Measuring AI’s impact within organizations: An interview with Sebastien Rozanes, global chief data officer at Carrefour
AI, Data & Analytics

Measuring AI’s impact within organizations: An interview with Sebastien Rozanes, global chief data officer at Carrefour

Sebastien Rozanes, global chief data officer of Carrefour, discusses the role of chief data officer, implementing AI and measuring its impacts, and how he is making sure Carrefour has the leaders it needs to navigate the use of emerging technologies.
Listen to the Heidrick & Struggles Leadership Podcast on Apple Podcasts Listen to the Heidrick & Struggles Leadership Podcast on Spotify

In this next episode of The Heidrick & Struggles Leadership Podcast, Heidrick & Struggles’ Frédéric Groussolles speaks to Sebastien Rozanes, group chief data officer of Carrefour, a French multinational retail and wholesaling corporation. Rozanes discusses the role of chief data officer and how he is managing the growing role of AI at his organization: how and where he is implementing it, how he is measuring its impact, and who he is involving in the decision-making process around tools. He also discusses AI’s impact on the talent landscape and recruiting process, what sort of leadership skills or capabilities he sees as most difficult to find or develop, and how he is making sure Carrefour has the leaders it needs to navigate the use of AI and other emerging technologies. 

Below is a full transcript of the episode, which has been edited for clarity.

Welcome to The Heidrick & Struggles Leadership Podcast. Heidrick is the premier global provider of senior-level executive search and leadership consulting services. Diversity and inclusion, leading through tumultuous times, and building thriving teams and organizations are among the core issues we talk with leaders about every day, including in our podcasts. Thank you for joining the conversation. 

Frédéric Groussolles: Hi, I'm Frédéric Groussolles, partner in Heidrick & Struggles’ Paris office. I'm a member of the Global Technology & Services Sector, and I co-lead our Data & AI practice in Europe. In today's podcast, I'm excited to speak to Sebastien Rozanes, global chief data officer of Carrefour. Following a 15-year tenure in strategy consulting, McKinsey, BCG, mostly in North America and later in France, Sebastien, you've transitioned to Carrefour recently, in April 2020. Carrefour being one of the largest retailers, I think you have more than 300,000 employees across more than 30 countries. Here, you lead and implement the group's data analytics strategy. In this capacity, you oversee the Analytics Factory, a unit comprised of proficient data specialists, and also responsible for the adoption of AI and data practices, tools, and solutions in the businesses across the whole Carrefour group. Sebastien, welcome, and thank you for taking the time to speak with us today.

Sebastien Rozanes: Thank you very much for inviting me to the show, Frédéric. 

Frédéric Grousolles: When you joined, how was your data and AI organized within Carrefour? What's your current scope, your reporting lines, and overall who within Carrefour leads the AI strategy?

Sebastien Rozanes: Well, a bit more than two years ago, we created what we call the analytics factories in our eight core countries, and the idea at the time was to gain scale, to transition out of our previous model where that analytics was incubated in a way into what we called the Carrefour Data Lab, and we wanted to move to a true factory model for analytics. So, beyond the semantics of a lab versus a factory, we made it clear that we wanted to shift out of the laboratory mode. It was great to innovate and develop proof of concept for years, but we really wanted to scale that better in the business. 

Thus at this time, we moved toward this factory model. We started in France, where our headquarters are located, but we rapidly replicated the model across our core countries. We increased the amount of investment and started an intensive internalization of the resources. Again, the idea was that if we were serious about building a factory, we needed more skilled workers in the factory to truly operate and scale up the setup. So, the delivery model was data analytics at the federal model, and that's when I joined Carrefour. 

Today, we have a small team of a dozen of high-profile professionals at the group level to steer the strategy, work on innovations, set the ways of working everywhere on data, and spearhead the data transformation in the business. But clearly, the 400 data technology and data analytics professionals are located in each of our company's operations, really close to the business. So in my role at Carrefour, today what I do: I steer this team, I define the overall data and AI strategy, and I report to the digital team at Carrefour. 

Frédéric Grousolles: And so when you when you arrived, I'm sure already there was quite a lot of AI being used. And by AI, I just meant here maybe classic AI, so before generative AI. Can you share any of these key use cases that were already existing?

Sebastien Rozanes: Oh, absolutely. We've deployed AI use cases at scale for a few years. Today, we are clearly seeing that we are gathering the fruit of the investment we made in the past years. AI now has a visible impact across the entire value chain of Carrefour. Let me give you a few examples to be very concrete. In the very early stage in the definition of our assortment, we are optimizing our assortments and customizing this assortment by store thanks to machine learning. And talking numbers, it helped us really reduce by 11%, since January, the assortment, while maintaining the sales, the margin, and of course the client satisfaction. 

