00:00:07 The hard luxury market and the importance of in-store experience.
00:01:37 Defining hard luxury stores, their locations, and target markets.
00:03:10 The challenges of hard luxury store assortment and the role of novelty.
00:05:01 Using network-wide data to make better assortment decisions.
00:07:23 Repeat business in hard luxury and connecting products to clients.
00:10:00 Quantitative analysis of cannibalization and substitution in luxury product assortments.
00:12:00 The necessity of understanding the presence and absence of products in stores.
00:14:01 Using product attributes to better allocate stock.
00:15:18 Implementing a probabilistic model to optimize assortment performance.
00:16:50 The process of adding a new product to a store and evaluating its impact on sales and cannibalization.
00:18:02 Using quantitative analysis to optimize product allocation across stores.
00:19:14 Assessing the potential gains and losses of moving products between stores.
00:20:06 Implementing optimization in luxury stores and addressing the cultural challenges.
Summary
In this interview, Kieran Chandler and Joannes Vermorel, the founder of Lokad, discuss supply chain optimization in the hard luxury market. Vermorel highlights the unique challenges of determining which products to stock and how novelty drives sales in this market. He suggests using probabilistic modeling to optimize product placement, which takes into account sales odds and potential cannibalization. Implementing Lokad’s approach can be challenging due to the culture shock in transitioning from intuition-driven businesses to numerical optimization. To successfully implement this approach, Vermorel advises companies to approach the problem rationally and dispassionately.
Extended Summary
In this interview, Kieran Chandler discusses supply chain optimization in the hard luxury market with Joannes Vermorel, the founder of Lokad, a software company specializing in supply chain optimization. They explore the unique challenges faced by luxury stores in optimizing their assortment and the role of novelty in driving sales.
Vermorel explains that the majority of hard luxury purchases still occur in-store, despite the growing online presence. The challenge for these stores is to determine which products to stock, as sales can be as low as 1 to 2 units per year. In this market, it is the in-store experience and the stock itself that create demand. Customers are unlikely to purchase a $10,000 watch from a catalog, with only a small single-digit percentage of people doing so.
Hard luxury stores are typically located in one area of large metropolitan cities, where the most expensive shops are concentrated. Contrary to popular belief, the target market for these stores is mostly the upper-middle class, as billionaires represent a small fraction of the overall consumer base. Customers in this market may not be exceedingly wealthy but are likely to spend on expensive pieces of jewelry or watches.
In the hard luxury market, novelty plays a significant role in driving sales, although not to the same extent as in fast fashion. Collections are constantly being rotated, and the challenge for luxury stores is deciding which products to stock from their catalogs. These stores may have limited data to work with, as they often sell only one unit per product reference, and collections do not last for decades. However, Vermorel emphasizes that this does not mean quantitative analysis is impossible.
Over the past few decades, the hard luxury market has undergone consolidation and growth. Many well-known brands now have dozens of stores, making them more than just small-scale operations. From a supply chain perspective, the primary goal is to maximize market impact and fulfill the largest number of desires. However, the classic time series approach to supply chain analytics does not work in the hard luxury market due to the sparse nature of the data.
The interview highlights the unique challenges faced by hard luxury stores in optimizing their assortment and the importance of novelty in driving sales. Despite the limited data available, quantitative analysis is still possible in this market, and luxury stores must find innovative ways to maximize their market impact and cater to customer desires.
Vermorel explains that traditional time series analysis and clustering methods do not work well for luxury brands due to the low volume of sales and highly specialized products. Instead, he suggests leveraging a more comprehensive data structure, such as a graph that connects clients with products.
Vermorel emphasizes the unique aspects of luxury retail, such as the known client base and the repeat business nature, which is similar to e-commerce. Clients are often identified, and they tend to make multiple high-end purchases throughout their lifetime. By understanding the individual client’s preferences, retailers can better tailor their inventory to meet their clients’ desires and anticipate future purchases.
To optimize the assortment of luxury products in a store, Vermorel proposes analyzing the graph connecting customers with products. This graph can help identify the affinity between clients and products, as well as the cannibalization and substitution that can occur between different products. By choosing an assortment that maximizes affinity to the client population, luxury retailers can better cater to their customers and increase sales.
In response to the question of how easy it is to implement this approach in reality, Vermorel acknowledges the difficulty but highlights that luxury retailers generally have excellent traceability of their products due to the valuable nature of their inventory. While the systems may not be state-of-the-art, the robustness of the data can be used to understand not just sales but also what was present in a store at any given time. This information is crucial for understanding what choices were made and not made by clients, allowing retailers to better optimize their inventory.
