1,300 Restaurants, Six Brands, One AI Roadmap

0:00Sophie Buresi: Five years ago, my challenge was to convince people that we could do amazing things with AI. Now, my challenge is not to say no to everyone. We need to work on everything but AI before we can create value with AI.

0:15Conor Sheridan: I'm Conor Sheridan, founder and CEO of Nori, and this is What's Cooking. My guest today is Sophie Buresi, group head of AI at QSOP, which runs around 1 ,300 restaurants across six brands, including Burger King, Quick, Otaxers, Dunkin' and Chopsticks. In our conversation, we get into why AI use cases like drive-thru ordering make sense in some markets but go on hold in others, how Soki sells a use case to a franchisee who owns their own P &L, and why she spent her first months, as in her own words, the annoying person asking why data, architecture and process are in place before building AI agents and workflows.

0:56Conor Sheridan: Here's my conversation with Sophie.

1:02Conor Sheridan: Sophie, great to have you on What's Cooking. Thanks for joining us.

1:06Sophie Buresi: Of course, thank you for having me.

1:07Conor Sheridan: So super excited to speak to you. Obviously, we're AI practitioners at Nori, but you're the first ever head of AI that's been on the show. So really excited to dig in and understand what that means for a group of QSORP size and what that means for operations in the business. And I'm sure lots of folks listening will be very keen to understand what you're working through and see how they can apply it to their business. So excited to dig in. Before we do, maybe you could give us a quick rundown of how you came to be the group head of AI for this business and what's your pathway to this role today?

1:46Sophie Buresi: Of course. So I've actually been working in, I would say, data, AI, technology transformation for the past 10 years now, a bit more than 10 years. Half of my journey was made in consulting. So I worked in a consulting company called Artifact, specializing data and AI transformation. and while I was working at Artifact, I was already very much involved in data and AI transformation for retailers or in retail, different kind of retailers. And so when I left consulting and joined the clients, I kept my focus on retailers. So I had experience in luxury retail, in fashion retail, and very recently I had that opportunity.

2:36Sophie Buresi: after LVMH and ETAM Group to join QSRP. So that's food retail. Obviously, the business is very different from LVMH or from a fashion brand, but it still is retail. So I'm sure that we'll dive a bit afterwards. But if you consider data or AI transformation, most of the topic is transversal. The use case are quite similar. So there was much of my previous experiences I could bring to QSRP.

3:10Conor Sheridan: Nice. And when you came into the business, what was the mandate for the role? What were you tasked with? It was a brand new position in the business. It never existed before, which is exciting, but also daunting, I can imagine. What's the mandate for this department?

3:26Sophie Buresi: So the mandate at first was quite easy. We need to create value with AI. Please, we need someone to do that. No, but with a bit more details, actually, I mean, QSRP has, I would say, two objectives. The first one, everyone has the same, is we want to improve a bidda. And AI, you have other tools to do that, but AI is a very powerful lever to improve a bidda, either top line or bottom line. But obviously, you need to structure your strategy to reach that objective. so they needed someone to really target a bit of improvement with AI. And the second objective, and so that's part of the mandate, was also to say, okay, most of our strategy at USRP is based on growth.

4:17Sophie Buresi: So making sure that we grow our brands, but also that we grow as a company, potentially acquiring new companies and new brands. And so we want to make sure that we can continue growing without multiplying our teams, especially the HQ teams, the back office teams. So that's the second mandate is how we can leverage AI and everything that we put under the AI umbrella to really keep our organization as lean as possible. It's really the mandate. And obviously with it goes the use cases, the foundations, building the teams, organizing everything. But I'm sure we'll get into the details.

5:00Conor Sheridan: Yeah, super cool. For a business of that size, 1 ,300 restaurants across multiple geos, different concepts, so ranging all within the QSOR space, but obviously different formats and whatnot. There's inherent complexities in different brands and concepts. If you think about AI for frontline EBITDA in those concepts, it might be slightly different. and then obviously the HQ. It's really exciting. But when you went into the role for the first, say, 90 days or first four months, anything surprise you when you looked across the different brands and concepts or anything scare you when you looked at it?

