and thank you for joining us. You are the CEO of the consulting firm Sopra Steria Next, which currently employs over 4,000 consultants. Of course, behind Sopra Steria Next, there is Sopra Steria. You are an entity of Sopra Steria group. And with us also today Hervé Thoumyre. Hello !

Hervé Thoumyre: Hello.

Frédéric Simottel: Hervé, thank you for joining us, Director of Customer Experience, Digital Services and Data and member of the Executive Committee of CNP Assurances. So we're starting to see more and more, as each month passes, we gain a little more perspective on Generative AI that is developing everywhere. Let me start with you, Hervé. We're talking a lot today about the benefits of AI in winning new customers. How is that working out for you at CNP Assurances?

[Banner " AI to attract new customers "]

Hervé Thoumyre: At CNP Assurances it's happening by Firstly, simplifying the daily lives of our employees, our partners, but also our customers, and finally, that's nothing new because we've been investing in AI for at least ten years.
Frédéric Simottel: Yes, it's true that we're talking about Generative AI because it's just arrived, but AI is something that everyone is working on, including you.

Hervé Thoumyre: In fact, I think we're more of a pioneer in this field. We set up our Data Lab in 2015.

Frédéric Simottel: Yes, indeed.

Hervé Thoumyre: And I have to say that today it's also materialised by the fact that we are convinced that AI has definitely become a lever to ensure that we create a competitive advantage, with our old partners. I would say the historical partners, the banks but also the new partners.

Frédéric Simottel: Yes, it's true that this is in your DNA. You have this ecosystem of partners around you.

Hervé Thoumyre: Completely. We're really in a multi-partner model, both with banks and with new sectors. Let me take the example of the retail sector. Last year we sealed a partnership with Carrefour, for example. So we're in this model. And to stand out from the crowd, the challenge is to offer insurance products that meet our customers' expectations, of course, but also to offer them services and ensure that the process, in particular, is as simple and as straightforward as possible. So, let's take a simple example of credit insurance. When we take out a loan for a house, we like to know very quickly whether our loan is covered by insurance under acceptable conditions.

Frédéric Simottel: Yes, that's always the sticking point.

Hervé Thoumyre: Well, today this process is completely automated using artificial intelligence so that in 85% of cases, we give an immediate response to our customers, as well as to the advisor looking after the customer. And I can tell you that in many cases, they're very happy to have this level of service.

Frédéric Simottel: It also makes negotiations in this area a little more transparent. Fabrice, from Sopra Steria Next, How is CNP Assurances exemplary? It's a data industrialist. Obviously, when you're in the world of finance and insurance, you have an enormous amount of data. So how are they exemplary when it comes to developing new businesses? As Hervé was saying, we've been working on this for several years, but we're raising the bar a little?

Fabrice Asvazadourian: Yes, we are. Hervé's testimony clearly shows that mastering data is almost a competitive advantage for CNP Assurances. It allows them to create seamless customer journeys, It allows them to make productivity gains in their middle and back office and it probably allows them to refine value propositions for particular segments or types of customer.

Frédéric Simottel: And then you have to see if we don't have data, a certain data hygiene. In any case, there's no point in embarking on AI projects.

Fabrice Asvazadourian: Absolutely. In the end, the rule is quite simple: If you don't have data, it's going to be hard to get AI to work properly, and we've been at this since this year, and Hervé is right, we need to be looking at ten years rather than two. Even if generative AI is changing the game. We need to feed these machine learning and deep learning models with massive amounts of data and with data from increasingly varied and different sources. It can be text, it can be images, and it all becomes data. And so for managers, we need to make sure that we regularly update these data platforms so that we can put this data to work for our business and facilitate interactions, in the case of CNP Assurances, with our partners. Because in the end, the data initially resides with the partner. We need to get it to CNP Assurances in a way that protects its partner's property, the rights of its partner's customers and secures all that.

Frédéric Simottel: So, it is precisely this data that comes back from partners and that goes back to all CNP Assurances employees. Now, as soon as we talk about AI, generative AI, we talk about augmented employees, assisted employees with more automated tasks. A lot is being done around this employee. Hervé, how will this augmented collaborator be able to add more value to the service provided to the customer?

Hervé Thoumyre:
[Banner « Hervé Thoumyre, Director of Customer Experience, Digital Services and Data & Member of the Executive

Committee of CNP Assurances]

[Banner « AI at the service of the “augmented collaborator” »]

Well, first of all, I would say that AI serves the employee and the human being in general, not the other way round. And in reality, and that's a very important conviction, which is really part of who we are, our raison d'être. In other words, we want people to be at the heart of our business model, at the heart of everything we do every day. So if I come back to the question, what this means in particular is that what we're trying to do first and foremost is free up time for our employees and ensure that they can express their expertise as fully as possible. Here are a few examples. Today, we know how to read the beneficiary clause of a life insurance policy and extract the key data from this clause to prepare an inheritance file. Secondly, in terms of fraud detection, we use AI technologies that enable us to detect fraud on identity card or bank details. Another example is e-mail processing. We know how to automate mail processing in order to direct them and organise the work of our teams.

Frédéric Simottel: Now that's almost traditional AI. But today, we're talking about generative AI. When we talk about customer service, generative AI often means, for example, we have all these verbatims from people who write or call. So, there you have it, there's a whole host of data sources coming in and generative AI can perhaps give us a better sense of what the customer is feeling, of what's going wrong.

Hervé Thoumyre: That's true. And since the beginning of the year, we've been using a platform that allows us to analyse and process customers' feedback, everything they tell us, so that we can respond, particularly if there's an emergency. The objective is also to learn from everything they tell us so that we can improve our customer experience.
Frédéric Simottel: This data comes from our customers. But there's also what we call synthetic data, Fabrice, which plays an important role. This is data created by other data.

Fabrice Asvazadourian: Exactly. First of all, what is synthetic data? It's data that was created by an algorithm and that aims to reproduce, initially, what happened two or three years ago, existing data. This means that you can take data from your customers or your operations and transform it into data that has the same statistical properties, but which is no longer real data. This means that data confidentiality is totally protected and that data sharing between partners is automated. And what we're now seeing thanks to generative AI is that not only does it reproduce existing data, it now creates new data, and this has led to the advances we've seen in pharmaceutical research, for example, where we've been able to drastically reduce the time it takes to decide which type of molecule to use. Because by creating data that generates new data, we've been able to reduce the time needed for experimentation and the creation of new data. That's the great advance, The great revolution.

Frédéric Simottel: Yes, we can even see this in the polling institutes, which sometimes use this synthetic data to get slightly more accurate figures.

Fabrice Asvazadourian: Exactly.

And when you consider the cost and time it can take to create databases of the right size for models that are increasingly demanding, it's clear that this is an asset in terms of productivity and automation for companies.
Frédéric Simottel: Hervé Thoumyre of CNP Assurances, another important subject, is this ethical and inclusive AI.
[Banner "Towards ethical and inclusive AI?"]

We talk about it a lot, but we were talking about trust earlier, about being able to inspire this culture of trust that was between customers and CNP Assurances agents. But this trust must be everywhere.

[Banner " Hervé Thoumyre, Director of Customer Experience, Digital and Data of CNP Assurances "]

Hervé Thoumyre: That's true. And I think it's based on three essential points. The first is to have a charter. Any company that wants to embark on AI must have an ethics charter and associated governance to really put it into practice. The second point is training, making employees aware of the use of AI. As part of the La Poste group, 70,000 employees have been trained in the use of AI technologies. And then, the third essential element is data sovereignty. We were just talking about it. And this sovereignty is demonstrated by the fact that we are in the process of migrating all our data platforms to the sovereign cloud Numspot, of which the major French public financial centre is one of the main shareholders and of which we are a part.

Frédéric Simottel: Yes, particularly with Docaposte, it's an offering that can be managed from that side. But it's true that this sovereignty of data is becoming, and I think we understand it a little better, I find, often when we talk about sovereignty, rather than talking about the cloud, it's the sovereignty of data. It's a little easier to understand what we're working with here. Regarding this ethically, sustainable AI, how does CNP Assurances seem to you today, Fabrice? Compared to the other companies you see? Are we in line or are we a bit ahead?

Fabrice Asvazadourian:
[Bandeau « Fabrice Asvazadourian, CEO of the consulting firm Sopra Steria Next »]
I think that CNP and Sopra Steria share the same high standards on this subject, because AI will either be sustainable, or it won't be. So, I can see that European companies are setting the bar higher than others on these issues, on this sensitivity. It's both about environmental issues. Today, generative AI, we know, is not sustainable, but a lot of billions are invested...

Frédéric Simottel: Perhaps, we're thinking ahead more than the Americans, if I take their example? It's when they're confronted with it that they react. We try to anticipate a little more.

Fabrice Asvazadourian: In any case, it's a subject that is one of the challenges for a manager. It's a question of "how can I still move forward? " Because, in my opinion, not moving forward is not the right solution, while at the same time, keeping an eye on all the new solutions that are emerging and that will enable me to reduce and control areas of risk. When you look at the billions and tens of billions of euros of investment being made by Google, Amazon, Apple, Nvidia and others, you know that the solutions are coming. But they are not all there. So, as a manager, how can I move fast without going too fast? And how can I be agile enough to integrate these new solutions when they will provide me models that consume less energy and water, models that will allow better control of data, of risks, of bias, etc. There are many dimensions in AI.

Frédéric Simottel: Yes, when we talk about 'responsible AI', there's the ethical part and then there's the sustainability and environmental part. Well, thank you for coming here enlightening us on these subjects. Hervé Thoumyre, Director of Customer Experience, Digital Services and Data, member of Comex, CNP Assurances where AI has been in use for quite a few years and where generative AI is accelerating in many areas. And thank you Fabrice Asvazadourian, CEO of the consulting firm Sopra Steria Next. See you soon for a new programme entitled "AI Uncovered". We've already got two or three on Replay, in the world of industry, in the world of banking, and today, in the world of insurance. We promise you plenty more throughout the year. Have a great week on BFM Business.
Voice over: Special Edition. BFM Business Files.