Later in the process, for example, in inbound logistics, we are significantly reducing our out-of-stocks and increasing product availability again, with machine learning techniques and operational research techniques to enable much better forecast, as well as the logistics chain. Also, of course, in our stores, in a way, AI has a direct visible impact. For example, as you might know we produce bakery in stores, and since 2022 we are using an algorithm based on gradient boosting, with 120 input variables to predict a much more accurate quantity of pastries to produce. With that, last year we saved 265 tons of products that would have been thrown away without the algo. The good news is that that's good for Carrefour’s budget of course, but it's even better for the planet. We avoided a lot of waste. 

And finally another example, even further down the chain, in sales, we are using supervised learning techniques to predict the performance of promotions. As we launch new campaigns, our sales and marketing team can predict whether the promotion will be making money and will be a high-performing promotion or not. And with that, we've reduced by 90% in one year the number of promotions that were either totally inefficient or making Carrefour lose thousands, or also in some instances hundreds of thousands of Euros per year. These are a few examples throughout the value chain but what you should take away, in my view, is that today in retail, AI is simply everywhere, from beginning to end across the value chain. 

Beyond these use cases in the business, two years ago we also decided that data for business was great, but we needed to go one step further. We created a business out of our data that we call Carrefour Links, and with this business, we sell directly our data to our suppliers, which are the consumer goods companies like, for example, Mondelez or Procter & Gamble. And they pay us to have this very granular data about the purchasing behaviors of the customers, our supply chain data, and most importantly, in my view, we share this data with them and we go on to the next level to collaborate with them. So we are jointly developing what we call AI products on top of the data to enable much better collaboration, jointly design promotions, and make sure that we have a well-adapted supply chain to minimize the inventory throughout the chain, from the producer down to the distribution in the stores. This is where I think data analytics is going to the next level, and it's really a game-changer in the way we work with our suppliers.

Frédéric Grousolles: Fascinating. So you've had already so many AI use cases, but to make it even more interesting, generative AI came into the picture this year like a storm. So, how did you manage even more and prioritize even more AI requests, AI use cases?

Sebastien Rozanes: Yes, that's a very good point. Frankly, generative AI has been discussed in labs, in expert conferences, in research paper in a way for years, but for a long time it had never reach the business world. For all of us, the storm hit us like last year in November 2022, when OpenAI released its public version of ChatGPT. This was a massive breakthrough frankly, and we had to adapt fast and in full transparency, I had not put a single Euro on generative AI in my 2023 budget, and this really caught everyone by surprise. So as always, in the retail business we reacted fast, we took action. And if you have a choice between deep thinking and strategy on the one hand or action, you always go for action first, you push for innovation to market, you gather customer insights, you gather reactions, and only maybe later you think about the deeper strategy and it’s always in that order. 

So to summarize in a couple of minutes what we did to get started in this totally unknown space nine months ago, we essentially did three things. Number one, we opened our minds, and we decided that we could do a lot of things with generative AI, but we didn't want to listen and interrogate the data scientist or the data specialist. So very practically, we booked one hour with 20 execs in the company and made a five-minute demo of ChatGPT. And then we listened to them imagining what major pain points they could address with this, that kind of technology, which activities they could totally automate with generative AI. So we started from the business view, definitely not from the tech or data view. Then came with all these ideas, the time to prioritize, and we rapidly triaged the use cases in a way to have on the one hand the use cases that were worth building and the ones that we would put on the back burner. And to do that, we heavily took into account the business impact potential, but also the availability of data because of course, not all the data that people were dreaming of were available. Also, we triaged in a way the use cases that we should do ourselves and the ones that we could wait for because it would come to us anyway. So for example, all the use cases around gen AI that were connected to automating email creation, slide creation, we knew that companies like Microsoft and Google were going to deliver that as part of their production shoot. So we got rid of this, and then we decided to be fast and first. That was a short step: let's be fast.

Frédéric Grousolles: Very impressive, so I can see and I can feel that you've been much more agile than you were a couple of months ago. So looking at the organization you described when you joined, has there been any impact on your current data/AI organization? 

Sebastien Rozanes: With the rise of data and AI in the business, frankly you cannot have data analytics, data platforms, data technologists, and business in separate silos anymore. 

Hence at Carrefour, we are progressively transitioning to a much more integrated model, a much more full stack delivery model in a way, and for us, it has made three impacts. Number one is that data technology and data analytics are more and more intertwined. Your data scientists do more and more data engineering work. The data analysts are starting to leverage low-code, no-code tools to develop machine-learning models, so that engineers are now much more embedded into the product team, they're actually delivering and not hidden behind the data platform. So it's a clear call for action for me to a more integrated model. That's why in this context, this year we decided to extract our data technology and data platform teams out of IT and integrate all the talents under the data analytics team, all together. 