5:47Sophie Buresi: So I think that I found something that's quite rare, especially in companies that are not super mature. And it was the case of QSRP, especially versus the companies I used to work for that were a bit more advanced in data and AI transformation. But they had a lot of good ideas. I would say real will and also real commitment from the teams, but also from the leadership. So that's super important because you don't have to convince people. They're already convinced. So it's more about execution and how we can move fast. but the one thing that wasn't there and actually it still isn't there really it's it's part of my job to build that it's it's the foundations and it's the real work like how what foundations you need to create AI value at scale so that's you know everything about the organization so I was the sort of a I was not the first AI person I was the second AI person because we already had one AI PO in the group, but there was nothing in terms of organization around actually building value at scale with AI.

7:00Sophie Buresi: No governance around it, no champions, no nothing. So we had to create everything from scratch. Also talking about governance, there was a lot of initiatives, lots of projects happening, but no really common way to choose to measure or to scale use cases. And in terms of technology, obviously we had things. I mean, GeoSurf is a big company, so there is data and there is a technical landscape. But there was no strong and central data and AI foundations from which we could easily build and scale use cases. So that's really what I discovered. And actually, it really shaped my first month and also the coming year.

7:50Sophie Buresi: It's really about building the AI platform. Because when we say QSRP, it's QSR platform. And so my mandate is really, I've taken the first months to really discover that. And the mandate is really to build the foundations and have that AI platform ready to accelerate on AI use cases in the coming months.

8:12Conor Sheridan: When you think about a mandate like that, you can think like how not to eat an elephant, right? I'd be talking internally around like elephant Carpaccio is like how do you break the problem into small slices? And you're talking around like an AI platform, you need obviously a broad data graph like in terms of workflows or use cases across all the concepts and geos. How do you know where to start or how do you select which use cases to begin with to maybe prove value? Or do you have to go more broad and say, actually, we need to have all of the core use cases that link to EBITDA or back office automation set and go after them all at once?

8:53Conor Sheridan: Or can you select some you think will have the largest impact on the P &L or the business? Or how did you approach the problem?

9:01Sophie Buresi: Well, I mean, as you're rightly saying, it's a huge topic. So at some point, we need to have a strategy. And so that's what we, I mean, we took a bit of time to precise that strategy. And part of the strategy is really about selecting priorities and making sure that the foundational work I was mentioning is driven by the priorities. I mean, the worst case scenario is to say, OK, here are all the great ideas that we have, but we need to take three years to build the foundations. And let's catch up in three years from now and maybe we will launch use cases. So when we built the strategy, we had to deep dive both aspects.

9:44Sophie Buresi: So obviously, exchange a lot with the business teams to understand what had been done already, what I wanted to do, you know, understand the sort of and build our portfolio of use case. and on the other side, look at the foundation and say, okay, if we take the portfolio and all the use cases and we take the ones that we believe are the most, like the priority because they concern maybe 80 % of our brands because they bring them more value. I mean, there are lots of ways to select the priority use cases. What are the priorities in terms of foundation? And then you build a story, you build the roadmap that goes with it.

10:24Sophie Buresi: And so that's how you start from an immense topic and you reach a structured roadmap with precise deliverables and precise goals on a quarterly or annual basis.

10:43Conor Sheridan: When you talk about the foundations, I can imagine lots of folks listening to this don't know where to start, right? They might have businesses in four to five geos, all using disparate systems with different architecture, data models, different outputs. How did you go about, maybe, or said another way, what did you do initially to look at the foundations and decide how best to set yourselves up for success in the future?

11:11Sophie Buresi: So I would say that good AI foundations are not about AI. It's about technical and data foundations. So when I arrived in the group, the sort of technical transformation was already happening. And especially when you're in a big group with lots of different brands, technical harmonization, I think it's the number one goal because it makes everything easier. So you don't want to, if each market and each brand has its own POS system, its own workforce management system, you know, it's all different systems, meaning all different data models, meaning different integration. It becomes a nightmare to manage.