[AI Uncovered – Episode 4] Luxury: These Three Major AI Applications You Should Prioritize

See the detailed description

[Appearance of the programme’s - "AI Uncovered: the leaders driving the AI Revolution”. Voice over]
Special edition. BFM Business Files. "AI uncovered: the leaders driving the AI Revolution" with Frédéric Simottel.
[Frédéric Simottel]
- Hello and welcome to this special issue "AI uncovered: the leaders driving the AI Revolution.” And we're going to be talking about AI and luxury today. But you can already replay our three interviews with Michelin, Crédit Agricole and CNP Assurances. Now we're taking a look at all these sectors with our partner Sopra Steria Next. This week, it's all about luxury, with lots to see! Improvements in shops, in the supply chain... We'll be looking at all of this with our two guests. Franck Le Moal - Hello Franck
[Franck Le Moal]
- Hello
[Frédéric Simottel]
- Thank you for joining us as LVMH's IT and Technology Director. And hello Fabrice Asvazadourian!
[Fabrice Asvazadourian]
- Hello Frédéric
[Frédéric Simmotel]
- CEO of the consulting firm Sopra Steria Next
[Appearance of a banner entitled "Artificial Intelligence at the service of luxury”]
So Franck, I'll start with you. So AI is appearing in many areas at LVMH. It's true that the list, we have a time limit, but it's quite long. You've trained a very large number of employees and then, when we were preparing this programme, you said "Hey, there's an interesting use case, it's in designer workshops".
[Franck Le Moal] Yes, that's right. So, first of all, you're right, we've trained a lot of employees in the LVMH group, since we have around 10,000 employees trained in AI and GenAI.
[Banner " Franck Le Moal, Director of IT and Technology at LVMH "]
And we're continuing with an approach that is increasingly segmented by profession, by geography, because what's important now is to bring AI and GenAI into the professions so that they can add even more value. We're slowly getting into the workshops, but when we say workshops, we mean workshops. We have a vision which is extremely important in luxury and at LVMH, and that is that we don't want AI and GenAI to replace our creators and designers, they are there to help and support. But behind the process of creating our products, our shops, our windows, there will always be men and women with the passion and aestheticism of their creations. Indeed, GenAI will add to them. Here are a few examples...
[Frédéric Simottel]
- Yes, very concretely, we're going to make a prompt saying "Here, I need this kind of thing" And then the Creator is going to work from these images?
[Franck Le Moal]
- So, for example, at Louis Vuitton, we've set up a little tool called the "AI Atelier", which will give style assistants and assistant designers a kind of wall of inspiration.
[Banner " Franck Le Moal, Director of IT and Technology at LVMH "]
Shapes and colours can help them in their research process. Trigger a little inspiration. The same thing happens in shops. As you know, we have very beautiful, very luxurious shops. So, for example, at Christian Dior Couture or Fendi, by the by the window designers, the merchandising designers, who will draw inspiration from the shapes, colours and atmospheres. So this will help to feed the creative process, without replacing it.
[Frédéric Simottel]
- Fabrice That's what we always say, it's not AI that will replace us, it's the person using AI who risks replacing us. We have a very concrete example of this. AI won't kill creativity, on the contrary?
[Fabrice Asvazadourian]
- Not at all. First of all, generative AI is not, it does not create. It's highly advanced statistics that allow you to
[Banner " Fabrice Asvazadourian, Managing Director of Sopra Steria Next "]
save a lot of time on inspirational tasks. But creation, the human process of creating generative AI, is incapable of this, so we won't be replacing creators, real creators, any time soon. On the other hand, generative AI can already help them a great deal. And that, in the end, is always about keeping the human being's greatest added value in focus. And in the case of designers, the aim is to lighten them, to multiply them, to increase them as they say, thanks to AI. And that's what it's all about.
[Frédéric Simottel]
- I like the idea of a trigger. We all want to write a post or an article, we're looking for inspiration, we're thinking, I've got some ideas, but I can't find the beginning. You do your little prompt and you come up with two or three beginnings. You say ah yes, here, I'll use this one. And then we unroll the whole thing. Frank, AI can be found in the boutiques, as you said, in the design department, but also, I imagine, in customer relations, with customers who now, even in the luxury sector, it has to be said, go online, use social networks and also go into the boutique.
[Franck Le Moal]
- Of course Frédéric, you've hit the nail on the head, First of all, it's in the world of the shop and the salesperson that we're making the most progress. GenAI can be very useful. Exactly, as for creators and designers. We're not there to replace the salesperson, we're there to help him, to enhance him and to give him a new lease of life.
[Banner " Franck Le Moal, Director of IT and Technology at LVMH "]
at the heart of an even richer relationship, with an even stronger experience, with its customers. So we really want AI and GenAI to enable our sales staff to focus on customer knowledge before our customers come into the shop. And when the customer arrives in the shop, they walk through the door, and there they are, with these tools. And I'll give you two or three examples, we're going to have another sales assistant who's going to support and enhance the experience and the dialogue with the customer. So what do we do? We try to find out what our customer is doing. So, of course, we're very respectful of our customers' data, but we know that our customers go online, they use digital platforms, they look at social networks, they call our call centres. So we're going to have a mass of information that is extremely interesting, and we're going to make it available to our sales people so that they can synthesise their value proposition even better. So that's one of the first use cases. We have deployed it on a large scale at Tag Heuer, at Bugari, at Tiffany, at Vuitton, at Christian...
[Frédéric Simottel]
- Yes, because every house has its own particularity...
[Franck Le Moal]
- But at the same time we're trying to re-use these cases. And the second point is that once the customer has come into the shop or left the shop, we're going to further enrich our exchange with them. For example, we're going to synthesise all the purchase history and comments that a customer may make, and our sales assistant, who will have his iPhone with his application, will be able to use the GenAI prompts that are integrated into the application to propose messages that are more qualitative and more personalised. That's what we're doing.
[Frédéric Simottel]
- I'm caricaturing here a bit, but how do you manage to strike the right balance, so that the sales assistant doesn't suddenly feel too... I'm caricaturing here a bit, but too much in front of his screen, so that he has all the data on the customer?
[Franck Le Moal]
- So we're working hard on ergonomics, we're working on applications that are going to be extremely streamlined and extremely fast. We're actually starting to test GenAI engines where the seller will speak and the iPhone will respond. These are the tests we're carrying out and the implementations we're in the process of making. So we're trying to make sure that during the act of buying, the experience and interaction with the customer are as seamless as possible.
[Banner " Franck Le Moal, Director of IT and Technology at LVMH "]
He has prepared, he quickly looks at his iPhone because the applications are very well designed and so I think that when we talk, if you have the opportunity to go to the Dior Couture boutique in Montaigne, for example, you will see a very fluid experience with the mobile phone which is always there, but in an extremely light way and in the discussion and exchange with our customer and our sales advisor.
[Frédéric Simottel]
- Because that's one of the difficulties. We may come back to the question of prioritising projects. How can we be sure that this type of application - AI in the service of the customer, customer personalisation, customer relations - will generate this value, Fabrice?
[Fabrice Asvazadourian]
- So that's the whole question. And Franck has already highlighted the key points. The first thing is that, thanks to the arrival of generative AI, we are now able to combine predictive AI, which is not new (companies have been doing it for years), to produce the famous 'Best next actions'. What is the next product or service that I can sell to my customer?
[Banner " Fabrice Asvazadourian, Managing Director of Sopra Steria Next "]
The problem was that it was quite cold, so now we're able, as Franck was saying, to contextualise it so that the advisor, the sales person, can not just come and push a product, but come and engage a customer.
[Frédéric Simottel]
- In fact, putting the relationship into context...
[Fabrice Asvazadourian]
- So that's a quantum leap in efficiency. The only thing is that for it to be a leap in efficiency, humans still have to use it.
[Frédéric Simottel]
- Yes, we're all familiar with those big customer relations projects that were set up a few years ago, but...
[Fabrice Asvadourian]
- Yes, and it's clear that working on the employee experience, on how employees will use it, on ergonomics, is what's going to make it a success, and I think that today, we're sometimes too much on the technical side of AI and not enough on the experiential side. In my opinion, this is what's going to make the big difference between the companies that make money out of it and those that make models, which is not quite the same thing.
[Frédéric Simottel]
- Yes, Franck also pointed out that. We're really going to have this exchange which will further enrich the relationship between the employee, the tool and the platform.
[Fabrice Asvazadourian]
- Yes, and less is more. We're going to have to hide the complexity for the user.
[Frédéric Simottel]
- So, Franck, we've seen the boutiques, we've seen the design, we've seen the customer personalisation, and what's also important is AI serving the supply chain. Because obviously in the luxury sector you have very demanding customers. So they want their product as quickly as possible, in the right place, at the right time. We're also involved in CSR, and that's important too, because it allows you to work your way up the chain and ensure that all the criteria currently applied in the company are respected.
[Franck Le Moal]
[Banner " Franck Le Moal, IT and Technology Director, LVMH "]
- Completely, Frédéric, because in fact the second very important subject for us is the supply chain. But more than that, it's also production. In fact, you are right to sum up our customers' expectations. But we must never lose sight of the fact that we are the luxury industry and this industry is a bit different from others. We're not a mass production industry. We handle and use expensive and rare materials in our products. That's all there is to it. We have men and women in our workshops who produce the smallest possible quantities on a just-in-time basis. And so, of course, AI and algorithms, especially on AI, are going to play a very important role. So, how we can be as relevant as possible in assessing the expectations of our markets ? So everything we do is geared towards optimising sales forecasts, so that we know exactly where the demand is, and where we need to send the right product to the right shop. So this is a very important subject. We have houses with completely different activities, we have a wealth of products and collections that means we have to get the right product to the right shop. I'm not even talking about seasons. We have the northern hemisphere and the southern hemisphere. Summer in Greece versus winter in Argentina. So these are extremely important issues, how best to prepare our product ranges and, above all, how to produce rare products. So producing at the right level, producing effectively to avoid transfers, not overproducing with extremely expensive materials, is very important. So anything that optimises the connection between sales and the signal we send to our workshops to produce the right quantities is absolutely essential. And, of course, to provide our customers with the best possible service in terms of what they expect, with the best possible experience. And that's where you're right, in an extremely strong way, we're going to have a model, AI models that will contribute to this phenomenon of sustainability and the environment, because what we don't produce is materials that we don't spend and above all products that we send to the right place. These are products that we're not going to move again, so they immediately have an impact on the environment and on supply chain performance.
[Frédéric Simottel]
- And we know that it's the supply chain
[Franck Le Moal]
- So the supply chain and production are extremely sensitive environmental factors. So that's our focus in most of our companies.
[Frédéric Simottel]
- It's many, many... a multi-part equation, not all unknowns because, thanks to AI, they are less and less unknown, but there are many parameters to take into account. Fabrice, through Franck's testimony, we see all this maturity with regard to AI. We're in a fine company like LVMH, a major group that has been assimilating all these technologies and this digital transformation for some time. And how can we ensure that this value-performance-investment ratio is taken into account in the best possible way elsewhere?
[Fabrice Asvazadourian]
- So that's a bit of a frustration for a lot of our customers, who say to themselves, well, we're doing POCs, we're doing POCs and then we fall into the curse of POCs that never get deployed.
[Banner " Fabrice Asvazadourian, Managing Director, Sopra Steria Next Consulting "]
We need to focus on use cases that are very mature today. Of course you have to devote perhaps 20-15% of your budget to investment and innovation, but if you put 70-80% of your budget into AI, in things where you know that solutions exist, you know that there is a key business value and that you just have to land it at home, which is already not bad, and secondly, land it at home. The first is no good data, no good AI. So how do I modernise my data platform to bring it up to the same standards as AI? That's a massive challenge. And secondly, how do I get AI algorithms into my IT system today? We have a figure: one AI algorithm in seven continues to perform satisfactorily when it goes into production. In other words, the transition from in vitro to in vivo is currently six out of seven that don't survive.
[Frédéric Simottel]
- Oh yes, so it's in your interest to calculate the cost from the outset.
[Fabrice Asvazadourian]
- Work on industrialisation right from the start. I think that's going to be the big challenge over the coming months. Industrialise AI.
[Frédéric Simottel]
- Yes, because everyone has ideas, but you have to prioritise in relation to the value you create. But obviously, industrialisation is the key to this work, to scaling up, and we've seen that clearly. Thank you again to Franck Le Moal for coming to talk about this. So much for these different aspects. I would also like to thank Fabrice Asvazadourian, Managing Director of the consultancy firm Sopra Steria Next, for joining us. There you have it, I hope we've shed some more light on AI. And as we can see from the examples here, it works, but there are a whole host of things to think about well beforehand, not least the data. Thank you for joining us, and we look forward to a new edition of AI Uncovered.
[Voice-over]
- Special edition. BFM Business Files.