The second aspect is we are deploying at the enterprise level a much more product-centric delivery model for all of our data and text projects. In a nutshell what it is, is that we apply the basics of agility and we put data, tech, business, and design into a single product team, and we totally eliminate the frontiers between the teams. And third aspect is that the business on this side is also much more demanding and wants more and more data, more and more analytics. So our strategy to cope with this demand is not to multiply my team by three within the next year, which is actually what I would need to cope with the demand. Why? Because that would be expensive, frankly difficult to do, and risky. So we've decided to invest heavily in making data available to them, equipping them with modern tools, and training them accordingly. And the idea is that we want to make these business people autonomous with self-service data, and self-service BI. And the concept behind that, which we are pushing hard, is to make all our category managers, management controllers, and top managers, capable of building an analysis on their own with certified high-quality data. So if you push that like in a way with a metaphor, an analogy, we are going to move to a self-service model. 

And in the past, like up to today, it was very much like a restaurant model where, you know, every single day you have a question around data, around analytics, you want a dashboard to be built. And you go to the restaurant, you go to your management controller or you go to your analytics team and you had them do the extraction, the data analysis. You place an order, you wait for the meal to come to the table—zero effort from the business. And what we found out is that interpretations of the data team is often inaccurate, it leads to a lot of redo and mistakes because we fail to understand each other. 

In the new mode we are setting up, we are now building what we call a data supermarket. It’s very simple, it’s like a supermarket in real life. It's a single portal where all the data are located, you can find them very easily, and you don't need to have any technical skills. Unlike, you know, when you go to a data warehouse, and you need to log in and have tech skills to extract the data, we don't stop there of course. Beyond, you know, placing the data in an accessible spot, we are deploying more modern big data analytics tools that enable our teams to go way beyond Excel and analyze huge volumes of data. And so that's kind of the kitchen that you attach to the supermarket and then, you know, to use this kitchen, what we do is that we train these people in the business to become chefs, so they can cook their meals. So in a way, there are these three things that go together: self-service data, the supermarket of data, the kitchen, the tools, the tooling to make good recipes, and finally the chef skills that you need to build in the business so that they can fully use the power of these available tools.

Frédéric Grousolles: As a French person, I like this metaphor of becoming more and more an autonomous chef at home. 

Sebastien Rozanes: Exactly. 

Frédéric Grousolles: But what's now the impact on the talent landscape? I mean we know that there is this war of talents out there, so what sort of impact generative AI has had on your talent and recruitment?

Sebastien Rozanes: On that front, I see gen AI frankly as a fantastic advancement for all enterprise employees. In a way, it brings AI into the hands of the people. That's the transition because, you know, historically we brought AI mostly in the hands of data scientists, and now we’re putting that into the hands of the business teams. But to deliver the gen AI-based products in an organization, you need a new mix of skills and you create new roles. 

There is good news for me behind that and for, I think, all senior executives in enterprises. Let me explain that a bit. As we rebalance the analytics investment to integrate generative AI products in our portfolio, we will need proportionally fewer data scientists in our teams. This is good news because we all know how hard and expensive these data scientists are on the market, given the demand for data scientists today vastly outpaces the supply. A data scientist has been the king of the jungle for the past few years, and I believe that this will help rebalance some of the power, and the hype cycle around the data scientist role might slow down a bit. I'm not saying we won't need those data scientists anymore, but it will help to rebalance the struggle and the power between the data scientists and the enterprises. 

If you think even longer term, we will all learn, you and I and all the business leaders, these prompting skills ourselves. We will maybe suddenly have assistants that will get created to help business people write the right prompts and hopefully this skill will penetrate the business. And the business, management controllers, business analysts, and business managers will be able to prompt on their own the large language models efficiently and more and more often. So we'll see how it plays in the future. In the very short term, there are these new skills that are really changing the mix, and I think that's really good for us and creating also new opportunities with gen AI. 

Frédéric Grousolles: Amazing, so a lot is happening already and clearly on your front. So if you look back, I mean after these few months of experience, how are you assessing the performance of all these gen AI initiatives? And perhaps who is involved in all these discussions and how do you decide what's next?

Sebastien Rozanes: Yes, very good question. Frankly, impact is the number one measure for us to justify the investment, so we track that very thoroughly. Ultimately, it's the business that defines what are the business objectives, what are the efficiencies we want to gain and what are the uplift in customer experience or in customer satisfaction that will be used to measure the impact. What we are doing, for example, with one of the initiatives that we did put on our website, which is a chatbot to enable clients to shop on our store using generative AI. We measure the uplift in the size of the basket, and the speed to get the basket complicated because in e-commerce there is a very strong correlation between the time you spend on the website to shop and the purchase. Because in grocery shopping, the basket can be huge and you're interrupted all the time, you get kids coming in and out, you get a phone call, and you stop your purchase process. So it's very easy to see, you know, with AB testing, how customers are behaving, for example in this case of the chatbot, and we are tracking very closely how the experience and how the business impact is changing with generative AI technologies embedded into our website, for example.