Vermorel advocates for a more comprehensive and data-driven approach to managing inventory in luxury retail. By leveraging the unique aspects of the industry, such as the known client base and repeat business nature, retailers can optimize their assortment and cater to their customers more effectively. This approach requires a deep understanding of both the sales data and the inventory present in a store, but the results can lead to more successful luxury retail management.
Vermorel explains that stock levels can be reconstructed with a fairly accurate vision, even if only available at the end of the month. Hard luxury items typically have a few thousand units in a catalog, and about twenty attributes that qualify each product, such as precious metals, type of stones, price point, and subcollection.
When deciding where to place newly produced items, Lokad’s approach is quite different from traditional methods. Instead of pushing items to the best performing stores, Lokad uses probabilistic modeling to determine the optimal store for each unit. This model takes into account the odds of a product being sold over any given period of time, as well as potential cannibalization, where sales of the new item may negatively affect sales of other items.
By simulating the placement of each new unit in every store, Lokad can determine the optimal store for each item, which is not necessarily the best performing store. The model considers the diminishing returns due to cannibalization and may recommend placing the item in a top ten store rather than the number one store.
Once the model is in place, it can be used to assess the optimal placement of any unit in the network, as well as evaluate the outcome of moving units between stores. This helps businesses maximize sales while minimizing the negative impact of cannibalization.
Implementing Lokad’s approach in a luxury store may be challenging due to potential culture shock, as companies that have been driven by intuition for decades suddenly rely on numerical optimization. Vermorel suggests that the first step in implementing this approach is to rationalize and approach the problem dispassionately, which can be difficult in an industry driven by passion.
Full Transcript
Kieran Chandler: Today on Lokad TV, we’re going to understand this challenge and discuss what these luxury stores can do in order to optimize their assortment. So Joannes, what is it that’s so interesting about the assortment the luxury store holds in hard luxury?
Joannes Vermorel: What is interesting is that it’s literally your stock that creates demand. People are not going to buy a $10,000 watch on catalog. Maybe a few people do; we did some analysis with some clients on that, but it’s a small single-digit percentage. It’s really the experience in the store that can lead to a successful acquisition of a piece that is remarkably beautiful and fairly expensive. That’s one of the key angles of luxury in the first place.
Kieran Chandler: It might be useful to clarify what we mean when we’re talking about these hard luxury stores. Where are they located, who’s their target market, and what do they tend to sell?
Joannes Vermorel: They’re typically located in one area of every large metropolitan city, where you will find all the very expensive shops. There are historical and logistical reasons for this concentration. The audience is mostly the upper-middle class. If you’re a billionaire, even if you spend a lot of money on hard luxury goods, you’re not going to spend the majority of your fortune. There are only a few billionaires compared to people who make, let’s say, starting from $100,000 a year. These people might occasionally spend $5,000 on an expensive piece of jewelry or a watch.
Kieran Chandler: They’re certainly the kind of areas we’d be avoiding in Paris to a certain extent. You mentioned the catalog earlier; what are the key challenges these stores face?
Joannes Vermorel: The key challenge is deciding which items to include in the catalog. This is a business like fashion that is driven by novelty. Hard luxury is also driven by novelty, not to the extent of fast fashion, but still novelty is very important, and it’s really what is driving the sales. You combine the fact that you have very limited sales per product reference, and the fact that products don’t last two decades. You rotate your collections, which gives the impression that you have no data to do anything quantitative. That’s not true; nowadays, there are many well-known hard luxury brands with dozens of stores.
The interesting thing from a supply chain perspective, where you really think about how to maximize your impact in the market and fulfill the largest amount of desires, is that the classic time series perspective that is ubiquitous in supply chain analytics is completely wrong for hard luxury. Everything is way too sparse, but it would be the wrong conclusion to assume that because time series doesn’t work, there’s nothing you can do.
Kieran Chandler: Series don’t work at all in our luxury. That’s suddenly the only thing that you can do is guestimation, you know, and that’s the best you can do. So, what can we do then? Because I mean, we want your kind of describing here is if you’re just looking at the data of one store, there’s not actually that much data to, in fact, make some kind of relevant conclusions. But if you look across a whole network, you start to have something where you can kind of build a bigger picture.