11:54Sophie Buresi: So I would say that trying to harmonize as much as possible because you can't harmonize everything. But at least looking at having a sort of core technical platform, you know, applicative platform is the number one priority. And the second one is data. Data, because we're talking a lot about AI. and I think that's the risk with companies that sort of missed the data transformation step and arrived now with AI that feels a bit more easy with, you know, Gen AI and all these technologies that seem a bit more like easier to understand and to use. But AI is always about data and so you need to have your data foundations right if you want to create value afterwards with AI.

12:42Sophie Buresi: And so that's actually most of my AI foundational work is working together with the data BI teams to make sure that we have these foundations, right? So that we have all the data centralized in one place correctly, you know, with the correct model, with like correct quality, accessible. because once you've built that sort of central data foundations, then the cost of launching any use case is almost nothing. Really, the cost of building use cases is when you have to rebuild each time foundations for each use case. So I would focus really on that, the technical harmonization and data foundations.

13:26Conor Sheridan: Definitely, for anyone listening, that's the first place they should start. if the data quality or the data structure is not in a good place you're not going to get the value from the business point of view it's maybe not the sexiest part of the transformation but the most important part speaking of use cases one of the super cool use cases you undertook as a business was around drive-thru voice ordering that I think that you pushed into a couple of geos So the Netherlands, Belgium, Italy across some concepts. Maybe you could talk to us a bit around why that use case was a priority and how did the implementation land for the project?

14:11Sophie Buresi: Yeah, of course. So when we were thinking about where AI can bring value in the restaurant operations, of course, you think about drive-through because there are a lot of operational constraints to drive through. And your objective is to reduce the cost per lane in your drive and to increase the revenue per lane as well. And there are lots of ideas of how to use AI to do that. And the easiest idea is to say, okay, we're going to use AI to take the order, maybe improve the upsell recommendation during the order. So the ID in itself, I mean, I think for anyone working in QSTAR, it seems like an obvious ID and you want to test it.

14:59Sophie Buresi: And it seems like a no brainer. But actually, it is a brainer. You need to think about it twice. And it's the case for this use case, but it's the case for actually all use cases. Because when you think about the business model of the case, well, the vendor is going to charge you per lane. And it's the same price everywhere. But the value of the lane really depends on the brand, on the geography, you know, even on the store itself. And so if you change one of these criterias, then the same contract, the same use case, the same idea can go from a pure yes, no-brainer, let's go, to an actual no.

15:44Sophie Buresi: or to an intermediate situation, which is our case in most geographies, to say, yes, let's do it, but maybe not in all the restaurants because that wouldn't make sense. So the way we, I mean, what we did and that's what we do with all use cases, we really looked at the value per lane and what we can improve with AI. so on the cost per lane and the value per lane and we really run it per market to really understand where it makes sense and it's true that for example in Belgium the results even before launching the use cases we knew that the use cases made sense there because of their the cost of the staff over there because of the average order value there because of all these criterias that were positive.

16:44Sophie Buresi: But if you look at Italy, the dynamic is very different. They don't really have, you know, the sites, they don't have as much staff. The staff is a bit cheaper as well. The AOV is a bit lower. And so the economic equation is not the same. So we decided to launch in certain markets to postpone on other. and we will re-evaluate once we have the results of the first months of pilot in the initial markets. So it's not a definitive no in some markets. It's more a temporary freeze. Let's re-evaluate and potentially we can even re-evaluate the solution. Maybe in Italy, the solution is not about bringing AI to take the order in Drive2.

17:36Sophie Buresi: maybe we can find another way to bring value with AI in the, you know, in the drive-through journey, but not exactly that one.

17:43Conor Sheridan: Yeah, really interesting. Like prioritizing where the maximum, because back to what you said at the mandate, EBITDA, right? So if you look at like brand, geo, unit economics, where's the biggest impact in P &L? Prioritize those geos. And as you mentioned, you pause Italy and potentially as macro changes, inflation moves, it moves into a bucket where unit economics makes sense for that use case and you bring it into the business, which is quite a nimble way to operate for a business of your size. Typically, you don't see that as nimbly. Often we see businesses of your size will just try to mandate everything abroad based across the portfolio because.