Frédéric Grousolles: And so we look now at the impact on your leadership within Carrefour, with all what's happening in this AI space, what sort of leadership skills or capabilities would you say are now most difficult to find or develop? And how are you making sure you have the leaders you will need going forward with AI and other emerging technologies?

Sebastien Rozanes: A very good question actually, this is something that is really top of my mind almost every day. In a way to develop data and AI leaders, it’s not only about technology, it's not only about mathematical skills. These are kind of the foundational skills that you need but as we recruit data product managers or data scientists, we look for, more and more for business acumen, communication skills, and leadership skills that are essential to engage in a partnership model with the business. Even more importantly in the product team’s model that I mentioned earlier, you cannot have this tech or data person, you know, hidden behind the curtain anymore. 

So to recruit these top talents, let me give you one example that we started last year and really got the impact this year. We created graduate data analytics programs for top talents which are within a couple of years post-graduation, and we totally revamped the recruiting process to integrate the evaluation of these softer skills, on top of the hard data and tech skills. We also created a unique value proposition where these young graduates can accelerate their career and learning by rotating across three different data jobs in two years, taking different roles across eight countries, so they have one of their rotations outside of their core country. To expose them to different environments, to have them, you know, try out maybe a job in the data environment that they wouldn't have applied to, but they, you know, might discover something really that might fit their skills. But also so that they can sit on the other side of the fence in a role that maybe you want to become a data engineer but if you work for six/eight months as a data scientist, you will really understand what's the job of data scientist and you will emerge better. That's the bet that we made, so that you build up your skills, you get into a mode where you learn fast, and you harness your leadership skills to step up in your roles every seven/eight months at Carrefour. 

Frédéric Grousolles: Amazing. I'm sure some others would like the idea and maybe replicate it. I mean, it's a good way to attract top talents. So we are getting close to the end of this podcast, and you've shared a lot already. So maybe in closing, what advice would you have for other data leaders in working with executives across the organization to implement AI? In other words, what will be the key success factors you need, I mean, to be successful with AI?

Sebastien Rozanes: Clearly for me, if I had one big piece of advice, it's to make sure that you have the C-suite onboard because this type of transformation starts from the top. And I've been very lucky to have this CEO's support to launch a very critical initiative for us at Carrefour. Last year, we trained our top 200 executives globally to data for two days. We put them in a room for two days and built the curriculum together with HEC in France, a prominent business school, and this has been a game-changer. It has for me, I have seen these execs turn into data evangelists in the organization and truly create a world where they help us, and they help me, push innovation with data to their teams. 

On top of that, to cement some of the learnings and expose them to real life, because it's great to be in the classroom for two days, but when the rubber hits the road, it's even more important. We organize learning expeditions in the US and Israel for our execs to meet the start-up ecosystem, see the technology in action, and mingle with other retail executives in other organizations. We discuss with VCs and these people in the space for product ideas that could really shake the world of retail. And all of this investment a year later now is paying off a lot, and it has made my job so much easier as a chief data officer. 

Frédéric Grousolles: And so to summarize, how would you define your chief data officer role today? 

Sebastien Rozanes: If I summarize my role, it’s very simple in a way. I have two roles. My number one is to make myself useful by helping the teams prioritize the most advantageous cases that could generate value, deploy the data and tech skills at the right place, deliver these use cases to the teams, and, in the end, measure the value. That's kind of my useful role. The second role is to make myself useless by, as I mentioned earlier, making the data accessible in self-service, equipping the business teams with modern tools beyond the traditional and very limited kind of Excel, and of course training them to become chefs. In a way, the successful transformation is that even the role of chief data officer, the role might disappear in the future because everybody will be data proficient. The data will not be a separate team, it will be something that will be embedded into the business.

Frédéric Grousolles: I think your role is still very, very, very much useful at this stage. But thank you, it’s absolutely a fascinating moment, and I know Carrefour will be quickly accelerating its data-driven journey. We'll be following you. Sebastien, once again thank you very much for taking the time and sharing all this wonderful insight with us today.

Sebastien Rozanes: Thank you very much.

Thanks for listening to The Heidrick & Struggles Leadership Podcast. To make sure you don’t miss more future-shaping ideas and conversations, please subscribe to our channel on the podcast app. And if you’re listening via LinkedIn, Twitter, or YouTube, why not share this with your connections? Until next time. 

About the interviewer

Frédéric Groussolles ( is a partner in Heidrick & Struggles’ Paris office and a member of the global Technology & Services Practice and co-leads the AI, Data & Analytics sector in Europe & Africa.

Stay connected

Stay connected to our expert insights, thought leadership, and event information.

Leadership Podcast

Explore the latest episodes of The Heidrick & Struggles Leadership Podcast