I’m Felix. I’m one of the founders of 7Learnings. My background is I was working for Zalando before for the price optimization department. For every price, basically all markets, all products for the Zalando before that I was a consultant at Kearney And I yeah, I really enjoyed the solution of predictive pricing, because it’s quite kind of has a simplicity and a beauty therefore I thought, okay, that’s a great idea for, for creating great product for retailers. Yeah, I’m super happy to have you today at our podcast, Robert. Maybe you could quickly introduce yourself as a first start. So my name is Robert. Now, my background of how I actually got into pricing isn’t that conventional route. So at university I studied physics because I basically loved all things to do with physics, trying to work out what was driving systems, understanding those changes. I then went on to do a PhD in semiconductor physics. And it was probably partway through that PhD. I realized that what was really interesting to me wasn’t necessarily all the physics. It was actually the system as you’re applying those problems to that ability to look and see what was driving something else that if I know A, and I know B, then I can deduce C. So for those underlying relationships, and that’s when, after that, I kind of moved into management consulting for a couple of years. And then over to retail, starting off with John Lewis partnership from Waitrose and then on to Currys. And it’s been that thing I find fascinating about retail. There’s so many different interactions. And for each of them, you can actually try and work out what is driving them. What’s the causal facts behind them to try and make their systems better and make them work harder for both the retailer and also the customer. And then over time slowly drifting into the pricing world as well. So you now only focus on pricing or you also do other things? So I’m looking after kind of what we call decision sciences. Now decision sciences that are to actually how do you use that data to make the decisions? It’s not about making that perfect model. It’s about working out where you can interfere or modify a system to make it work better. And what are those and what are those fundamental drivers behind it? So we’re looking at Basic pricing, promotions, things like sales forecasting, customer lifetime value, starting to look a bit at ranging in the past. We’ve looked at basic delivery networks, anything you can break that system down into a set of manageable components to optimize. That’s kind of what fits within the decision science space. Okay. That’s quite broad. Talk about, talk about pricing, but it’s not that you’re not afraid that large language models will replace your job anytime soon. Not really. I think the stuff that could do with those large language models of actually interpreting and not getting them to understand or asking the questions about what’s driving something. But I think behind every great LLM, there has to be a really solid data model of causality. Yeah. I mean, you know, as well as I do, when you look at machine learning models and AI, it’s very easy to get random factors coming in. So I think large language, great ability to integrate the data as well as driving things, but you’ve got to have those right, you’ve got to have the right data models underneath it. Mm hmm. And like for, for the use cases that you mentioned, like what, what was, what do you think is the, the most impactful use case? I mean, in terms of like, I see applying what LLMs or just applying ML. No, no. The, the, the one before, like in terms of like, was it pricing or is it assortment or is it marketing promotions or which one we think is the, for the average retailer, or maybe also for what was having the biggest impact. I think when you talk about what’s having the biggest impact, impact is a function of two different things. One of them is your ability to actually change those metrics, which you’re, which you’re modeling and influencing. The second one is actually how many touch points are you having with that customer? So if you think about it from a marketing perspective, there’ll be a few really big things you can deploy, which might impact your whole assortment, your whole customer base. You don’t actually need to have very much change on those. A small impact on a large number of people will have a big difference for the business. Now, when we look, when I think about pricing, pricing clearly impacts everything in the business, but it’s one of those foundational levers. If your pricing strategy is right, you can drive a lot of value. If you’re slightly off, it’s easy to fall behind, either becoming uncompetitive, losing that customer trust or basically investing too heavily. So there’s a balance there. So I think pricing is one of those highest agency models. I also saw that Gartner, sorry, but I also saw the Gartner list pricing as the highest impact AI use case. I think it’s super important for the audience also to understand, like people talk a lot about LLMs, but then when you look at actual applications for retailers I think the chatbot comes like at number 10 or something. And, and that’s where you can use LLMs. But there are nine topics before which have even much higher impact. Yeah, because if you think about it, let’s just say