18:24Sophie Buresi: yeah well in in an ideal world it would be easier to validate you know you validate one use case and you scale everywhere but we see that it's not the case and even in one brand in one market from one restaurant to another the economy equation can be very different and also as you were rightly saying even the context can change so even if we validate a use case uh for a market or for a brand we need to be very cautious about performance measurements on the long run because at some point the use case that was positive can become negative and then we need to be able to stop the case. So it's like the double challenge.

19:02Sophie Buresi: Before launching the case and while the use case is running you still need some ownership and making sure that we monitor performance.

19:12Conor Sheridan: So I segue into like KPI tracking or metric tracking. So often operators can be worried around, okay, I'll implement AI and I'll see efficiencies and profitability and productivity gains. I don't want that to come at the expense of guest experience or customer sentiment or customer retention. Like if you think around that use case, what were you looking at from the drive-thru point of view to determine if it was successful? There's obviously the labor cost or the revenue per lane, but you look at other submetrics within that use case to see if it's performing as desired.

19:50Sophie Buresi: So, yeah, the two most important points is AOV increase because actually you see a big impact when you have a recurrent product recommendation during order. So AOV was really important. And also staffing optimization. It doesn't mean sort of reducing the number of people, but maybe, you know, you can reallocate the time of the person taking the order to the kitchen. And especially during peak time, it can be really helpful. So these are the two most important metrics. But obviously, we're looking at other things as well. you want to look at you know the like the throughput and drive through speed making sure that in terms of customer experience people are you know taking less time to order if you have these two first metrics that are good but then each car takes 60 more seconds you know to have their order taken then it's going to be operationally speaking it's not going to be a good use case And also another element that we take into consideration, but it's not something that you can really monitor precisely with data being collected, is also customer sentiment about the case.

21:15Sophie Buresi: Because it typically is a very good example. even in Europe depending on the markets you have very different feedbacks from the customers on AI use cases especially when it's AI taking the order so you would have people in the northern countries are a bit more used to it and they feel a bit more at ease with sort of AI or digital technology to take orders people from the you know south countries like even in Italy they're not so comfortable having an AI taking the order. They're okay if the AI supports the person, but just having a pure AI taking the order is that in terms of customer sentiment and customer experience, it's not received in the same way.

21:59Sophie Buresi: And that's something very important because even if you don't see it in the pilot metrics on the long run, it may impact your brand equity. So you need to take that into consideration as well.

22:11Conor Sheridan: How did you get to that understanding and those nuances, was this something you were aware of going into market or during a pilot process, was that surfaced up to say actually feedback is materially different between markets in terms of comfort using the system?

22:27Sophie Buresi: Well, actually, it's, I mean, personally, I didn't know that point. But, I mean, I'm leading the group AI team, but I'm working on a daily basis with the business teams. And for each use case, even in pilot phase, we are working very closely with the local business teams and they know their market, they know their brand, they know their customers by heart. So we had that input from the local teams from day one. And then it's just confirmed, you know, with studies and, you know, things like that. But that's where collaboration between groups, technical teams, either data, AI, whatever, collaboration with business teams is very, very important because they know their business.

23:14Sophie Buresi: And sometimes when we launch use case, we have, you know, that precise framework with the listed KPIs and everything. And on paper, maybe everything will seem green, but the local person, the local operator will see that there is something missing. will spot the operational points that could block the use case. So that's why we need, it's not even about integrating and collaborating with local business teams. We need really operators. So when we design use case, when we pilot use case, we always have restaurant managers, district managers that are integrated from day one because they are the one people that will spot the problems ahead of any kind of use case launch.