[AI Uncovered – Episode 5] Generative AI, Sales Leadership & Customer Experience

See the detailed description

[Appearance of the programme’s - "AI Uncovered: the leaders driving the AI Revolution”.]
[Voice over]
Special edition. BFM Business Files.  "AI uncovered: the leaders driving the AI Revolution" with Frédéric Simottel.

[On-set cameras]

[Frédéric Simottel]
Welcome to our programme "AI uncovered: the leaders driving the AI Revolution", in partnership with Sopra Steria Next, where we're going to talk about the implementation of generative AI and trusted AI within companies. As you know, if you watch the replays, we're interested in industry, banking and insurance, and then we're going to take a look at sales departments and customer services within the La Poste group with our guest Pierre-Etienne Bardin. Hello !

[Pierre-Etienne Bardin]
Hello.

[Fréderic Simottel]
Thank you for joining us, Chief Data Officer at La Poste Group. So, more than 15 billion objects distributed around the world every year, a turnover of 34 billion euros in 2023, 230,000 employees, 500 experts in data and AI and a group that is truly innovative in both the digital and AI fields. We're going to talk about it with you and then with us.

[Appearance of a "Deploy AI with confidence" banner]
Fabrice Asvazadourian, hello!

[Fabrice Asvazadourian]
Hello Frédéric

[Frédéric Simottel]
Fabrice, thank you for joining us. CEO of the consulting firm Sopra Steria Next. It has over 4,000 consultants. And it's a subsidiary of the well-known ESN Sopra Steria! Pierre-Etienne We're going to start with you on the ideas for use cases. I imagine there's no shortage of them at La Poste group. You must manage them and see them arrive on your desk every day, especially with generative AI. So you had to prioritise and you thought, well, maybe there's an area where we can get started. Well, there are several, but this is one where we might see some applications straight away. It's for sales departments and customer services.

[“AI: how to prioritise projects" banner]

[Pierre-Etienne Bardin]
So let's go back a bit, let's go back a bit. We've all been victims of this generative AI wave. And when I say we've all been victims, I'm talking in inverted commas because it's not as dramatic as all that. And all the business lines have taken an interest in these transformations, these innovations, and the IT teams have also taken an interest, particularly in understanding and knowing how to use these new technologies. So we decided to filter the use cases and take those with the biggest impact.

[More than 15 billion objects distributed 233,000 employees (more than 179,000 in France) €34.1 billion turnover in 2023 More than 500 data and AI experts].
We could have taken the easiest, the simplest ones. No, we wanted to take the ones with the biggest impact, to demonstrate that generative AI was really something that was a major breakthrough compared with traditional AI. We developed an in-house solution, a single point of contact, a single point of entry, based on an LLM. We didn't want to use ChatGPT for this solution, so we called on our experts to build this solution, this in-house solution, to meet our needs based on our data. This is 'La Poste GPT'. That's its name.

[Frédéric Simottel]
Oh, I don't think anyone is original in this field!

[Pierre-Etienne Bardin]
So La Poste GPT means using generative AI technologies on our knowledge bases, on postal data. That's what was important for us, and it's a first. And it's an assistant. And indeed, when we chose this use case, we favoured the one with the greatest impact, and therefore for the sales teams and customer relations teams, to make it easier to respond and to speed up the responses we give to customers, whether they are major accounts, business customers or individual customers. That's what was important to us. We're now entering a phase of industrialisation and scaling up. We've been experimenting for a year now, testing different solutions and choosing certain algorithms. We've also done a lot of work on RAG technologies, i.e. technologies that enable us to understand our knowledge bases and structure them in such a way that they don't hallucinate. And that they respond correctly to needs. Because that was very concrete for us.

[Frédéric Simottel]
what does this mean in practical terms for users?

[Pierre-Etienne Bardin]
In practical terms, what does this mean for these users? Today you have very complex knowledge bases, hundreds of products, and so a user logs on to Poste GPT and has a response to make to a customer, either a commercial response or a customer relations response. So they log on, chat to Poste GPT in natural language and ask questions about a particular product, a particular customer is interested in a particular feature, and what Poste GPT does is connect to all these product sheets and summarise the information, still in natural language, but with links - and this is very important if we are to gain credibility for the solution - by linking all the product sheets that were used to produce this response.

[Frédéric Simottel]
By smiling at all the information?

[Pierre-Etienne Bardin]
Here, smiling at his answers.

[Frédéric Simottel]
You could criticise ChatGPT...

[Pierre-Etienne Bardin]
You could criticise it for giving us an answer and we didn't know where it was coming from. So to reassure users, we're saying, this is a summary of all the product sheets in relation to this customer request. But these are the files that correspond to this response. Of course, you can access these files via hypertext links, and then you can access the database that is the source of the LLM. So we rely on our in-house experts and that's how we can speed things up.

[Frédéric Simottel]
Fabrice, it's important to start with a POC, of course, but also to have something of a flagship application. We say, ‘Look, it's working here, now we're going to look at other directions.

[Fabrice Asvazadourian]
Exactly. I think we came up with a point of view at the beginning of the year on how to deal with generative AI. And we said ‘2024, acculturez, acculturez, acculturez’. So we had and we saw this era of POCs left and right, with the positive side of POCs, which is that it allowed us to get a bit closer to the real thing, and then the sometimes disappointing side of POCs, which is that it is ultimately difficult to move from in vitro to in vivo. So we told them that 2025 would be the time for flagships. Choose major areas in which you won't take the subject in bits and pieces, but from start to finish, because that's how you calculate a real business case. Because these days you can calculate ROI on anything. Which, at the same time, Excel knows how to do a lot of things, but it doesn't generate any gains that you can see afterwards in the profit and loss accounts. And so we can see that in the world of industry, the flagships today are more on the supply chain, if we want to see them it's more on supply chain optimisation. We talked about this on a previous programme. In the world of services, there's a lot of focus on assistants who are making productivity gains for employees. That's an interesting example from La Poste. Yes, these seem to me to be two moments, two major subjects that we're seeing industrialised at the moment.

[Frédéric Simottel]
And behind that, Pierre Etienne, you have to involve everyone in these projects. So how do you do that? Because there's the fantasy that ‘it's going to replace me’ and so on. So today we're saying ‘no, it's those who are going to use AI who are going to replace you. So go ahead, concentrate on ‘how did you manage to get everyone on board and then take AI to other jobs?

[Pierre-Etienne Bardin]
At La Poste, we've been working on AI for ten years now, and our approach is to say that we're all players and we're all affected by this transformation. We have 75,000 people certified in AI within the group, so it's something that was taken very seriously very quickly. And it's true that over the past year, we've seen an explosion in requests for acculturation and training to understand what this generative AI means and how I use it. And is there a risk to my job? And so two thousand training courses have been launched in one year, which is quite a lot. We worked a lot on acculturation. We are now moving on to more serious training, prompt techniques and hackathons on certain professions. And what we wanted to explain during these training sessions was to reassure people that the jobs at La Poste are very complex and very diversified. We're not talking about jobs that are going to disappear, but rather about activities within these jobs that are going to be automated and simplified, and that's the message we wanted to get across. And the other message we also wanted to get across is that there are a lot of opportunities behind it. Generative AI is an opportunity to access information more easily. And so with the example I gave.

[Frédéric Simottel]
Yes, having the flagship a bit

[Pierre-Etienne Bardin]
Before, it was complicated to access knowledge bases, it took up a lot of my time and sometimes I couldn't find what I was looking for. I had to call in an expert. But now I feel empowered because I'm the one who can access information quickly. So we have this opportunity. The second opportunity is for those professions that were looking at AI from a distance, saying ‘it's not for me’, but in fact they've been caught up by the patrol and we've now been able to launch programmes to accelerate the transformation of these professions. And the third opportunity is to say that we are all capable of creating content, we all have this possibility in a fairly simple way, to create documents, to create computer code, to create images, to create videos. I'm not saying that we all become artists, but in a way we all have this capacity to be augmented by this generative AI. So these are the three opportunities we're trying to push. These are the messages we're sending out. And obviously we won't be replaced by AI, but by those who know how to use it.

[Frédéric Simottel]
And be bold, that's what a study by Sopra Steria Next says. And it's about increasing the number of employees, showing that it's a disruptive innovation, but also an incremental innovation, Fabrice?

[Fabrice Asvazadourian]
Absolutely. I think that when we look at how we can convince employees to use AI, first of all, we need to do what you did at La Poste, which is to train them, because there's nothing worse than not knowing how to resist. So that's the first step. We see many of our customers using and supporting the deployment of AI through training in empathy and postures. Ultimately, the time that will be saved on tasks that can be automated needs to be redeployed to create more of a bond, more human warmth if you like. And as we've just come out of a fairly long period where we were driven by Compliance, high compliance, the softer qualities of the relationship need to be reasserted. And so we are seeing customers providing support, to demonstrate the full added value of the human element, when augmented by these AI solutions.

[Frédéric Simottel]
And on these programmes, at “AI Uncovered” we are also interested in trusted AI. And that's where you were able to talk about ethics, to say that we're here to support you. That's what we said. But what is your definition of trusted AI within the La Poste group?

[Banner ‘AI: the ethical challenge’]

[Pierre-Etienne Bardin]
For us, it's actually three dates: 2016, 2022 and 2024. 2016 is two years before the RGPD and it's the creation of a data charter that sets out the group's values. And to ensure that we have an ethical approach to the use of data. 2022 is two years before the RIA, and it's the same approach, but on AI and trusted AI. In 2024, these two charters will be merged to form an AI data charter that explains to employees and also to our partners how we use artificial intelligence. It's a set of principles. It was defined by a team of experts, a cross-functional team that brought together all the Group's business lines, including banking, insurance, logistics and support functions. It is a system of governance that makes it possible to control and verify the correct application of these principles. A governance structure that I co-chair with the Group's CSR Director, Stéphanie Dupuy-Lyon, which brings together the DPOs (the people in charge of compliance), data experts, legal experts and external experts who provide us with a viewpoint, a vision, a perspective on what is being done in other groups to help us go even faster and even further. We also have a whole range of control systems that will review projects involving the use of artificial intelligence to check that they comply with these principles and meet compliance requirements. But it also involves acculturation and training, and an analysis grid to check the risks associated with AI, because that's what the RIA is all about. There you have it.

[Frédéric Simottel]
And so this need to be a trusted player for the company. Just a word on that, Fabrice?

[Fabrice Asvazadourian]
So we know that this is a major issue and I think that Europe is ahead of the game in this area. We have to say so, because people often say that we are not ahead of the game, but on this subject, we are very much ahead of the game and we have to continue to stimulate the world to maintain a trusted AI environment.

[Frédéric Simottel]
Thank you both. Pierre-Etienne Bardin, Chief Data Officer, La Poste Group, Fabrice Asvazadourian, CEO of the consulting firm Sopra Steria Next. See you soon for a new programme entitled “AI Uncovered”.

[Voiceover]
Special edition. BFM Business Files

[AI Uncovered – Episode 6] Trust AI to support the evolution of lawyer professions

See the detailed description

[Fabrice Asvazadourian]
Absolutely. I think Jean-Godefroy said it very well: no data, no AI. Or in any case, ‘Bad data, no AI’, that's for sure. Let me remind you of an interesting statistic: we only have one algorithm out of seven that continues to have the same level of performance when it goes into deployment mode, because, among other things, we don't manage to have sufficient data quality when we're in industrial mode. So I can see three data challenges. The first is what data? And I think that today there is a real challenge for a lot of companies, especially companies that don't have a lot of data. There are the big banks, they have a lot of data, but many industries are entitled to ask ‘Where can I find data?’ That's the first question. Now there are lots of sources of data that we didn't think of. Let me give you an example.

