Retailer Carrefour is transforming procurement with a process approach to generative AI

“We realized comparing three quotes would take around 30 minutes for a buyer to do it manually,” said Florian Tué, Head of Procurement Transformation at Carrefour. “If we do it with ChatGPT and the POC that we've been running, it takes only 10 minutes. So the 20 minutes productivity gain is huge for our organization, and that's why we'll be pushing it so much.”

Tué was part of a generative AI panel at Celosphere 2023 and also sat down with Faika Halici, Sr. Director of Customer Advocacy for Celonis, and Tom Smith, Editor-in-Chief of Acceleration Economy, to discuss the company’s current artificial intelligence initiatives and future plans.

https://delivery-p141552-e1488202.adobeaemcloud.com/adobe/assets/urn:aaid:aem:e3e015b6-273a-490d-a6f9-b958a7e5d30c/as/ng3KWEKrmd6rbehP312jD8.jpg
https://videos.celonis.com/watch/ng3KWEKrmd6rbehP312jD8

The following is a transcript of the interview, edited for readability.

Faika Halici: Carrefour's Florian Tue, Head of Procurement Transformation joins us at Celosphere, and in conversation with Acceleration Economy’s, Tom Smith, and I, to discuss how this internationally active French retail and wholesale company has been applying process mining to the business. Florian, I understand your work with Celonis began several years ago. Let's kick off with an introduction to Carrefour, and a bit more perspective into your personal areas of responsibility.

Florian Tue: Sure. So Carrefour is one of the major European retailers. So we are operating in 40 countries, and we have around 14,000 stores. And our turnover is around 80 billion euros. And my role at Carrefour is to enhance the digital transformation of the procurement function for both the indirect and private label.

Faika Halici: And so perhaps you could talk to us a bit about AI and generative AI, and how does that factor into Carrefour's strategy today and going forward?

Florian Tue: Yeah, sure. So, actually, we identified generative AI a few months ago, so before summer 2023. And we realized that it could have a good potential for Carrefour and for all of our functions, not only procurement, but also marketing, HR, and plenty of other functions. And so we decided to take it from a process perspective, and to really look at our procurement journey, and identify the different use cases that can bring productivity and value to Carrefour's procurement. And so we've identified that for the quotes comparison step, we could generate a lot of value from it.

So, for example, analyzing a quote at Carrefour from an indirect buyer, will take around 30 minutes if he does it manually. But with the PoC that we are currently running with ChatGPT, it will only take them 10 minutes. Which is a very high productivity gain, when we know that every buyer, from an hourly perspective, makes 1,000 euros of EBITDA in terms of a saving generation, yeah.

Faika Halici: Wow, exciting.

Tom Smith: Those are really impressive results. Can you talk to us about any lessons you've learned in the use of generative AI? Obviously, we're in pretty early days here, but what have you learned so far that other customers might be able to use in their business?

Florian Tue: Yeah, of course. First of all, it's really impressive, in terms of the results, the accuracy of the results, the reactivity, also, the productivity.

Tom Smith: Accuracy is good.

Florian Tue: Yeah, yeah, it looks like [it has] tremendous potential. But the thing is you need to be really careful and up to date with the new release. So, today, we're at Celosphere, and we've seen that Celonis will bring up very new functionalities that are very impressive as well. You also have [other] companies [that] are currently releasing new functionalities. So I think people should be really focused and stay up to date. Because every month there is a new functionality that is released.

And I think it's really tough for big corporations to follow, in fact. Because the time that a new functionality is released, it's pretty hard for them to try to implement it inside their organization. So I think corporate organizations need to find the right balance between the different use cases that they can apply to each and every function that they have. And in fact, yeah, it's pretty tough to follow the rhythm of the releases of generative AI, to be honest.

Tom Smith: Speaking of new generative AI technology, Celonis released Process Copilots at this event. Can you give us some feedback on that and how that might play into the work that you're doing with Celonis today?

Florian Tue: Yeah, to me, it sounds really impressive. Why? Because I'm just a business contributor on the Celonis project that we've been running at Carrefour. So it was led by our finance team. And, personally, as a business contributor, I found really impressive what the work that we've been doing with Celonis, in terms of procure-to-pay, account payable, and super creation process mapping because it can help you have detailed data of each step of the process.

But the thing is, it's even too much data. And so, as a business stakeholder, what's of importance for me is to really identify the bottlenecks, and to have a detailed action plan to make sure we are correct, what's wrong, and what we can improve. And so I am sure with that digital twin that Celonis has been developing through generative AI, we'll be able to understand really the bottlenecks, to generate graphs without a lot of efforts, and to generate action plans, and to concretely improve our processes, I would say, yeah.

Tom Smith: And another big announcement here was the Process Intelligence Graph, which is, as I understand it, something of a compilation of all the process knowledge that Celonis has collected over the years. But for you in the retail industry, do you expect to really get a lot of value out of that? How would it help you with some of your retail initiatives?

Florian Tue: Of course, because it can help us ask any questions to Celonis, and generate these graphs, as you said. So maybe identify double payments, identify a lack of reactivity of several stakeholders on some part of the process. To answer plenty of questions that we have at the moment, but that we didn't take the time to verify through accurate data. And so, for me, it's, of course, a great way to generate value from Celonis here.