24:00Conor Sheridan: Yeah, really nice. The operator has to be at the table, obviously to shape the outcome and shape the solution. But usually where a lot of these deployments fall down or that execution gap comes from adoption risk, right? Folks don't buy into the use case. When I look at your business, you're predominantly franchise-driven in some of the concepts. So in our experience of large-scale deployments, often what you would do with a business like yours is you work in some corporate-owned stores, maybe some tech-forward franchise-y stores or restaurants to get a view on differences in adoption and feedback because often they can be run differently on the two sides.

24:45Conor Sheridan: But beyond that pilot period, there's actually a whole different internal enablement and sales process that has to happen for a franchise system to then move the use case through the business to a high level of adoption. How do you think around when you've got a pilot, do you do this in like your own company-owned restaurants, prove the value and the ROI and the metrics and then bring it to franchisees and then bring it through the system? Or do you just roll it straight through? Be interested to understand how you take a use case from testing through to system-wide adoption.

25:21Sophie Buresi: So you're right that as all the heavy franchisee business, we start by testing the most of the time we test use cases in our company-owned restaurants because it's easier. We can go faster. But even if we test in the company-owned restaurant before opening the case to franchises, we always take into consideration the potential differences in operating model or any difference we could think about before proposing the use case to franchisees. Because, I mean, if you want to sell something to someone, you want to make sure that you understand your client's constraints. And so that's why we think about them from day one.

26:07Sophie Buresi: But so that's what we do. We start testing in the company owned. And then we want to make sure that we have all the good arguments for the franchisees to accept use case rollout. And if we've done our work correctly, most of the time it's a no brainer and they're okay. But because if we see during the pilot, any counter arguments, then it means that we wouldn't even push the use case to the franchisee. So, I mean, at the end of the day, the franchisee, he's like the CEO of his own company. So what he's going to look at is the payback, you know, he's going to get from the use case. It's their P &L.

26:52Sophie Buresi: It's not our P &L. So we want to make sure that the payback is super clear. And it's clear not in years, but how many months you will get a payback on that investment. Exactly the cash you will get. We want to, in terms of financials, we need to be super square on the figures that we're pushing. That's the first point. I think that obviously they also, you know, they trust what they see. So it's important, even if it's company-owned restaurants, it's always important when we select pilot restaurants that we select these restaurants correctly and maybe have different types of restaurants, not only flagships in city centers, but have a good, you know, diversity of restaurants to make sure that the franchisee will identify with maybe one of the company-owned restaurants, because obviously, you know, a franchisee that has a very small restaurant in the countryside, if I tell him, yeah, it's amazing, the use cases work so much in the center of Paris, it's going to be okay, great, but I'm not sure it will work exactly the same where I live.

28:01Sophie Buresi: And also the one point that we really take into consideration, and it's not really an AI topic, but that's why we work really closely with the business teams, is the operational impact of the use case. Because sometimes when you're too much into innovation, AI, data and everything, you're so thrilled about providing an AI innovation that you could forget about the operational impact. And at the end of the day, their work is to sell burgers, tacos, whatever, to the clients and to sell as much as possible. you know, keeping the customer happy. And so it's great to bring innovation, but if innovation creates complexity instead of removing complexity, then it's going to be a problem.

28:53Sophie Buresi: And franchises, they're super, super, super cautious about that point because they want to make sure that we reduce complexity and not increase complexity. So in our sort of argumentation, we always have that operational consideration and we know what will be the impact, if there is an impact. We want to make sure that we can answer their questions on that.

29:20Conor Sheridan: Yeah, it's a really great approach. It leads me into the next kind of use case or topic. So you talked about the change to an operating model or the kind of complexity curve for a franchisee. One of the use cases you've looked at involves hardware, which is kiosks. so i think in the long post implementation it would typically reduce complexity of an operating model but introducing any kind of hardware obviously involves capex a change to guest flow a change to the systemization of the business which has initial complexity how did you manage how do you manage deployments like that where you're introducing not just software but hardware a change to the guest flow.

30:07Conor Sheridan: Maybe you can talk to us a bit around how you navigated that and how do you select where to test those kind of deployments in the business?