[Frédéric Simmotel]
Or do you think they're all good? For example, in a law firm, there was no loss, I imagine? every piece of data counts. That's what Jean-Godefroy has just said.

[Fabrice Asvazadourian]
Secondly, there's the issue of data quality. And the good news is that there are AI solutions to check the quality of the data and help us to continue to improve on the subject of data quality. And then there's the third subject, which is, let's say, a bit old-fashioned, but which needs to be brought up to date with the issues specific to AI. It's all about the governance that a company has to put in place to do what people see. It means sorting, selecting, making sure there are data owners, and so on. Who are the people who have to ensure that the data remains of high quality?

[Frédéric Simottel]
So how do we work around all this? Jean-Godefroy, at FIDAL, I said it's 2,000 lawyers, there are more senior people and also young lawyers who arrive and who have different attitudes to AI. And then you really put... one of your priorities is learning around all this?

[Jean-Godefroy Desmazières]
Yes, it's experimentation and learning. Ultimately, we're working with a tool that's going to change the day-to-day lives of our professionals in general, and in different ways for different categories of professionals. A lawyer who is more senior, who is a partner in the firm, will be able to use this tool but won't necessarily trust it directly. When you use AI to produce content. In the end, you're going to have a level of efficiency that may be a little lower than when you work on pre-existing documents. We also have young professionals to train, and these young professionals have been training for years, for decades. In other words, we ask them a question. They were going to look for answers, they were going to get lost, they were going to be able to come up with solutions. They would come up with an answer that we would fine-tune for our customer. Today, we have a slightly different system. In other words, when we ask a question to one of the young professionals who work with us, they will be able to ask this question to the tool and get an answer immediately. And with that answer, rather than looking for an answer they don't yet know, they'll try to validate the answer we've given them. This is an absolutely central element, as it means that professionals will no longer lose their way, in other words, they will no longer have the opportunity to invent and create the law of tomorrow. AI works on documents that already exist, on data that already exists. So AI is looking in the rear-view mirror. What we are interested in for our clients is also inventing the law of tomorrow, developing case law, giving advice that is of a different quality because it is off the beaten track. And to do that, we need ingenuity, we need inventiveness. So we're going to have to train our professionals in AI, and also train our young people to develop a critical mind despite AI.

[Frédéric Simottel]
Yes, it's easy on the one hand, but it's not easy on the other.

[Jean-Godefroy Desmazières]
Exactly.

[Frédéric Simottel]
This side is important. And when it comes to learning tools, there's an interesting initiative that Jean-Godefroy and I were talking about when we were preparing this programme, a prompt library, Fabrice?

[Fabrice Asvazadourian]
Yes, that's right, it's a very good practice that we recommend because I see it with my customers and I see it with my consultants. Initially, everyone wants to have their own FIDAL IA or ‘X-GPT’. And then they deploy it, install it and all that, and then they look at it. And then there are those who use it every day. Because they naturally want to, they're resourceful. There are those who don't use it much, sporadically, for very specific things. And then there are those who refuse, or who haven't got the hang of it. Or don't consider a use of AI to be part of their life. It's not a way of saying ‘there's a tool out there, use it’, it's a way of saying ‘we have specific uses, here's how to get into them’. And then you can re-prompts, refining and specifying. And so we very strongly recommend that our customers, in order to anchor their employees' daily practices, have libraries of prompts and then let the creativity of the in-house communities enrich, enhance and complete them. These prompts. You have to crutch the entrance. For the vast majority of employees, you need a crutch.

[Frédéric Simottel]
I found this initiative interesting, because what we often wonder about is the front door. Once you've asked a question, you've got the answer, but have you really asked the question in the right way? And I thought that this library of prompts was an initiative that you at FIDAL could certainly emphasise. All these uses are also transforming the profession and the business model of your firm. But also of the legal profession. Generally speaking, Jean-Godefroy Desmazières

[Jean-Godefroy Desmazières]
Yes, absolutely. In fact, this AI approach means that we can go much faster in aggregating knowledge that was previously scattered. So that leads us to two things. It leads us to question our business model. I'll come back to this point. It also leads us to ask ourselves what the profession of a business lawyer really is. Because when it comes down to it, a lawyer's added value when it comes to aggregating scattered knowledge is not very high. And because it's not very high, we're going to find it hard to sell it to our clients. Until now, and it's been like this for years in most law firms, our profession, our turnover, has been a matter of adding and multiplying, of adding time and multiplying it by hourly rates and finally arriving at a value. But it is almost exclusively in these professions that the value of a service is linked to the time taken to produce it. If you look at a mobile phone, you don't ask yourself how long it took to build it. We look at how useful it is, how effective it is, how innovative it is. That's exactly what we're doing, using time-saving tools. In the end, we spend less time on our customers' questions and if there is less inventiveness, this multiplication of less time by a rate leads to less sales. So the question is either Yes, or we have a lot more customers and we can get by. If the modernisation of our businesses results in a drop in sales, then we're missing the point. So we want to provide a high-quality customer service that's as innovative and modern as possible, while at the same time focusing on the finer points of value for our customers. And this value is served by AI, but it's not just AI, so it's a tool for our professionals.

[Frédéric Simottel]
And so today, are you reviewing your model in relation to the use of this AI?

[Jean-Godefroy Desmazières]
Today, we work with our customers on the value that is provided, regardless of the time taken to provide the service. And do you already have customers who come to you and say, ‘Look, generative AI is the same as what I've been able to do as a customer’, and then come back to you and say, ‘There are easier things to do’? It really depends on what the customer or our customers want. We have 69,000 customers, so we don't have all the same 69,000 points of view, but it's still a significant number. Some of them come to a law firm to get perfect service. Today, AI helps us, but it doesn't provide a perfect service. So when we produce content, a document or a deed, we use digital tools, whether they be tech contracts, AI with Fidal IA, or other tools. In reality, we always have to rework the whole thing to ensure that it's as perfect as possible, i.e. in line with the rules, the rules applicable to the law, case law, etc. But above all, it has to be tailored to our client's needs. So obviously, our clients are going to tell us you're going to save time on certain tasks. Low added value. It's up to us to demonstrate that 100% value is something that matters to us.

[Frédéric Simottel]
And here we come to the theme of our show: trusted AI. As Jean-Godefroy Desmazières has just said, every company is becoming a trusted player for its employees and customers.

[Fabrice Asvazadourian]
Yes, AI will be trusted or it won't be. So this is a critical issue. At the same time, it's not a new issue, because companies have always known that if they don't have the trust of their employees and customers, they'll have trouble surviving. So... But here, the stakes are a little higher, since...

[Frédéric Simottel]
Yes, now we have a tool that sometimes does our thinking for us... Yes. So we have to recruit. So, there is, there will be the compulsory floor and the AI act will create. The seven dimensions of trusted AI that everyone will have to face up to. They're common sense. I don't know if I can name all seven off the top of my head, but they're common sense. They're going to become obligations. So we're going to have to, we're going to have to demonstrate that we've effectively put in place within our organization the capacity to manage a trusted AI. And then, I think there's a second stage where we'll go beyond that and say, “How can we really implement trusted AI in these critical business processes? We're taking part in a consortium called Confiance.AI, alongside seven other major partners, which we're working on. How will we be able to handle business-critical processes in the future with trusted AI and with confidence in relation to an environment, sensitive data, etc., etc.?

[Frédéric Simottel]
Well, thank you both for coming to talk to us about all this, about this trusted AI. Fabrice Asvazadourian, CEO of consulting firm Sopra Steria Next and Jean-Godefroy Desmazières, Deputy Managing Director of law firm FIDAL, and then in charge of this digital transformation, this AI transformation. Thank you both for sharing your stories with us. We'll be back very soon for a new episode in our series “AI Uncovered”

[Voice-over]
Special edition. BFM Business Files.

[AI Uncovered – Episode 8] – AI facing productivity and sovereignty stakes

See the detailed description

[VOICEOVER] Special edition, BFM Business Files. “AI Uncovered: Driving the AI revolution” with Frédéric Simottel.

[Frédéric Simottel]
Hello and welcome to this special edition of AI Uncovered, in partnership with Sopra Steria Next. We are exploring the behind-the-scenes of generative AI and its impact on businesses. Over the past few months, you have been able to watch replays covering sectors like luxury, industry, and banking. Today, we are focusing on a truly strategic domain – submarine cables – with our guests. Let me introduce them: Fabrice Asvazadourian, hello.

[Info bar: Time for industrial 5G]

[Fabrice Asvazadourian]
Hello, Frédéric.

[Frédéric Simottel]
Managing Director of the consulting firm Sopra Steria Next. More than 4,000 consultants today, and of course, a subsidiary of the ESN Sopra Steria, which we know well. And with us, Christophe Bejina.

[Christophe Bejina]
Hello.

[Speaker bar: Christophe Bejina: Chief Information Officer of the ASN (Alcatel Submarine Networks) group]

[Frédéric Simottel]
Hello Christophe. CIO of ASN Group. People now just say ASN, but let’s clarify – Alcatel Submarine Networks. Otherwise, there are plenty of acronyms! Christophe, let’s start with you. ASN is an industrial company, a giant in submarine cables. For over five years you have significantly digitalised the group. Your digital transformation is well underway: a complete overhaul of your infrastructure, integrating industrial 5G. We’ll also talk later about this important year for you, as the French state has taken back 80% of ASN’s capital. But first, tell us about this industrial 5G.

[Christophe Bejina]
Of course. Industrial 5G is part of a digital transformation programme that began five years ago, as you mentioned. It’s a four-phase programme. The first phase was to lay the foundations of this transformation – starting with network infrastructure and data centres, hence 5G, and setting up a partnership ecosystem to support us on this journey.

The second phase was implementing or integrating an enterprise architecture – an information system architecture geared towards Industry 4.0, while also putting in place principles of securing industrial IT systems, which is becoming increasingly important.

[Frédéric Simottel]
So, it really was necessary to renovate, to start from scratch and rebuild.

[Christophe Bejina]
Exactly. Renovation was crucial, firstly because this is a historic industry with legacy systems that had to evolve. Secondly, there is a boom in the business, and the nature of contracts and projects has reached a level of complexity that can no longer be handled with the old tools.

So, this was a major imperative – a fast-paced digital transformation.

The third and fourth phases involve deploying what we call data and AI.

[Frédéric Simottel]
That’s right, because above this infrastructure – you’re building it in parallel – there’s a whole data infrastructure as well.

[Infographic: Alcatel Submarine Networks (ASN)
• 2,000 employees
• 7 vessels
• 410,000 km of cables under maintenance
• 180 systems installed]

[Christophe Bejina]
Exactly. We need to refine the data, expose it securely, and use it for AI. The final phases we are in now involve deploying sovereign computing capacity – that is, computing power within our own data centres – to gradually train AI models on our data without exposing them.

That is the roadmap: implement the digital transformation so that, in the end, we can fully exploit data and enable the advent of AI.

[Frédéric Simottel]
And this industrial 5G – it’s now a key element for you?

[Christophe Bejina]
Absolutely. It is progressively becoming the data backbone of our factories. More and more, we are connecting to 5G rather than fibre. All industrial security systems will eventually be connected via 5G. It becomes the nervous system enabling IoT deployment, smart systems, new robots, energy control systems, and other intelligent devices across our plants.