pricing impacts pretty much every decision that customer is making in the store. Options will be a close second because how you actually highlight what value means to you, drawing the customer’s eye to them is going to have a big impact. If you drop down the list, you’d start to look at marketing, the ability to average and segment those campaigns, and understanding which customers are most likely to respond to others. And I think you’ve got to go a little bit further down the list before you get to more of like the elements of micro personalization. Because in Asia, you may have a really big impact at the customer level, but the number of touch points is declining rapidly. So I think pricing is definitely out there. Okay, cool. Yeah, so maybe, maybe you can quickly describe then your yeah, touch points on, or like, maybe your how you developed your pricing in the last five to 10 years, then is that like at a broad level? I think at a broad level. In terms of how we’ve dealt with pricing, well, there’s two different aspects to that. The first one is actually how you start to migrate and use the models to make better decisions. But those models don’t really mean anything unless you’re from the hearts and minds of the categories and the people you’re trying to implement and support with. And I think you need those two really in parallel. So one of the things which we’ve looked at is, it isn’t just about what are those models you put in place, it’s how you get them working for the categories. How you actually understand the decisions they’re normally making. And how do you, so that sounds like you are using that more as a support for the, so there is a human decision in place, but you’re supporting that human decision with your models, with the output of the models. Yeah. Because if you think about, like, if you look at most retailers, right, what is one of the biggest challenges in pricing? One of the biggest challenges you see is time, right? In retail time costs money. So any retailer, you may have a range of sort of like 10 to 20,000 SKUs. Now, not all of those are going to be of equal importance for the customer base. You might be competing against 10 or 15 different other retailers, either online or through store tells the volume of signals you have coming in is huge, and it grows every year as systems become more sophisticated. Online retailers get slightly more aggressive or change prices faster. So the hardest part there is actually saying what is it that you need to react to when. It’s literally you go back to some of the most basic pricing systems and I’ve seen kind of like the price scraping companies. Yeah. They can give me a list of this is your price. This is the list and different competitors. It’s not really helping me work out what is it I need to do. And for each of those decisions, there’s a set of things. What is my stock position? What’s the base margin? Actually, how do I think the customers are going to respond? There’s a tremendous amount of information which you would have to put into each decision. Yeah. So I think where you start to get to is those automated systems. The first thing you want them doing is helping prioritize what are the key decisions you need to make each time. But still like my point, how do you support that decision? So you’re generating a prediction for different price options and then the category manager decides or how does it work in practice? So in practice, I think most of these systems as we’ve talked about the level of agency. I’m completely against just letting the machine set all the prices. Right what I believe you have to do is that is how the algorithm recommends what action you should take. And then those can be passed through the buying teams because the buying teams have to say, does this fit with what is my value with how I’m trying to convey value to the customer? Is this a move I want to make right now? Is this something we’re actually looking at? I’ve actually got the line going on promotion. Maybe I need to do like that as you go down that list of the biggest decisions all the way to the smallest ones. There’s a point where you get the impact of this. Is the risk level is sufficiently low that I can start to automate . And over time when you’re working with the back teams, you can see what level of cutoff is appropriate. So you give a recommendation to the team and then they are still allowed to overwrite it or change it? Yes, because I think ultimately you’ve got over time then the buying teams get more used to the recommendations. They can see that the algorithms are thinking in a similar way to which they would, you’re building out the trust and then you can roll it out. So you’d start, you’d start with the, like the slowest moving lines, the lowest risk ones, and as long as you’ve got monitoring in place to check, nothing’s right, you can slowly increase that level of automation. And to get that buy in from the, from the, from the merchandisers, like how like, how do you explain the decision of your yeah, of your optimizer? It’s a good question, but at the heart of it, what, what is the optimization engine really doing? You have an idea of what the price sensitivity of that product is comparing and saying for a certain, for a certain set of prices. Where am I going to be maximizing my volume? Where am