00:00:00 Introduction
00:02:01 Collaboration project timeline and focus
00:03:03 Classic Lokad scope and reverse sales
00:04:18 Stock inflows, demand forecast, and AI buzz
00:06:00 Probabilistic algorithms and parts classification
00:07:30 Dispatch recommendations
00:09:15 Automation, AI execution, and stock shifting
00:12:26 Estimating stock and sales scenarios
00:13:48 Long-term stock assessment and ROI logic
00:15:13 Stock out penalty and customer trust
00:17:07 2025 project completion and expansion
Summary
In a recent interview, Fabian Hoehner, Commercial Director at Lokad, and Elliot Langella, Lead Supply Chain Scientist at Lokad, discussed their participation in the ATR digital conference in Athens, hosted by the Tokic group, a Croatian automotive aftermarket company. They highlighted the integration of AI in the automotive aftermarket sector, focusing on their three-year collaboration with Tokic. This partnership, which began in late 2021, aimed to optimize stock levels and improve supply chain efficiency through AI-driven decision-making and demand forecasting. The results included a 5% increase in service levels and a 10% rise in sales, showcasing the transformative potential of AI in supply chain management.
Full Transcript
Fabian Hoehner: Hello, here from our Paris office.
Elliot, we’ve been in Athens together a few weeks ago together with Josip from the Tokic group, and now, well, we’re here in our new beautiful studio and talking about what we were doing in Athens. Can you give us a little bit of an insight from your perspective? What were you doing there?
Elliot Langella: Athens was a conference about AI in automotive aftermarket applications, results, you know, just some insight from the different companies that were attending. And so we were there with Tokic, presenting the result of our collaboration of the last three years.
Fabian Hoehner: Maybe you can start us out by just reintroducing the Tokic group. Josip has already done it, but who are they, especially from a supply chain perspective? Just give us an idea.
Elliot Langella: Tokic group is one of the leading auto parts retailers in Croatia and Slovenia. More than 130 stores now, I believe, and growing. More than 150,000 parts in the catalog. So you multiply that by the stores, that’s a lot of variants, right? A lot of ones.
So they’re working with several hundreds of suppliers, proposing either, you know, usual repair shop products, just spare parts that you use to fix the cars, but also some specialty products like agricultural trucks. And so they do serve obviously local demand from Croatia and Slovenia, but also export customers. So they’re kind of an important player on the Balkan markets.
Fabian Hoehner: So when we talk about their clients, well, now we have a super fancy screen in the background. Is that what we have to imagine as their clients? Are these the kind of products?
Elliot Langella: Yes, exactly. That’s a good chunk of their customers, actually. Mechanics and, you know, local owners of repair shops that just come into the Tokic store and buy whatever they will be needing to fix their customer cars in the next couple of weeks.
Fabian Hoehner: Okay, then talking a little bit more concretely about, well, the collaboration with Lokad. What was the project? What was the timeline? Give us some ideas.
Elliot Langella: So we started out at the end of 2021, and well, that was a very difficult time for supply chain at this moment. You know, COVID was disrupting supply and demand, obviously with lockdowns. So it was a challenging time for retail companies. You know, you’re not sure whether things are going to be back to normal, either supply or demand side.
So the first thing that we started out with Tokic is to work on the supplier lead times and supplier service level estimation so that they could get a better feel of how far off they would be from the optimal stock levels they should be holding. After a few months, we ended up going live on a pilot phase for replenishment, so daily push of the stocks from the warehouse to the stores. And a few months later, now we’re fully live both on purchase planning and daily dispatch.
Fabian Hoehner: Okay, that sounds like a fairly classic Lokad scope. Was there anything special? I mean, you’ve done quite a few implementations over the years. What was interesting for you? Was that something that you hadn’t seen before?
Elliot Langella: I think that what Tokic group calls reverse sales is quite interesting. So we have all been buying things online on Amazon or Zalando, and you know, sometimes there’s a problem, it’s not the good size, and you can return back the products. This is not so often easy to implement and to propose to your customers in auto parts, spare parts. But what Tokic group does is provide their customers the possibility to just come in, get a bit more units than what they would be paying today.
You know, they get basically three weeks later or end of the month’s payment, and only then can they choose to actually buy the part or rather return it back to the Tokic store.
Fabian Hoehner: That’s also for the physical, I mean, in e-commerce, I think that’s even for auto parts somewhat, well, even by law, but they’re doing that in their physical stores?
Elliot Langella: On physical stores directly. So you could literally be, you know, getting additional clutches and filters. You get 10 instead of the few ones you were planning to get, and you know, three weeks later you can give back two because you ended up using three instead of four.
Fabian Hoehner: Okay, and what’s the consequence for you from a, well, supply chain scientist perspective? What are the complexities that this is creating?
Elliot Langella: This creates additional stock inflows and outflows that you have to take into account in both purchase planning at central warehouse level, but most importantly, daily replenishment to the stores. Well, if you think that you’re short today in stock at this given store, the normal decision would be to push more from the warehouse. But you also have to take into account the fact that maybe some customers will be returning a part of the units they have currently. And so actually, at the end of the day, you have to balance those two in and out flows.