[Frédéric Simottel]
We are seeing this everywhere. Fabrice Asvazadourian, just a few days ago was the Mobile World Congress in Barcelona. There was a lot of talk about 5G and industrial 5G. I’ve always felt that operators somewhat mis-sold 5G to the general public, while in industry things are really picking up speed.

[Fabrice Asvazadourian]
Absolutely, Frédéric. I believe 5G is not simply the next step after 4G. It was designed first and foremost to meet industrial needs. What does it bring? In factory environments, it ensures there are no more network problems, delivers much faster bandwidth, drastically reduced latency – enabling true real-time operations.

It allows decentralisation within factories, and coupled with edge computing, it brings computing power closer to the data. This helps to move faster and meet the extremely low-latency demands of industrial processes. Additional security layers are also crucial. So 5G combined with edge computing becomes the enabler of Industry 4.0, multiplying sensors and IoT, allowing machines to communicate, and optimising flows and supply chains – even between suppliers. This is a major productivity driver.

[Frédéric Simotel]
Exactly – Christophe, ASN had a significant year, as I mentioned. The state took back 80% of your capital, because ASN is clearly a strategic industry. When we talk about AI, we talk about data – and when we talk about data, we talk about sovereignty. Does state ownership heighten the focus on sovereignty issues?

[Info bar: AI: the challenge of sovereignty]

[Christophe Bejina]
Yes, absolutely. Being state-owned – or at least having the state as majority shareholder through the APE – strengthens the sovereignty focus. But this was already a priority when we began our digital transformation. This simply accelerates the process.

Sovereignty brings two key points of focus: First, when deploying AI and 5G, we must remember that everything runs on data. Data from a strategic industry like ASN cannot be shared or exposed. So, the first priority is to work with the right partners to build this data and AI ecosystem – ideally French or European partners, since we also have plants in Norway and the UK.

The second point is building a sovereign infrastructure: ensuring computing power is located in our own or our French partners’ data centres, and that applications are hosted on-premises, with data remaining inside our systems and not exposed externally.

[Frédéric Simottel]
That sovereignty criterion is important – and not just for strategic industries, given the current geopolitical climate.

[Fabrice Asvazadourian]
Exactly. Three or four years ago, this was an intellectual discussion – except for defence, the army, etc. – but sovereignty was not really a key decision factor. Today, every CIO, across all industries, considers sovereignty when choosing cloud providers, AI solutions, and other technologies.

France is, in my view, ahead of the curve here. Europe has put in place defensive regulatory mechanisms, but perhaps we are not yet offensive enough. Hopefully, the billions announced just last month will help us take a more proactive stance. The President’s strong emphasis on Mistral, for instance, should prompt French and European companies to think carefully about how to make sovereignty a lived reality – because ultimately, it’s about making choices.

[Christophe Bejina]
Yes, and it is a constraint.

[Frédéric Simottel]
On AI specifically, could you share one or two projects that have scaled successfully?

[Info bar: AI: scaling up]

[Christophe Bejina]
There are several, but I’ll highlight two or three. First, in our R&D teams, we already use machine learning to develop our systems. Ultra-high-speed optical networks are very complex, particularly in terms of resource management and filtering capacity.

We have AI compute capacity – using those famous Nvidia GPUs – to solve these extremely complex problems in our R&D products. These systems process 60–70 terabytes of data – so we are operating at a whole new scale.

A second example: we run highly complex marine operations. Cable deployment is not just maritime transport – these are factory-ships travelling back and forth. We faced serious planning complexity for these operations. This year we implemented an AI-based system using metaheuristics to handle marine planning – a task that had become too complex for human intelligence alone.

[Frédéric Simotel]
Fascinating. AI is clearly a productivity driver – and even a driver for reindustrialisation.

[Info bar: AI becomes a productivity driver]

[Fabrice Asvazadourian]
Absolutely. Eighteen months ago, together with ASN, we were proud to win a major innovation award for deploying 5G in industry. Because ultimately, the goal is competitiveness.

The combination of technologies – 5G, AI – allows us to deliver higher performance. One interesting example is using AI to improve circular manufacturing: reusing waste as an input for production, rather than relying on chemistry or raw materials sourced from far away. AI models help boost performance in ESG objectives, sovereignty, and competitiveness simultaneously. The reindustrialisation of France will necessarily rely on these technologies to restore competitive advantage.

[Frédéric Simottel]

That’s the positive message we wanted to share, particularly in these times of geopolitical tension. Our industry is strong – and will become even stronger thanks to AI. That’s what we’ll keep promoting on our programmes, alongside industrial, service, and consulting players.

Thank you for joining us. See you soon for another episode of AI Uncovered. And remember, you can watch all the other episodes on replay with our partner Sopra Steria Next. Have an excellent day.

[VOICEOVER]
Special edition, BFM Business Files.

[AI Uncovered – Episode 9] – Business models facing AI

See the detailed description

[BFM Business Logo, text appears: “Special Edition, BFM Business files” then “AI Uncovered: Leaders deploying AI with confidence.”]

[Voiceover]

Special Edition. BFM Business files. "AI Uncovered: Leaders deploying AI with confidence," with Frédéric Simottel.

[Frédéric Simottel]

Welcome to our show "AI Uncovered: Leaders deploying AI with confidence," in partnership with Sopra Steria Next.

We will obviously talk about the implementation of AI, generative AI in companies, and we will focus more specifically on AI within a global group that has a multitude of brands. Digital technology has long been a strategic focus for this group, and we will discuss it with you, Hélène Chaplain, CIO of the Pernod Ricard group.

Hello.

[Hélène Chaplain]

Hello.

[Frédéric Simottel]

Thank you for joining us. And also with us is Fabrice Asvazadourian.

Hello.

[Fabrice Asvazadourian]

Hello Frédéric. Hello Hélène.

[Frédéric Simottel]

Fabrice is the CEO of the consulting firm Sopra Steria Next. So, who exactly is this group? Four thousand consultants, right?

[Fabrice Asvazadourian]

Exactly.

[Frédéric Simottel]

And obviously a subsidiary of the IT services company Sopra Steria.

Hélène, I’ll start with you. Beyond productivity, how does Pernod Ricard use technology today, especially artificial intelligence, to promote its products, generate demand? And how do you convince yourself that you’re making the right technology choices?

[Hélène Chaplain]

That’s a very broad question, but maybe to start, the strength of our model, or our group, is indeed to rely on two very structuring and demanding competitive advantages: a very wide brand portfolio and, on the other hand, a very broad distribution presence.

So, in addition to the quality of our spirits, we need to work on visibility and value proposition to bring the right brand through the distribution model to our consumers.

Overall, we have worked on this visibility in a very granular, structured, and advanced way by using AI to define again within the brand portfolio and our multichannel presence, what is the right combination for a given consumer and ultimately for each individual consumer.

What we saw come as a wave of disruption initially was that we had to get referenced on Google, so it was standard SEO, whether organic or acquisition. Then we saw that Amazon actually became the product search engine because we have consumable products, so we had to be visible there. That involved a commercial relationship with Amazon or its marketplaces to be available with the right value proposition, the right content, the right message, the right price, at least, with the consumers.

And what we see today is this new disruption with a new intermediary in the consumer experience, in the relationship with the consumer via generative AI.

So we are fundamentally disrupted or impacted, or at least very vigilant to bring our brands — since we are fundamentally intermediated — to bring our brands and work with all intermediaries. Generative AI is a new intermediary in the landscape.

[Frédéric Simottel]

And we will even talk about the new wave behind generative AI, which is the agentic wave. Fabrice, how can AI lead to changes in business models?

[Fabrice Asvazadourian]

First of all, I think everyone is convinced it will take some time, and slower than what we might have dreamed two years ago.

But, AI will enable companies to generate productivity gains by optimising processes beyond what is already embedded in company systems, which is probably the primary challenge. The cultural adoption phase is, in my opinion, largely behind us.

[Frédéric Simottel]

Yes, yes.

[Fabrice Asvazadourian]

That is probably a victory never fully won, but well. Today we are entering the phase of AI very focused on productivity, on all administrative tasks which today have enough added value, but are quite reusable by not super sophisticated AIs. That is the basic quantity-driven AI.

Then, as Hélène mentioned, AI is not just that.

[Hélène Chaplain]

No.

[Fabrice Asvazadourian]

It is also disruption. What I mean is optimisation — companies are used to optimising. Big companies, in particular, are optimisation machines. Disruption is more complicated. Disruption will come in value propositions, in how to showcase these products that can be fundamentally disrupted. This is where we will see real innovation AI will bring.

The key question for a leader is not to miss the first wave, which is essential — the one that generates the first ROI. Today everyone questions the ROI of their investments.

[Frédéric Simottel]

Yes, even if sometimes there is a disappointing effect for some...

[Fabrice Asvazadourian]

Yes, it is hard; it’s harder than what was promised.

[Hélène Chaplain]

It’s not magic.

[Fabrice Asvazadourian]

So, it’s not magic. Even though we discovered something magical two years ago. So we have to keep doing this, and in my opinion, this is the very short-term challenge to keep investing because the sums involved in budgets are still significant.

On the other hand, we must remain alert and active on these disruptions that are coming and that are simple but not easy to fully integrate into these models.

[Frédéric Simottel]

And precisely, I was saying to Hélène Chaplain that the agentic wave is arriving behind this generative AI, with its power. For you, it could change a lot, within your offerings, your own works, the usage, the impact it can have for user companies like yours.

But it also has an effect on the IT or digital suppliers you deal with.

[Hélène Chaplain]

Exactly. Beyond productivity and competitiveness it can generate, there are other implications because, ultimately, the users of generative AI are consumers.

That means in interaction with GPT, the user will provide information about who they are, their feelings, emotions, seeking an experience, just by simply asking questions, they give a lot of information. What happens in this new intermediary, GPT or other AIs — OpenAI just announced opening their e-commerce section.

So, it’s significant. We observe this new intermediary which presumably has very advanced knowledge of the user, who is also a consumer with knowledge, and will propose, push or not our products. How it selects, what content it pushes, if we talk about e-commerce, which channel it uses — these consumer interactions are things we do not control because we don’t know what data underlies this routing.

So this creates a very strong implication: a new intermediary we have to address, prepare for, and more than prepare for, since it’s already here. It’s like before, referencing on Google; now it’s the ability to exist with our content, offers, and brands — but without knowing how decisions are made. This has implications, and I wouldn’t call it a risk, but rather an opportunity, on our value chain.

[Frédéric Simottel]

Yes, because that’s the difficulty: usually when we hear companies using generative AI, it’s within their own environment, with data they control.

You, at Pernod Ricard, have many brands, distributors, and so on. You cannot control all these brands.

[Hélène Chaplain]

No, and also, some data is generated by our users, by our consumers — UGC (User Generated Content). We have data generated by third parties, by our distributors. And since models source data we don’t own, or even if we do, it has been exposed, we don’t know exactly which data is used as a source and how the value proposition is formed.

We all seek experiences and offers that are individualised and personalised. In this case, we don’t know the user who has become a consumer, and we don’t know how the algorithm or the model operates and what it is sourcing to finally push one thing, push our product, or push another product.

[Frédéric Simottel]

So, Fabrice, does this agentic wave mean there will have to be agents, agents of agents, super agents, right? Orchestrators?