I going to be maximizing my sales? Where am I going to be maximizing the margin? And against that, you’ve got a set of other constraints, which is, what is the market pricing? Because nobody’s cracking the vacuum. There’s always a reference point which you’ve got to hold it to. And you can never be too far away from that point. So I think it’s a case of actually saying to the commercial teams, This is why we’ve come up with that recommendation. This is that logic. And basically make it as simple as possible. Kind of take all the complexity, how you calculated those elasticities, how you run all those optimization functions, the degree to which you need to understand everything? No, because in this case, you’re looking for what is that response? Do you think that it should be a central team like a central pricing team making decisions, or should it be really the merchandise team and then in this case, how many should review these prices? Well, a lot of it then depends on what is your overall organizational structure and where does the responsibility set because one of the key things we commenced about product and one of the reasons it is in that high impact decision is it can instantly impact your sales. It can instantly impact those margin targets now with any organization be it Currys be it Waitrose where I worked certain levels of the organization or certain departments have certain targets. So my pricing perspective, if it’s the, if it’s the buyers and the trading team who have that sales target, they’ve got to hit, you have to balance the margin, then they’re the ones who need to be approving those pricing decisions. Because ultimately the machines are acting on their behalf to try and hit a set of targets. No, that’s a governance problem. We had that problem actually at Zalando as well. Like it’s the category managers were getting the target, revenue, target, profit, targets, and so on. And so basically we had a central pricing team, but you had to gather and give them some impact on the prices because otherwise they’d say, okay, how am I supposed to hit my targets if I’m not allowed to change the prices, basically. But that’s where it comes to winning that hearts and minds. And that’s you saying, what is that overarching strategy you’re running for each category? Because when you talk about using an algorithmic optimization, there are many different things you can optimize for. You can be optimizing for sales, you can be optimizing for margin, you can be optimizing for certain stock turnover volume, or trying to hit certain events. So by having those discussions with the categories and saying, well, what is it that we’re looking to do within the next period? You can set those up as closely to replicate the decisions they’d be making. Yep. And then you can feed that information down to the merchandising teams in terms of is there going to be a material impact on the stock forecast? Are there going to be any other external pressures you’ve got to make to measure? So pricing is really the heart of it, but it’s how you then disseminate that information and those outputs across the business in order to make it work. Yeah. And I would also say that the underlying thing that has to be there as a ground thing is that you have a certain governance where you know who makes the decision on the target in the first place. Yeah. I guess also you’re right. It could work in both cases. I still have a tendency to say that shouldn’t be like 100 people reviewing these prices because it’s all still like a little bit too difficult to to teach or I kind of to explain to all of these people at the same time, how the system works. Yeah. So that’s just my recommendation. But yeah, I think what that means is when you’re When you start to think what is the type of system you want to put in place. You go for that in source versus outsource model. It’s thinking about where is it that you’re that the tooling you’re you’re you’re buying or creating? Where is that possibility and that to change versus where is the organizational flexibility to change? Is you can have the best tool in the world if you don’t get the buying it doesn’t work You can have the best buying but if the tooling isn’t quite right It is really trying to work and it’s trying to find that balance. Yeah. Okay. So you already touched that. I mean, generally, how do you assess the performance of the pricing solution? Like, even like when you come up with a new idea how do you assess that new idea? I think it is a tricky one because within pricing, you’re making a lot of small movements, which collectively have a big impact, but on their own can be hard to measure. You’d be looking for things like, what is, what level of recommendation are you seeing? But you’d also be trying to understand if you aren’t getting high level recommendations, what’s driving it? So you can work with the buying team to get past those challenges. One thing which we tried to do is saying, how do you track the ongoing impact of pricing? Now it’s hard, because it’s not something you can necessarily do easy A, B tests on. So we’re trying to, we’ve been looking at how do we use simplified digit twin models to track that impact. Do you understand what would happen if we made different sets of decisions? So there’s a series of moves you can take to measure