30:16Sophie Buresi: Well, I mean, it's just like the AI use cases. For the case where we have hardware deployment, we need to be even extra cautious in terms of process consideration, operational impact. So, and also on the economic equation, because as you were saying, it's not only about OPEX, there is a strong CAPEX. So, what we're doing is we are working very closely with the group operations and local operations team to really not only, you know, select the technology and have a deployment plan, but have an adoption and a transformation plan. and most of the time it's owned by the business teams themselves. So it's not Sophie, head of AI, who's like pushing the adoption of that new scale or that new kiosk or that new, you know, whatever hardware.

31:17Sophie Buresi: It's, I don't know, Peter, head of operations for his brand, you know, really designing the new processes. And we are, I mean, that's why, you know, It's the strength of QSR. It's having really strong processes, really clear processes, and making sure that they are applied everywhere. So they want to make sure that the teams are onboarded correctly. And we put a lot of effort, a lot of time, and actually a lot of money in making sure that it's not only technology being dropped on the teams, but they have training, they have acculturation, They have, you know, people checking that they know how to manage these new tools.

32:01Sophie Buresi: And so most of the time we try to avoid a sort of, you know, deployment blast. So we would sort of do it step by step. And once again, the steps are being defined by the ops teams because they know their restaurants better than anyone. so they know in which store it's going to be super easy, in which store it's going to be a bit more complex. And so we design different waves of deployment and we really try to measure correctly adoption to potentially adapt our ways of working. We have different hardware-intensive use cases around AI. We have one leveraging scale, scales in Italy, which it's a great use case, but operationally speaking, it's not an easy use case.

32:58Sophie Buresi: You bring in new technology in the kitchen that's already filled with like hardware and steps and everything. And so typically on this kind of deployment, we often take more time on the pilot and the different waves to make sure that we have sufficient feedback from the teams on the process impact. impact but one good thing is that if you when you work with operational team and restaurant teams even if you take more time to really understand the impact etc uh you see if if the technology is not clear enough if it's not explained well enough you see directly because the teams they just won't use it so it's like you have a straight response from the field uh like if it's not uh if we haven't done the work correctly then it's not used and we see directly in the metrics so it's quite easy to see if we haven't done our work correctly honestly really interesting around like

33:57Conor Sheridan: the first mandate for ebita at brand level at restaurant level you mentioned the second mandate was hq automation for wanting a better term so how do you scale from 1300 restaurants to two and a half dozen restaurants without linearly scaling the hq function in the business What are the use cases or processes that you identified or are working through at the moment to try and achieve that outcome?

34:27Sophie Buresi: So for that second objective, we have two different streams that are complementary. The first one, we call it AI for All because it's always great to have nicknames to our streams. But it says, I mean, the name says it all. It's about making sure that we improve individual efficiency, you know, with AI and especially with LLM and Gen AI. So in that first stream, it's not about completely transforming the way people work. It's about making sure that anyone from the marketing teams, the finance teams, the ops team, whoever, especially in HQ, they have access to the selected secure environment, you know, that we have at QSRP.

35:12Sophie Buresi: and they can use as much as possible AI in their daily routine. And so on that first stream, our objective was, first of all, select, sort of build that secure environment. It's step one. Obviously, onboard, open licenses for everyone. It was not the case. Unboard everyone because not everyone was onboarded. Train people because most people, they use ChatGPT, Cloud, or whatever in their personal life, but they're not used at using these tools in their professional life. So they need a bit of training. There are a lot of things that they think they know and actually they don't know. And even when you've done these first steps, then it's a long run effort because you need to make sure that people, they keep using it.

36:01Sophie Buresi: You need to have a governance in place. You need to make sure that people understand what are the interesting use cases. So we selected AI champions in every department to be the sort of leader for, you know, AI use cases. And so I would say it's the easiest stream in terms of what we have to do as a group AI team. The effort is really in the long run, making sure that people keep using the technology and also in making sure that it brings value. And the value is not easy to monitor because it's individual value. But we know that it's here. And we have designed surveys to understand the impact.