[Fabrice Asvazadourian]

Orchestrators, commanders, captains so agents communicate well.

Yes, we already see it today. When you see a fleet of drones flying synchronously, they talk to each other and interact — so we see it maybe in a more applied way.

There will be agents working in warehouses managing entries and exits who will communicate with agents in charge of logistics optimising fleet routes.

So, they must communicate. The question is: at some point, who decides?

[Hélène Chaplain]

The super agent.

[Fabrice Asvazadourian]

For large companies, super agents or commanders will decide when different agents have proposals or inputs that are potentially conflicting.

I think that will be the future challenge.

What Hélène said is very insightful because when Google was there, when Amazon was there, large companies learnt how to decode how to influence and position themselves well.

[Hélène Chaplain]

Yes, organically or by acquisition.

[Fabrice Asvazadourian]

Here, we have much blacker boxes. Before understanding algorithms which say: because you have such a feeling, or such information we don’t know well, we propose this or that. We will have to learn this.

So, the sooner we start testing, the faster we will learn.

And then, I think we will see what strong brands do — they always manage. Average brands will struggle a lot.

[Frédéric Simottel]

We will see that on the market. So, in all this, Hélène, what should be the role of the CIO, the IT director in this orchestration, this organisation?

[Hélène Chaplain]

It is very broad.

First, we talked about disruption for actors like Pernod Ricard, but also disruption among our suppliers, the software industry.

We see that they built themselves around process structuring, selling productivity — I won’t name the German company — with extremely rigid processes governing many companies.

But AI brings disruption to this very monolithic vision of processes, creating the opportunity to generate code on the fly, flows of interface, even synthetic data to feed the model.

So, there is a potential rupture among our suppliers and partners, and we must make choices.

And the first point I want to recall is that in these choices, the main point is the technology investment strategy of a company — at least a CAC40 group.

It is extremely significant because it usually sits around 3% of turnover.

When you invest 3% of turnover every year, you are manoeuvring to correlate technology investment strategy with what it should bring in productivity and competitiveness internally, but also in what it should enable in terms of value chain disruption. We talked about the e-commerce business model on ChatGPT.

So, we have very high requirements. We must treat investment strategy as if it were at the core of the company’s engine. We must also choose technologies that have a foundational character.

Again, these are significant sums committed for the long term because we normally build things to last and never start from scratch.

So, the agent, the super agent, the agent of agents, we must build around what we call legacy, which is far from modern.

We must consider the complexity of the information system, its integrity.

Obviously, we must scale all that.

Before this interview, we talked about defining guiding principles, evaluation principles for self-financing loops.

We must connect technological strategy and business strategy complementarily — one no longer serving the other.

[Frédéric Simottel]

So the CIO also becomes Chief AI Officer at the same time.

[Fabrice Asvazadourian]

Yes, well, we should ask Hélène what she thinks, but if we look at history, we’ve had generations of Chief Digital Officers, and today you don’t hear much about them anymore.

So now, we need to have Chief AI Officers. This is the moment because companies need focus; otherwise, we can’t structure today.

We see many of our clients building AI factories to concentrate all resources.

In five years, we will talk again.

[Frédéric Simottel]

Thank you both for coming to talk about all this in this show, "AI Uncovered: Leaders deploying AI with confidence." Thanks to Hélène Chaplain, CIO of Pernod Ricard group, and Fabrice Asvazadourian, CEO of the consulting firm Sopra Steria Next. See you soon for a new programme.

[Voiceover]

Special Edition. BFM Business files.

[AI Uncovered – Episode 10] – Strategic integration of AI

See the detailed description

[Voice-over] BFM Business special feature
“AI Uncovered, driving the AI revolution” with Frédéric Simottel.

[On-screen caption: How SNCF integrates AI into its strategy]

[Frédéric Simottel]

Welcome to our programme “AI Uncovered”. In partnership with Sopra Steria Next, we are going to talk about the implementation of AI and generative AI, about trusted AI within companies, and we will focus more specifically on AI within a transport giant in France and worldwide. Because yes, SNCF operates all over the world, and we are going to discuss this with our experts.

Nicolas Conso, hello!

[Nicolas Conso]

Hello.

[Frédéric Simottel]

Nicolas, thank you for being with us. Director of the transformation of information systems at SNCF Voyageurs, which includes TGV, Transilien, TER and maintenance activities. That represents 65,000 people and a turnover of 20 billion euros. Just to give a sense of scale, in very broad strokes.

And with us as well, Fabrice Asvazadourian, good evening. Uh, good morning Fabrice, sorry!

[Fabrice Asvazadourian]

Hello Frédéric!

[Frédéric Simottel]

Managing Director of the consulting firm Sopra Steria Next, which is Sopra Steria’s consulting entity: 4,000 consultants and 56,000 employees across the whole group, and around 6 billion euros in turnover.

Nicolas, AI has become a reality. We see it, we see it without really seeing it. For us as passengers using SNCF Voyageurs. But you use it both for predictive maintenance and to better understand your 20 million customers. So how do you use all this information, given the massive amount of data coming from your customers, and how does AI help you sort through it all?

[On-screen caption: Nicolas Conso – Director of IT Transformation, SNCF Voyageurs]

[Nicolas Conso]

Yes, Frédéric. AI at SNCF Voyageurs has a long history for two reasons. First, because at SNCF Voyageurs we have had a strong digital culture for a long time. SNCF Connect is one of the first e-commerce sites selling tickets for our 15,000 trains per day.

Second, we do indeed have great wealth in the era of AI: our data. The 22 million customers generate almost 250 terabytes of data every day. Just imagine everything we can do to analyse and optimise customer services with that. And there is also maintenance data. The new TGVs that we will have from next year will have almost 2,000 sensors, so you can imagine all the data generated by these connected trains.

[Infographic: SNCF Connect supporting passengers]

Visits: 1 billion
Unique visitors: 32 million
Train tickets sold in 2022: 190 million (60% long-distance)
Source: SNCF Voyageurs

[Frédéric Simottel]

And how do we, as customers, see and feel all this as users?

[Nicolas Conso]

There are three areas. First, of course, improving customer relations. I would say this is a classic in companies, but we use it very deliberately, particularly for chatbots, to support agents working with customers in call centres and to facilitate responses.

[On-screen caption: Leveraging data for better services]

We also used it, for example, when we welcomed the whole world for the Olympic Games. We gave each agent wearing the purple vests (I was wearing one too) a tool that allowed translation into more than a hundred languages. So AI also helps deliver better service.

That is customer relations. Another very important area for our passengers is passenger information. AI allows us to better predict delays with greater accuracy. When, unfortunately, a wild boar meets a train, we can assess the consequences and the delay, and in the future provide more personalised information depending on the situation. Passenger information is a very strong demand from our travellers.

[On-screen caption: How SNCF integrates AI into its strategy]

Another major customer demand is simply to have more trains available. As you mentioned, Frédéric, predictive maintenance, using all the data we have, allows us to optimise maintenance, meaning aiming for zero breakdowns.

This means detecting problems early, repairing trains quickly, or even extending maintenance cycles so trains can run longer. This allows us to offer more trains, more seats, and meet demand in France and across Europe.

[Infographic: TGV, industrial excellence]

• 35 technical centres, including 10 industrial maintenance sites
• 17,000 trains maintained and 450,000 parts repaired per year
• One visit every three days on average per train to a technical centre
• 40% of the TER fleet fully refurbished within 10 years (OPTER programme)
• 15 patents filed per year
Source: SNCF Voyageurs

[Frédéric Simottel]

Fabrice, at the heart of all this, customer relations and the fact that more trains can run thanks to predictive maintenance, data is central.

[Fabrice Asvazadourian]

The figures Nicolas mentioned are impressive: 20 million customers, 15,000 trains, new TGVs with 2,000 sensors. One has to imagine the volume of information to process. For data to be usable, it must be well organised and well governed to truly become useful.

What is interesting is how a company like SNCF builds a data architecture, platforms, solutions and tools around data, as well as processes and governance, so that all SNCF employees can access this data and do their jobs better.

[On-screen caption: SNCF Voyageurs in the era of generative AI]

That is why this example is interesting. It is because of this foundation that AI can be deployed. You can deploy as many models as you want, but without this base, AI would not deliver additional value, whether for predictive maintenance, passenger information optimisation or call centre support.

[Frédéric Simottel]

Today at SNCF Voyageurs, you are obviously using generative AI. After the discovery phase and initial experiments, where are you today?

[Nicolas Conso]

We take a proactive but reasonable and responsible approach. Reasonable because there is hype, and we want to invest in structuring elements with ROI.

We have made SNCF GPT available to all employees: a multi-LLM tool that allows use cases to be developed and identified. It saves between one and three hours per week, depending on usage, which is fairly typical.

We also run an acceleration programme for high-ROI use cases, particularly in support functions such as HR, finance, legal and procurement, where there is a lot of documentation and strong potential.

Another area is assistants for staff, whether customer-facing agents or maintenance operators. For example, when a door is broken, an operator can use a tool based on historical data to identify the cause of the fault and how to repair it. This assistant helps them work faster and more efficiently.

[Frédéric Simottel]

This replaces informal knowledge management from more experienced staff. We clearly see the value of generative AI. What we see with SNCF Voyageurs, Fabrice, is determination combined with caution, focusing on high-ROI use cases that benefit the company, employees and customers.

[Fabrice Asvazadourian]

Yes, there is that balance. We see it at SNCF and elsewhere. On the one hand, widespread upskilling, as we are only at the beginning of the AI and generative AI wave. Employees need to want to adopt it and get used to working with it, so that when truly transformative use cases arrive, they are ready.

On the other hand, we are now seeing generative AI deployed at scale, mainly through process optimisation rather than new offerings, which will come later. Providing AI tools for everyone, such as SNCF GPT, alongside process optimisation, is what many large organisations are doing today.

[Frédéric Simottel]

You also mentioned responsible AI. Are your employees adopting this mindset? We are all tempted to turn to ChatGPT, Mistral Chat or Perplexity for every question.

[On-screen caption: When SNCF raises awareness of responsible AI]

[Nicolas Conso]

This is very important for us because we are the most decarbonised mode of transport, so we must be exemplary in all areas.

[Infographic: Train as a sustainable mobility solution]

CO₂ equivalent emissions per passenger/km
Train: 11g
Electric car*: 58g (x5)
Petrol car*: 146g (x14)
Plane: 260g (x65)
*1.6 passengers
Source: SNCF Voyageurs

[Frédéric Simottel]

You are a major electricity consumer…

[Nicolas Conso]

And we must also be exemplary in our use of digital technology. AI helps us further reduce emissions, for example by optimising train driving. A train weighs 400 tonnes at 300 km/h, with very little friction. Using gradients, slopes and algorithms, we can optimise driving and save up to 10% of electricity.

As the largest electricity consumer in France, around 8 TWh, roughly the output of a nuclear power station or a city of four million inhabitants, this is significant.

We also want everyone to be responsible. In SNCF GPT, we display the financial and carbon cost of every query. Each employee sees their monthly and annual footprint, encouraging responsible use for genuinely useful purposes.

[Frédéric Simottel]

In conclusion, with SNCF Voyageurs we see a comprehensive example of AI, with clearly defined priorities for employees, business growth and customers.

[Fabrice Asvazadourian]

Absolutely. It shows determination combined with control, particularly regarding environmental impact. AI will be part of the solution, provided it is used properly, with the right models for the right needs.

[Frédéric Simottel]

And sovereignty is also a key issue.