Joannes Vermorel: Exactly. I mean, first, even for hard luxury brands, even if the amount of units being sold is small, we are still usually talking about tens of thousands of units per year, even for brands that are not initially among the largest brands. So, this is, in aggregate, very statistically significant. But if you start with certain classes of statistical methods that are going to be exceedingly poor, because those methods are very demanding in terms of the amounts of data.
Time series, for example, works very nicely if you’re like Procter & Gamble and producing shampoos in very large quantities, but time series requires a lot of data. So, time series are out. You have other methods like clustering that also work nicely if you have a lot of data, but clustering is going to be very bad by design in these situations.
The first insight is that you want to leverage more than just a time series perspective. Usually, every single unit you sell is to a known client. For these types of products, you have loyalty cards, loyalty programs, and for many reasons, you know your clients. So, hard luxury is a bit like e-commerce. Your clients are usually identified, and you know them even if it’s in a store. That’s a very good property.
Another thing that might not be super intuitive for hard luxury is that it’s repeat business. People who buy expensive watches are most likely going to buy several in their lifetime. It might sound a bit unfair, but it’s a sort of market where you have passionate amateurs that are going to buy an expensive piece every couple of years.
So, the starting point is to think of your products as a graph of your sales that connects your products to clients. It’s not like a time series where you see a unit being sold; you know that this unit has been sold to this particular client. And thus, you have a very rich data structure that connects your clients with your products. That’s probably the starting point to start thinking about this problem.
What does a good assortment mean? It means having the pieces in a store that maximize the coverage of desire for the clients that can walk into the store.
Kieran Chandler: So, basically, by having that understanding of what that individual client wants, you can ensure that most likely, in two or three years’ time, when they make their next purchase, you have the next latest, biggest thing ready for them in the store when they might want it.
Joannes Vermorel: Yes, the key idea is that if you want to start understanding the power of an assortment, you’re going to make a selection of pieces where at most, you will have one unit in the store. And in order to know what that even means, the power of an assortment, you need to think about the fact that you have a lot of cannibalization, substitutions taking place. There is no mechanical compatibility; you’re not selling car parts. So watches, you know, pretty much, you will have models for men, models for women, and maybe you have some models that are better suited for large men and some models.
Kieran Chandler: That are not suited for people that are shorter or thinner. That being said, there are tons of cannibalization substitution that are possible. People can choose between things that are widely different. So the bottom line is, how do you quantitatively address that? Most people think incorrectly that it’s just impossible and it’s only human insight who can grasp this sort of analysis. I am absolutely not in agreement with that.
Joannes Vermorel: The way you can quantitatively assess cannibalization substitution is by analyzing the graph that connects customers with products. That’s the way you can basically establish some kind of affinity where every single client that you’ve ever observed is basically defined by the products that have been purchased in the past. Every single product has also a profile, and every product has an affinity to certain clients. When you want to choose an assortment, you want to choose an assortment that maximizes your affinity to the population of clients that can actually enter the store.
Kieran Chandler: So the main aim of the game is to have the optimum assortment for each store and for all of the clients that frequent that store. From a data perspective, it all sounds great, but how easy is it to actually do in reality?
Joannes Vermorel: I would not say that it’s easy, but my experience with hard luxury is that it’s not the worst. First, those companies have excellent traceability, which is unsurprising because when you’re selling expensive things made of precious metals, traceability is very important. In this sort of business, an absolute traceability of every single part or piece has been in place, typically for decades. It’s not necessarily state-of-the-art cloud-based IT systems, but usually, it’s fairly robust. Given enough time and effort, you can get all the data that you need. By the way, the data is not just the sales; that’s only half of the picture. You need to know what was present at any point in time in any store because it’s not just the sales that matter, it’s understanding what was present in the store when this unit sold or conversely when this unit was not sold, what was actually lacking from the store.
Kieran Chandler: So the idea is you’re understanding what were the choices that weren’t made as well as the choice that was made?
Joannes Vermorel: Absolutely. You need to know day by day what exactly was the assortment at any point in time. You don’t need to have super high precision; for example, because inventory rotates very slowly, even if you only have historical stock levels at the end of the month, it’s not the end of the world. You can reconstruct a fairly accurate vision of what were the stock levels, and by stock level, I pretty much mean what was present in the store or not at any point in time. Then you need also to have some attributes about the products, but again in hard luxury, your catalogs are large, several thousands of units, but they are not exceedingly large. You can have a lot of attributes, such as the precious metals that are being used, the type of stones, the price point, the style, maybe the subcollection, etc. So you can have typically about twenty attributes that qualify your products.