36:47Sophie Buresi: And we can see it in, even if it's small impact, if, I don't know, you took maybe five hours during your week to build slides, to build presentation documents, and now you only take one hour, then it's already a great win. Because if you multiply by the number of weeks and the number of people in HQ teams, then you see the amount of hours that were not saved, but really replaced on higher value topics. So that's really the first stream. And the second stream, which is, I would say, way more complicated, but the impact is huge, is really process transformation. So here, it's not so much about individual efficiency.

37:31Sophie Buresi: It's about department centric efficiency. And so what we want to do is it's a bit of a consulting approach, but we need to go through that is understand how the team work. What are the processes? So it's like doing process mapping, understanding how they work. And sometimes we discover that we have no process or sometimes there are as many processes as there are people in a team. And so we want to understand how people work. the processes, what are the constraints, what are the processes that are actually, you know, too complex, that take too much time. And we want to rework on these ways of working on this process.

38:13Sophie Buresi: And why do I say that this second stream is very complex? It's because it's pushed by AI, because AI can be one solution. And when I say AI, I'm mostly talking here about agentic and automation, agentic generative AI, not so much about machine learning. So AI can be a solution, but it's only one tool. When you're considering process optimization, simplifying the ways of working, the big mistake would be to think that AI is the only response to that. No, actually, it must be a business-driven effort. and at the end of the day, most of the time, simplifying the process is not about adding an agent.

38:59It's about just breaking the process,

39:02Sophie Buresi: like rebuilding something way easier with less steps, improving communication, maybe making a role clearer and then you can add that extra layer of agentic or generative AI that will be sort of the cherry on the cake. But so that's why I'm saying that I said that it's a challenge is that the second stream process transformation can bring a lot of value, is pushed by AI because there is potential with AI. But it's something that must be owned by the business teams and shouldn't be reduced only to bringing AI on top of existing processes. that's very important. And in terms of transformation, it's not so easy.

39:49Sophie Buresi: And so at CSRP, I mean, you can't work with all the departments at once. So we really selected the departments where we saw the biggest impact potential. Finance is one, because most of the time, finance, even if they have softwares, there are a lot of things to make their life easier. and the second department we're working with right now is marketing and as I was saying it's it's not as if we we were done with finance and we switched to marketing it's like we started with finance and then we have an organization and a governance to support long-term transformation for finance and then we sort of duplicated that approach to marketing and then we will have another department and so that transformation will be department per department but it will take

40:41Conor Sheridan: months actually yeah really interesting and i love the code name so we have one internally called project 90 which is on the finance stream so how do we get initially 90 automation and say accounts payable flows so from supply chain to po to credit note to grn and then you want to have exception handling at the top end so there's still human in the loop but i think we what we found was when we went in from a first principles point of view so many systems so many different ways of running the process the workflows you have to start there and it goes back to what you said it's like the architecture the model the data if that's not all in a place where you can have an agent run a continuous process the hit up against walls they'll be doing the you're automating a mess right and it doesn't work and it doesn't end up saving people time yeah so it's super interesting

41:33Sophie Buresi: And that was a challenge when I arrived at PSRP because everyone was super excited about all this agentic and AI transformation. And at the beginning, I kept saying, well, yeah, but we need to work on processes. But yeah, we need to work on the data. I was like the annoying person saying that we were missing a lot of things before reaching that sort of cherry on the cake AI use case. but now after a few months we are in a good dynamic on that one but I think it's the new complexity for AI people before you had like five years ago my challenge was to convince people that we could do amazing things with AI now my challenge is not to say no to everyone and not to be that annoying person always reminding that we need foundations we need that, we need to work on processes we need to work on everything but AI before we can create value with AI

42:26Conor Sheridan: yeah definitely I think it's a great message for anyone listening about where most of the time all she needs to be spent for the business from an investment and a mandate point of view so we're going to move to the quick turn which is how we finish out the show so I appreciate everything so far it's rapid fire questions so this can be as quick as a couple of seconds or as long as a couple of minutes so you can determine how you want to answer them first ones first Monday morning, one number which one what do you look at what's the first metric you look at on a monday morning

43:01Sophie Buresi: um i mean even if i'm not an operational people i would look at the sales uh from last week especially the sales per labor hour uh because i mean it's the the one metric we want to see how the business is doing because based on how the business is doing it will have an impact on my work so that's very important to know these uh these metrics um what's one ai use case in qsr

43:25Conor Sheridan: that you think gets too much attention than it deserves? It's too focused on where there's many more value elsewhere?