[Nicolas Conso]

Yes. We announced a partnership with Mistral at VivaTech. Using responsible, sovereign AI is important, especially with smaller models.

[Frédéric Simottel]

Thank you. Nicolas Conso, Director of IT Transformation at SNCF Voyageurs, and Fabrice Asvazadourian, Managing Director of Sopra Steria Next.
AI Uncovered, driving the AI revolution.

[AI Uncovered – Episode 11] – Trustworthy AI

See the detailed description

[Voice-over]

BFM Business special feature

“AI Uncovered, driving the AI revolution” with Frédéric Simottel.

[Frédéric Simottel]

Welcome to AI Uncovered, this BFM Business special feature, “driving the AI revolution”, with our partner Sopra Steria Next.

Today, as you can see in replay, we have explored many sectors: luxury, industry, banking…
We have covered a very wide range of sectors, and today we are going to focus more specifically on the world of aeronautics.

The use of trusted AI in critical systems, as you can imagine, to ensure greater safety and reliability, both in civil and defence environments.

With us to discuss this are our two experts. David Sadek, hello David.

[David Sadek]

Hello Frédéric.

[Frédéric Simottel]

You are Vice President Research, Technology and Innovation and AI CTO at Thales.

And with us as well, Fabrice Asvazadourian, hello.

[Fabrice Asvazadourian]

Hello Frédéric.

[Frédéric Simottel]

Fabrice, our expert consultant, Managing Director of the consulting firm Sopra Steria Next, 4,000 consultants, a subsidiary, of course, of the ESN Sopra Steria.

David, first question, a fairly general one: what characterises AI at Thales? As I said in the introduction, it has to be highly demanding, because we are dealing with critical systems. What does that mean in practice?

[On-screen caption: Thales focuses on reliable and secure AI]

[On-screen caption: David Sadek – Vice President Research, Technology and Innovation, Thales]

[David Sadek]

I usually say that when Netflix recommends a film you do not like, it is not dramatic. You choose another film and life goes on.

Now, if you take an aircraft, knowing that two out of three aircraft worldwide take off or land using Thales systems, either because there is what we call an FMS on board, a Flight Management System, the brain of the aircraft, or because the aircraft is handled at some point by Thales air traffic management systems, if you put AI into these systems, you had better be able to trust it.

You cannot say, “the aircraft will roughly land”. It must land. To embed a component in an aircraft and certify it, you must prove that the probability of incidents per flight hour is lower than 10 to the power of minus 9 or minus 6, depending on whether we are talking about a system or a component. Roughly speaking, that is 0.0000000001. Extremely small.

So AI must also meet these requirements and comply with them. That is why, for several years now, we have introduced at Thales a fully structured trusted AI approach for critical systems.

[Frédéric Simottel]

This is AI that is trained and trained again continuously, is that right?

[David Sadek]

Of course. It is AI trained on datasets, but not only data, it can also combine data and expert knowledge.

Above all, it is AI that must comply with validity requirements, which are the foundation of operational safety. Validity means proving or guaranteeing that the system does exactly what it is expected to do, no more, no less, and being able to prove it. This is not declared safety; it is proven safety, which is exactly what certification is about.

It must also meet security requirements, meaning resilience to malicious attacks, especially cyberattacks. At Thales, we have teams dedicated to what we call “friendly hacking”, attacking AI algorithms to identify vulnerabilities and propose countermeasures.

This team even won a challenge organised by the French Defence Procurement Agency about a year and a half ago, where they managed to recover training data used by an AI system. More recently, they developed techniques to detect fake images generated by AI. Security is therefore the second pillar of trusted AI.

The third pillar is explainability: the ability to explain and justify why an AI did what it did or made a recommendation. At Thales, we even push for explainability in operation.

For example, if a digital co-pilot recommends that a pilot turns 45 degrees in 30 miles, the pilot must be able to ask, “Why should I do that?”, especially if they were considering doing something else. And the system must be able to answer: because there is a threat, because there is a storm, not because neuron layer number three was activated at 30%, which is not an explanation.

Finally, the fourth pillar, and a crucial one, is responsibility. This includes detecting and minimising bias in data and algorithms, data confidentiality, regulatory compliance, and concern for environmental footprint.

This is where Green AI comes in: designing AI technologies that minimise energy consumption. We work extensively on what we call frugal learning, minimising data requirements, and also on using AI itself to reduce the carbon footprint of applications, particularly in aeronautics.

[Frédéric Simottel]

Clearly, there is no room for error in many companies, but especially at Thales with critical systems. Fabrice, are these four points the key priorities for trusted AI?

[On-screen caption: Trusted AI – what are the priorities?]

[On-screen caption: Fabrice Asvazadourian – Managing Director, Sopra Steria Next]

[Fabrice Asvazadourian]

Absolutely. France is fortunate to have several major industrial players that are globally ahead in applying AI to critical processes.

We have built a consortium with Thales and around ten other major French industrial groups called Confiance.AI, designed to establish a framework for deploying AI in increasingly secure ways within critical environments.

We are not necessarily talking about generative AI here, but more predictable AI. And today, these are the areas where we see the most obvious business cases, where added value is delivered once predictability and security reach the levels David described.

[Frédéric Simottel]

David, given these constraints, how do you manage to industrialise AI applications in such a competitive market?

[On-screen caption: How to industrialise AI applications]

[David Sadek]

This is indeed a crucial point when moving from proof of concept to fully industrialised and deployed systems.

At Thales, we launched an accelerator last year called Cortex, designed to mature AI-based solutions as quickly as possible and integrate them into products that are already deployed or will be deployed.

In addition, we must establish the full industrialisation chain: tools, methods and processes. The Confiance.AI programme, supported by Thales and other partners, has produced nearly 150 tools and methods covering everything from design to deployment and operational maintenance, including qualification, verification and certification.

[Frédéric Simottel]

And that industrialisation approach is essential.

[Fabrice Asvazadourian]

It is essential, because in these contexts there is absolutely no margin for error. This is not about optimising a minor administrative process or deploying a virtual assistant that might give a wrong answer.

This is critical. AI is designed as part of the product itself, embedded into the engineering processes of major industrial groups.

[On-screen caption: Generative AI – next steps]

[Frédéric Simottel]

David, we hear a lot about generative AI, but Thales works on many forms of AI: embedded AI, autonomous AI, human–machine AI, distributed AI…

[David Sadek]

At Thales, we have identified eight key differentiating AI technologies. Generative AI is one of them, but in its current state, it does not meet the criteria of trusted AI. That does not mean we will not use it, not yet.

We have launched a dedicated programme called Trusted Generative AI to make generative AI valid, secure, explainable and responsible for critical systems.

We also work on hybrid AI, combining data with expert knowledge, frugal AI to minimise data and energy usage, and AI to reduce energy consumption and carbon footprint, for example by optimising aircraft trajectories or descent paths.

[Frédéric Simottel]

Human–machine interaction is also key.

[David Sadek]

Absolutely. When AI supports decision-making, human–machine interaction must be intuitive. We work on embedded AI across sensors, radar, sonar, cameras, satellites, and on AI-driven decision-support systems, notably through Cortex Factory.

[Frédéric Simottel]

Safety and reliability, that is the key takeaway?

[On-screen caption: AI – best practices]

[Fabrice Asvazadourian]

Yes, and humility. Industrial players are characterised by a pragmatic approach, beyond hype. Buzz helps justify investment, but execution must remain grounded.

[Frédéric Simottel]

Thank you both. Fabrice Asvazadourian, Managing Director of Sopra Steria Next, and David Sadek, Vice President Research, Technology and Innovation at Thales.

See you very soon on BFM Business.

[AI Uncovered – Episode 12] – GenAI Transformations

See the detailed description

[Voice-over] BFM Business special feature “AI Uncovered, driving the AI revolution” with Frédéric Simottel.

[On-screen caption: Implementing trusted AI – how?]

[Frédéric Simottel]

Welcome to this special feature produced in partnership with Sopra Steria Next. We are going to talk about the implementation of generative AI and trusted AI within companies.

Several episodes are already available to watch on replay, featuring senior executives who have joined us. Today, we are focusing on AI within the Orange Group, and more specifically on identifying key projects that are scaling up or already in the process of doing so.

How do you engage employees? There are many themes to discuss. With us is Alexis Trichet. Hello Alexis.

[Alexis Trichet]

Hello everyone.

[Frédéric Simottel]

Director of Data and AI at Orange France, and with us as well, of course, Fabrice Asvazadourian. Hello.

[Fabrice Asvazadourian]

Hello Frédéric.

[Frédéric Simottel]

Fabrice, thank you for being with us, Managing Director of the consulting firm Sopra Steria Next.

Alexis, you have been working on AI at Orange for several years now, four or five years, and things have clearly accelerated given the structure of the Orange Group.

[On-screen caption: 10 AI domains identified at Orange]

One might have feared a somewhat fragmented approach. There are engineers everywhere, everyone wants to move forward. So what are the main areas you are focusing on today?

[On-screen caption: Alexis Trichet – Director of Data and AI, Orange France]

[Alexis Trichet]

That could have been the case, but in fact, in 2020 we realised we needed to structure things.

We could see that everything was starting to take shape. At that time, we carried out a fairly simple assessment. We already had several hundred people at Orange working on these topics, close to the business teams, which was an asset.

However, we struggled to have an overall view of what they were working on, and we also had some difficulty, we have to admit, in using standardised methods to assess the value being generated. So we decided we needed to focus on two or three major priorities.

**[Infographic: ~300 AI use cases at Orange France

55% under review
45% deployed or being deployed
Source: Orange France, September 2025]**

We realised we could not leave the situation as it was, and that we needed to organise things. This was before the arrival of GPT, at the end of 2020.

We decided to identify the key domains where we were willing to make significant bets, building on existing strengths, with business teams working closely with data scientists who had already begun exploring these areas.

[Frédéric Simottel]

So what are these areas? Network, customer relations, marketing?

[Alexis Trichet]

Exactly. The major domains are fairly classic, but they make sense for a telecoms operator: customer relations, marketing, communication and the network.

There is also a very important domain for an operator called field operations. These are the employees and partners we send either to customers’ homes or into the network to deploy or maintain services.

These are the small orange vans you see everywhere, carrying field technicians. This is therefore a major area for us.

[Frédéric Simottel]

So you focused on a small number of major themes. Fabrice, this is often the challenge highlighted throughout AI Uncovered: identifying which use cases will truly create value. How should companies approach this?

[On-screen caption: AI – identifying high-value use cases]

[Fabrice Asvazadourian]

I like what Alexis said: they made bets. We see many companies spending countless hours building Excel business plans. When it comes to AI or generative AI, I am not sure that is the best use of time.

You need convictions, but also the ability to regularly challenge them. And before trying to invent entirely new use cases, we should draw inspiration from what already exists.

Orange’s network operations, marketing and customer relations are well-known domains, and many companies communicate extensively about their AI initiatives. Identifying promising use cases that have already been tested elsewhere builds confidence.

At the start of the year, we had already deployed more than 500 use cases, which we can share with our clients. That does not mean they will work everywhere, but it shows it is possible.

Finally, companies need to be modest and ask whether the conditions for large-scale deployment are already in place. Too often, organisations wait too long before asking whether a successful proof of concept is ready to scale, and that is where things get stuck.

[Frédéric Simottel]

In other words, we need to see whether service quality has improved and assess the ratio between revenue generated and savings achieved. These indicators matter, Alexis.