that impact. So we’re twin models. You mean you simulate the impact of that. And then, yes, we’re trying to simulate what would have happened if we’d done something different. Yeah. Because that thing gives us confidence. Are we making the right decisions? What would the alternate strategy have brought? Okay. And that gives you flexibility of playing around saying if we run scenario A versus B, What change would we have seen within the market? Is that within, is that, is that along the lines we want to go? Okay, but you never did an A B test on pricing. We’ve done A B tests in some areas. I’ve run those in the past within Waitrose. We’ve done it in the past within Curry’s. One of the challenges though is in a lot of markets, you’re changing prices globally. I don’t know that in a lot of A B tests, it helps have an A and B group where you’re treating things differently. So it’s common within marketing. Someone gets version A, someone gets version B, which is the winner. You don’t actually have that freedom within pricing because you can’t chart a different price in store group A versus store group B because there’s a lot of overlap. And so it does get harder to measure those. You can try and find similar categories which are behaving similarly. If there’s a little bit of luck there. If the market changes, Yeah. Can get those impacts. Absolutely. Yeah. It’s yeah, there’s always flaws in every setup. I mean, online, you can try the traffic split, but I also don’t have very good experience with that. And what works well for us actually is that you do a yeah, like a product based split, you give 50% of the products to one group and 50% to the other group. But also there, there, there are flaws in that as well. You know, if you’re running out of stock, for example, on one product, that’s a problem. Well, I think the challenge is if those two groups are essentially the same product. The lift from one intrinsically cannibalizes the other. Yeah, that’s possible. There’s some sort of duality between them. Yeah, no, that’s, that’s true. Okay. So maybe to that question which you already touched, this insource and outsource how do you make that decision? As I said, I mean, one of the key things before I’d advise anybody who’s looking at making that decision is to start by looking at what your own present pricing process is. Okay. So start understanding what are the needs of the buying teams you’ve got, what’s the nature of the decisions and what is unique about how you operate at the moment? Because if you’re going out to the outsourced market, you’ll start off with some vanilla software. Which you’ll then have to try and customize now that can be quite difficult from an operations perspective. So you’ve got to be clear: What is it that is unique to your business which is hard to replicate as well? Versus where is it you can pull in and start to use standard systems and standard processes? The other thing is thinking, actually, what is the point? What is what? What? What is one of your cores? What are your core competencies you need as a retailer? And where is it that things which are slightly outside of that sphere of influence? So if you think, for example, about competitive price tracking, is that the most efficient thing for any retailer to be doing on their own and most cost effective? Probably not. There are solutions out there where you can get market data. Now that’s the example of where it doesn’t make sense in my mind that anyone would embark on that on their own. But in terms of identifying your own workout, your own price elasticity or price optimization engines, that’s something where you can create a point of difference. And so that’s something which you would consider bringing in house, because small differences there can have a big impact on, on, on the business. You can still make the decision in house, but use an external solution helping you with that. Right. So that’s, think what I was trying to get to. So but, I think you’ve got to look at that external solution of how well does it, how well does the way it operates within your existing business, existing business aims. Because it’s the end of the day, you can have the best toolkit in the world, but if you haven’t got the hearts and minds of the bars behind it, people struggle and I, and I think that for me is the real talent is finding the alignment between the external versus the internal processes you’ve got. Yeah. Okay. Maybe last question going forward. Like, what are you most excited about, like in terms of roadmap in the coming one or two years, let’s say, where do you think the market and maybe also your own development is going? I think the big which excites me is if I look back over the last five to 10 years. The availability of compute through cloud providers has grown exponentially. So what it means is it’s opening the door to answering problems we didn’t think we’d be able to look at in the past. It’s bringing a whole level of sophistication in terms of basic data mining, understanding what is really driving different parts of the business, arranging, stocking point of view. You’ve got a lot more AI based tools coming in, identifying where there are opportunities within, within, within the business, what’s performing well, what isn’t. On the one hand, that’s really exciting. There’s so many