43:33Sophie Buresi: I think it's not one use case. It's more a category of use case, but everything about kitchen robots. Obviously, you have fancy demonstration, but at the end of the day, especially in QSR when you have, the models are always really well thought and sometimes it's like mono products or really, you know, the menus are really well designed and they're quite easy to make. You know, kitchen robots, they do not bring that much value. Like having a robot to make your burger, it's really capex intensive and it's not going to save that much time. So I think there's a bit too much focus on that and not enough on, I would say, data AI use cases.

44:15Sophie Buresi: So everything about machine learning, forecast and improving orchestration based on the data and not so much based on the hardware in the kitchen.

44:25Conor Sheridan: Yeah, nice. We could do a whole episode on that topic, I think. So at your scale, build versus buy, how do you make that decision internally?

44:36Sophie Buresi: So, I mean, my answer is going to be really for QSRP. Maybe I wouldn't have had the same answer in my previous companies, but at QSRP, it's sort of buy per default because we're not a technology company. Even if we're building technical foundations and technical platforms, I prefer buying the technology. The one exception is for everything that really brings strategic differentiation. And so when it's really a core business case, then I want to be in capacity to build it. And so typically, I want to own my own demand forecast. And so I want to have the right data and AI foundations to have my demand forecast homemade and to be able to maintain and run that demand forecast.

45:24Conor Sheridan: And last one, one piece of advice for a CTO who's just been handed an AI mandate similar to yourselves, where should they start?

45:35Sophie Buresi: It goes back to my, I think my very first answer, but start with the data. Start with the data, because AI scales really on the foundations. So if they are good, then you will have good AI and good value. If they're bad, you won't have any value. So start with the data.

45:54Conor Sheridan: 100%. Brilliantly said. Well, Sophie, thanks so much for your time. It was great to have you on the show and awesome to meet you.

46:00Sophie Buresi: No, it was great. Thank you so much.

46:05Conor Sheridan: What an awesome conversation with Sophie. I want to say a huge thank you for joining the show and in particular for being so open about why some use cases like drive-thru make sense in some markets like Belgium but go on hold in others and why it's a pause and a revaluation rather than a no. Three things I've taken from our chat. The first is the business case is local even when the technology is not. Value per lane moves with the brand, the market and the store you need to price use cases market by market and keep measuring after launch because the use case that pays today can stop paying tomorrow and you need to be ready to switch it off and conversely when it doesn't pay today it might tomorrow it needs to be ready to go second the operator decides always sophie and qsrp testing company-owned restaurants and then when they're picking and rolling out the franchisees they pick locations that they'll recognize for example, a countryside operator wants to see data from a countryside location, not just a city center powerhouse.

47:08Conor Sheridan: And franchisees and operators need to see payback in months, not years. If a use case adds complexity in the restaurant, teams won't use it and the usage data will tell you straight away. Third, AI is the cherry on the cake. You need to harmonize your systems, centralize your data, and fix your processes first. Most of the time in deploying an AI use case is actually spent in simplifying a process, not adding an agent. And once the foundations are right, the cost of launching the next use case falls to almost nothing. But what is now clear is that the head of AI job was now flipped. Five years ago, it was convincing people AI could do amazing things.

47:49Conor Sheridan: Now it's not saying no to everyone, but making sure you've got the data foundations, the architecture, models, and pace, but actually let the AI pay. So if you know an operator about to roll out an AI pilot across every market at once, send them to this episode and subscribe to What's Cooking wherever you listen. Thanks for listening. See you next time.