[Alexis Trichet]

Absolutely. Let me give you a few examples.

The first is a marketing use case, still ongoing today: personalisation. When you log in, if you are an Orange TV customer, what you see is not the same as what I see. The same applies on the website, the app, customer service calls or in-store interactions. Employees have information tailored to each customer’s situation.

This is a classic use case, as Fabrice said, but in 2020 we decided to bet on it. We already had rule-based technology, but we felt it was time to move to state-of-the-art AI.

We built a system that clearly delivers between 5% and 10% additional value, depending on the case. These projects also involve the CFO, whose role is sometimes underestimated in AI initiatives.

**[Infographic: Orange objectives

Revenue → Savings (OPEX, CAPEX) → NPS → Revenue]**

Finance and management control play a crucial role. From the outset, senior management legitimately asked: what return are we getting?

So we invested a lot of effort in measuring value consistently, using sophisticated methods and control groups to ensure the value generated by these systems was real.

**[Infographic: Contribution of AI projects to Orange Group financial results

~€200m generated in 2024

€300m expected in 2025
Source: Orange Group financial results 2024]**

[Frédéric Simottel]

Another key topic we often discuss is data management. At Orange, there is a major overhaul of your data assets.

[On-screen caption: Reinventing the data estate for AI]

[Alexis Trichet]

Yes, data management is essential. We handle tens of petabytes of data, tens of thousands of gigabytes. It is a massive data estate.

You cannot tackle everything at once. Our rule is simple: start where the priority use cases are. That is what we are doing now.

Structuring, documenting and sometimes discarding data takes time. What we underestimated was the effort required to explain where data is located and to change habits. Change management costs more than expected.

[Frédéric Simottel]

This brings us to data governance, which may sound abstract but is critical. Fabrice?

[Fabrice Asvazadourian]

It is extremely important. Many companies have already made progress with golden sources, data owners and governance frameworks.

Generative AI adds complexity with unstructured data and rapidly evolving technologies. Today, however, the real challenge is investment in data platforms. These investments may be invisible, but they are indispensable.

[Frédéric Simottel]

Has the data governance stage been completed?

[Fabrice Asvazadourian]

Completed? No. But companies like Orange are well equipped and moving fast. The frameworks, ethics and training are in place. Now they need to be fully implemented.

[Frédéric Simottel]

What are the key best practices for scaling AI?

[On-screen caption: Industrialising AI – best practices]

[Alexis Trichet]

The good news is that we are scaling. Across the group, AI use cases generated €200 million in value in 2024, and more than €300 million is expected in 2025.

There are two main dimensions. First, project and change management must be flawless: clear problem definition, strong sponsorship, robust governance, KPIs, and involving end users.

Second, you need the right technical stack to scale under the right conditions. When both are in place, scaling is possible, and that is what we are doing every day.

[Frédéric Simottel]

We do not often talk about social dialogue, but it is crucial when transforming jobs.

[On-screen caption: AI and transformation – social dialogue is key]

[Fabrice Asvazadourian]

Workforce planning needs to integrate AI. The major employment transformations will come around 2030 and beyond, but preparation must start now.

[Frédéric Simottel]

Thank you both. Alexis Trichet, Director of Data and AI at Orange France, and Fabrice Asvazadourian, Managing Director of Sopra Steria Next.

See you very soon for another BFM Business special feature: AI Uncovered, driving the AI revolution.

[AI Uncovered – Episode 13] – GenAI & Customer Experience

See the detailed description

[BFM Business Logo, text appears: “Special Edition, BFM Business files” then “AI Uncovered: The leaders driving the AI revolution“]

[Voiceover]
Special edition, BFM Business files. AI Uncovered: The leaders driving the AI revolution. With Frédéric Simotel.

[Frédéric Simotel]
Welcome to our show AI Uncovered, where we explore how business leaders driving the AI revolution, in partnership with Sopra Steria Next.
We’ll be talking about the implementation of generative AI and trustworthy AI within companies. And specifically, we’ll take a closer look at AI within FDJ United. You probably know the name—it’s La Française des Jeux, but today it operates internationally, present in ten countries, with a strong focus on redesigning the customer experience.
We’ll discuss this with our two guests: Sébastien Rosanès, hello.

[Sébastien Rosanès]
Hello.

[Frédéric Simotel]
Sébastien, thanks for joining us. You’re the Digital, Data and AI Director at FDJ United. And Fabrice Asvazadourian, hello Fabrice.

[Fabrice Asvazadourian]
Hello Frédéric.

[Info bar: FDJ United – AI for Safer Gaming]

[Frédéric Simotel]
Managing Director of consulting firm Sopra Steria Next. So, Sébastien, let’s start with you. FDJ United—every year, you’ve got 33 million players. I guess we’re all part of that, some more often than others, depending on the jackpots.
You’re already using AI in your organisation—for fraud detection, customer experience, and what you call responsible gaming, right?

[Sébastien Rosanès]
Exactly. AI is a key element of our long-term growth strategy.
We use it for marketing, targeting, better customer knowledge, and optimising game creation.
We’re even exploring how AI can help us design new games, imagine new ones.
And of course, we also use it to protect players. Gaming has addictive risks, and AI helps us be much more precise in detecting and deciding when to intervene or stop play, or support players.

[Frédéric Simotel]
And not just checking the birth date to see if someone’s over 18.

[Sébastien Rosanès]
Exactly. That’s just the basic requirement. But beyond that, we look at how many games are played, how much money is spent, and so on.
AI helps detect patterns and even suggest more personalised ways to intervene instead of sending a one-size-fits-all alert.
And behind that, humans step in to take over and have real conversations with players.

[Infographic: FDJ United]
• 5,000+ employees
• Presence in 10+ countries
• 33 million players
• €3B in revenue (2024)
• Digital = 30% of activity

We regularly make thousands of calls to some of our online players, with whom we’ve built long-term relationships.
It’s important for us to rely on insights flagged by AI.

[Frédéric Simotel]
And that’s the paradox—you want people to play, but not too much. Because of the addictive side. You’re aware of it, you set the right alerts, and AI helps you do that.
This customer relationship, this personalisation, is one of the main successful uses of AI across sectors.

[Info bar: Customer personalisation, key to AI success]

[Fabrice Asvazadourian]
We see it a lot, especially in fraud, risk, and customer support with a responsible approach.
I believe companies that push AI too far for short-term revenue gains may pay a high price for irresponsible AI later.
That’s why FDJ United’s initiative is remarkable. It’s a big issue.
We see companies hesitating between ultra-personalisation—which AI allows—but also worrying about data quality. Because personalisation only works if based on accurate data. Otherwise, you risk fake personalisation that customers quickly dislike.

On the other hand, proper personalisation helps retain and support loyal customers over time.

[Frédéric Simotel]
Sébastien, can you explain how this AI integration works at FDJ United—both in terms of your Data & AI team organisation and your infrastructure, since you’ve chosen a cloud-first approach?

[Info bar: Strong impact on data & AI teams]

[Sébastien Rosanès]
Sure. Both go hand in hand. First, to scale AI, teams must be more autonomous, thinking strategically about their AI products, developing them, and operating them.
So we’ve grouped all these skills together under one unit, under my direction, to give teams autonomy.
We call them “squads.” It’s a model inspired by tech companies like Amazon or Google, now spreading into larger organisations like FDJ United.
But beyond organisation, we also need flexible tools that support autonomous work.
That’s where the cloud is powerful—it gives teams the ability to build, test, and deploy directly, right up to customer-facing environments.

[Frédéric Simotel]
So what does being cloud-first bring you?

[Sébastien Rosanès]
It lets teams control their applications directly and push them to the cloud.
They also benefit from the latest AI innovations. And it avoids the usual inertia of big IT departments managing heavy projects that can’t pivot quickly.
For instance, when GPT-5 came out this summer, or with Microsoft Copilot—the generative AI tool in Word, Excel, PowerPoint, and Teams—it became available within days to all FDJ United employees through the cloud, with zero IT intervention.
That’s a revolution. Because in AI, every 2–3 months, a major innovation reshuffles everything. Having access to the best tools as soon as they’re released is transformative.

[Frédéric Simotel]
That’s what fascinates me about IT today. How do you keep up with that pace of innovation, when before cycles were so much longer?
And Fabrice, with FDJ United, we clearly see a balance between the Data-AI team structure and their infrastructure. Many companies struggle with that.

[Fabrice Asvazadourian]
Yes. Many clients have well-thought-out organisational setups.
But fundamentally, we’re technologists. Ideas only matter if they go beyond POCs into real-world deployment.
That’s why investment in data platforms and cloud is critical—to move from lab thinking to industrial thinking.
Many clients feel frustrated after the initial hype phase and now face the challenge of scaling.
When you have 33 million customers, like FDJ United, a pilot that reaches 50 people is useless. It has to scale to everyone.
That’s where infrastructure investments come in—beyond organisation and governance.

[Info bar: Industrialisation and AI culture, the keys]

[Frédéric Simotel]
Exactly. Scaling isn’t just inflating a small project—it’s a complex industrialisation process.
And successful AI projects combine that scale with employee culture and adoption.
How did you ensure this transformation was well received? Because it changes jobs, and people need to be brought along.

[Sébastien Rosanès]
Absolutely. My view is that AI actually means more human interaction.
AI takes over repetitive tasks, freeing people to focus on more human, value-added ones—like spending more time with clients or colleagues instead of doing back-office work.
So we train and upskill employees. We built a Data and AI Academy with two goals.
First, to train all employees to use AI daily—using tools like Copilot, for example. It sounds simple, but prompting properly is a real skill.

[Frédéric Simotel]
And they need to make it a habit.

[Sébastien Rosanès]
Exactly. It becomes second nature. Our goal is to make 100% of our employees AI-literate, with basic skills and data-savvy analysts in every department.
The data analyst role is spreading fast—it’s no longer a separate job. It’s becoming a core skill. Marketing? You’re also a data analyst. HR? You analyse employee data.
That’s reflected in our hiring too. For example, we launched a graduate program this year to train new hires in all AI product roles—data engineering, data science, etc.—so they gain a 360° view.
And for existing Data & AI teams, we help them evolve too. Ten years ago, we hired statisticians. Today, those roles are becoming data scientists. We’re helping them upskill into the AI era.

[Info bar: AI – Getting teams on board]

[Frédéric Simotel]
And how do you make sure it’s well received?

[Sébastien Rosanès]
It starts with understanding: your job isn’t being replaced, it’s being augmented.
If you embrace AI, your work becomes more interesting and higher value.
So, we conduct a fairly detailed analysis of each of the professions to imagine the jobs of the future, how they will change and tell employees that we are not here to replace them. We’re replacing tasks, not people.
We analyse each role to map its future evolution and show employees we’re replacing tasks, not people.
I’m convinced AI won’t replace employees—but employees with AI skills will replace those without them.

[Frédéric Simotel]
Fabrice, that’s where mapping tasks and anticipating change is key, right?

[Fabrice Asvazadourian]
Exactly. Every leader must commit to bringing people along—it’s their duty.
We’ve seen many transformations before; this one is no different.
We now have enough data by job role to model how fast and how much AI will help, so we can plan ahead and manage the change.
The key is anticipating, not reacting.

[Frédéric Simotel]
That’s the perfect closing line. Thank you both, Sébastien Rosanès from FDJ United and Fabrice Asvazadourian from Sopra Steria Next. See you soon for another episode of AI Uncovered – Leaders driving the AI revolution.