Kieran Chandler: Okay, so if I’m a store manager for example, and I’m a brand and I’ve got twenty units that I need to know where to put and when to place them, what would you actually do?
Joannes Vermorel: Typically, stock allocation is going to be decided by the central.
Kieran Chandler: So, Joannes, let me ask you this. You produce 20 units for a novelty item. What do you do with them?
Joannes Vermorel: First, you need to have a probabilistic modelization of the power of your assortment. That means that you take a store, you have those units and that will directly tell you what are the odds, expressed with probabilities of having every single one of those units being sold over any period of time. That can be next month, next quarter, next year. It’s probabilistic forecasting because we’re talking about numbers that are very small.
Kieran Chandler: Okay, so what’s the next step then?
Joannes Vermorel: Well, you have a new novel item, and you need to push it. You take the first store and you kind of throw my model to basically say, “Dear model, tell me if I add this one unit, what is the outcome?” By the way, you will simulate what are the future sales with respective probabilities and you will see what is the priority that this unit gets sold. But also, how much cannibalization is involved because if you push this novelty to this store, it might actually sell well, but if you’re just cannibalizing units around, then basically you’ve not gained much. You can assess for one unit that you pass in a store exactly what is the financial outcome, thanks to a probabilistic forecast, and that takes into account cannibalization. Then you can simulate the same being pushed to another store and to a third store and the fourth until you can simulate to all your stores. From that, you can assess what is the store that is going to give you the best results.
Kieran Chandler: Okay, that makes sense. So, how do you determine which store is the best?
Joannes Vermorel: You would think it’s just going to be the best store, but not really because your best store has already a lot of stock. It has already a lot of stuff, so you get diminishing returns because of those cannibalizations. Typically, when we are going to start to have this quantitative analysis, we are going to realize that the first unit that you want to push somewhere in your network, well, the best location is not the number one store. Usually, it’s going to be still a fairly good store in the if you look at the world list of stores, but it might not be the first one. It might actually be something in the top ten stores instead of being the number one store.
Kieran Chandler: I see. So, what about the second unit? How do you determine where to allocate it?
Joannes Vermorel: You’re going to repeat this process that we just did, except that you’ve already placed the first unit, so you know that one store has now an assortment that is slightly different because it has one more unit of something. You can repeat that again and again, and that will give you a complete spread. The interesting thing is that once you have this kind of modernization in place, you can go and look at any unit that is currently in the network and ask yourself, “Is this unit really where it should be? Can I just move it somewhere else, and what is the outcome?” When you move an item, you will basically probe your mathematical predictive model, which is a representation of the power of the assortment, by basically asking the question, “If I remove this unit from this store, what do I lose?”
Kieran Chandler: You probably lose something because suddenly, you know, there is less, and you’ve kind of diminished the power of the assortment of this store. What do you gain because you’ve actually moved that unit somewhere else?
Joannes Vermorel: There’s some friction which represents the time in which the unit will be traveling.
Kieran Chandler: Okay, final question then. If you are a luxury store and you want to implement this, what are the first steps you should take? Is it all about ensuring you’ve got all of the client data, ensuring you know exactly what are the sales you’ve made, but also knowing what you’ve held in the store as well?
Joannes Vermorel: The first question would be, are you even ready to start, you know, in terms of culture shock doing the optimization? Because usually, it’s kind of a culture shock. Maybe the best store was just considering itself as entitled to 100 percent of the novelty, and maybe suddenly, when you stop playing this game, you would say, “Well, you’re entitled to a fair share of the novelty, maybe more than half of the novelty, maybe not 100 percent.” So it’s a culture shock, you know. It’s a real shock when you start doing optimization, especially when you’re optimizing in terms of dollars of error, dollars of opportunity. When you’re in a company that has been driven by intuition quite successfully, I would add, for several decades, it’s a real shock. And also, sometimes it’s a real shock to even realize that your intuition was incorrect. I mean, people can put a lot of egos on that. That’s the problem with numeric optimization – it doesn’t care about your opinion on the facts. The culture shock can be quite strong, and that’s probably the first step. It’s probably to start by trying to rationalize and approach this problem in a fairly dispassionate way, which can be very difficult in an industry driven by passion. That’s probably the first step and the most difficult step for most luxury brands.
Kieran Chandler: Okay, pretty much. That’s everything for this week. Thanks very much for tuning in, and we’ll see you again in the next episode. Bye for now.