options. The slightly scary part is working out how do you get all of those interacting properly with the business? Yeah. How do you get the right prioritization? How do you manage the expectations of what can and can’t be delivered? I think this goes back to what you opened with. There’s a whole set of challenges which are, which are there in the retail environment. And it’s picking and going, which one do we choose next? Which is going to have that biggest impact? But crucially, what’s the best way of ordering them? So you get them, so you can build the capability as you go. Yeah, yeah, that’s maybe, I think it’s really, it’s a good observation that there is a lot of potential. And I mean, I think certainly we see a lot of development also in AI for sure in LLMs at the moment, but also in normal machine learning models. But then somehow applications are not really catching up with that. That’s the feeling, right? So you have models which are much more capable than before, but the applications are not really getting that much better yet. I also see that that’s something that’s hopefully going to happen. Because most retailers are really struggling getting out of that manual rule based crawling based approach from a pricing perspective. I would still say that 95 percent of the retailers are still in that world at the moment. And we, for example, have seen like based on, for example, some like really low margin retailers like an AI based approach can give you like 50 percent more profit or something like that versus a purely competitor crawling based approach. So the impacts are gigantic. But yeah, it does give me the recommend to a V any retailer. Where would you start? The reality is pricing is probably one of those easiest areas to start in Because you’ve only you’ve got a handful of really of key levers. You can apply globally You’ve got pricing you’ve got promotions You’ve got what do you do with your range by the time you get down to marketing? It’s harder and harder to take those models and apply them at a global level. It’s hard results when you go down to lower levels of personalization. There’s more deals. You need to consider offer orchestration and the like, but with pricing, you kind of are in control, you’ve got that value chain, you’ve got a limited number of people making big decisions. So you can support them and crucially the high level outputs of those models. What are those elasticities? What’s driving the promotions? Those insights come out really nicely to start to dictate what should your marketing plan to be. Actually it’s coherent. Good. You’ve got that one pillar of ownership. It’s easy to get that in. And it’s also relatively easy to change the price. Yeah, that’s another thing that especially when you’re online or you have electronic shelves then super easy to make the change actually to execute the decision. Now if you were talking about a new warehouse setup or something like that, that’s probably also possible But it’s much more difficult to implement. It’s also one way you can do simple proof of concepts and actually see the change. Yeah, A/B testing is hard, but if you’re deploying it and going one part of your range first versus others You can see the benefits.It becomes self funding after a while that investment front. Yeah others wait, it’s bigger investments. It takes longest to pay back. It’s harder to get that halfway house working of semi automated semi heuristic. You’ll find your struggles to get those in as quickly. Cool. Okay. I think that was really really good insights from from you super excited that you are also seeing many things like I see them actually. That’s really cool. I think there’s a lot of like exciting times ahead for the retail world based on all those changes and new capabilities. I think the one key insight from Robert is really that you have to think about the users and the merchandisers who are looking at the decision coming out of your system. Basically, the AI systems, even if they automate the decision they have to be, they have to explain what they’re doing and where the decision is coming from. If you don’t have that in place, then it probably won’t work because even if, even if you can somehow prove that it’s optimal it probably won’t, it might actually still fail in the implementation phase. For me, I think there’s, look, there’s an awful lot of opportunity within, within pricing or any of the AI systems. But the key thing is winning the hearts and minds of those you expect to use it. Because if you’ve not got that implementation, you’ve got a really interesting model. But it’s just maths. What we’ve got to do as data scientists is find a way to actually get the trust on board, start to use it. Because that’s where you see the value coming from. In some cases, you may say, look, we’ve got a perfect solution, which is wonderful and gives us the best optimization. Sometimes it’s better to actually fall back and go with, we want something which is better than what we’ve got today. And let’s get that in practice. Let’s learn from why people are accepting where people are recommending it and then build out from there and test your way to better. Thank you so much Robert for your time. And yeah, enjoy the rest of your day and the rest of your week. Thank you for hosting me and I’ll see you again soon.