Fabian Hoehner: Okay, so influencing the demand forecast and those projections?
Elliot Langella: Demand forecast and decision making on a daily basis.
Fabian Hoehner: Okay, well, talking about that, the decision making, we were at an AI conference. So tell me, what are you even doing? I mean, now it’s AI, everything’s completely automated Yeah, so everything’s automated and you’re not doing anything anymore. Why are we still paying you?
Elliot Langella: No, no, no. So, first thing is, you know, AI is super famous, you know, it’s buzzing these days. And people usually associate AI as the buzzword to…
Fabian Hoehner: So are you telling me that we are not doing AI or what are you saying here?
Elliot Langella: Give me a few seconds. So these days, people are mostly associating AI with LLMs, so ChatGPT, chatbots, agents basically, or robots or whatever. What we’re doing at Lokad is partly that, but that’s not the core product that we built for Tokic. It’s rather AI in the decision making and in the demand forecasting. So it’s happening under the hood, in the back end of the software, and then it gets reflected in the interfaces that end users consume on a daily basis.
So important blocks, I would say, would be for forecasting demand. So we do use probabilistic algorithms to do so, and you know, we rely on what we call differential programming that is some kind of variant of deep learning to do that.
Fabian Hoehner: Yeah, well, who is interested in that can go back to Johan’s explanations on that, but…
Elliot Langella: We’ve got a few hours of content on the topic.
Fabian Hoehner: Okay, where else would you see applications?
Elliot Langella: We also use LLMs and, you know, classic clustering algorithms for classification of parts, time series, and categorization of the products. So it’s notably important for forecast demand, but also to, let’s say, bias in the good numerical way the suggestions of purchase and dispatch that we do on a daily basis at Tokic.
Because in the Lokad way of thinking, there’s not more exactly ABC categorization. It’s more a competition between all of the different SKUs for common resources that are either the capacity at the warehouse or the space that you have available at the store. And well, you have to balance out that maybe a big body part is creating a lot of margin, but it sucks up the space that could have been used by, you know, a lot of smaller parts that would be also fulfilling customer demand.
Fabian Hoehner: Okay, so bottom line, a large variety of, well, different kinds of AIs, and it’s your job as a supply chain scientist to choose the appropriate tooling to respond to whatever question there is, be it data cleaning or the decision-making process?
Elliot Langella: Exactly. To make it a production-grade process, the Tokic team is using our dispatch recommendation on a daily basis every weekday morning. So it needs to be ready at 6:00 a.m. when the warehouse opens. So we can’t just be toying around with fancy algorithms. It has to be, you know, production-grade, and it has to deliver consistently every day good results for the business.
Fabian Hoehner: Okay, what would happen if you wouldn’t deliver? If they don’t get the data at 6:00 in the morning?
Elliot Langella: So we do have some backup logics to make sure that there’s always something they could rely on, or we can just speed up some processing to just approximate a few things so that, you know, an hour, an hour and a half later, they’ve got something to rely on and, you know, they can go through the rest of the day.
Fabian Hoehner: Okay, but you need to make sure because otherwise there’s, well, a few trucks that are not leaving.
Elliot Langella: Yeah, or not delivering. Like, there’s a lot of, you know, real life in supply chain is pressing. It’s happening every day, so you got to be delivering.
Fabian Hoehner: Okay, well, now from the real life, let’s go to the macro level. I assume it’s working, otherwise we wouldn’t be standing here. What are some of the results that you can talk about?
Elliot Langella: So I would say the first important thing is transparency and comfort of use on a daily basis. The Tokic team moved from pulling data from databases, crunching in Excel spreadsheets, doing some, you know, mundane computation, and not much time left for real high-level analysis. They are logging in Lokad in the morning. Most of the recommendations we provide them are like 100% good to use, and there’s just a few tweaks here and there that they should be working with because they have additional knowledge from their supplier.
That supplier is going to be late, or, you know, they need to fulfill some purchasing condition. You know, the real-life things that happen that can be reflected in the data that we used, but that they know about because they’re either experts or, you know, firefighting in the operational daily reality. So this is much more comfortable for them to operate now with Lokad.
Fabian Hoehner: So transparency, automation, time beyond firefighting mode.
Elliot Langella: And focusing on more strategical tasks, you know, like they’re more at the definition of what they want to do, getting the good information from the market, from the suppliers, unless at the execution, operational execution, that’s delegated to the AI.
Fabian Hoehner: Okay, and well, obviously, we are numbers people. Do you have any numbers that we can share, anything concrete?
Elliot Langella: Okay, so I would say there’s going to be two angles. There’s really like operational execution about how the warehouse was able to shift from, let’s say, the COVID assortment to the new one that Tokic group is now relying on.
What we help Tokic do is basically shift about 40% of their assortments, like opening new stores, replacing older products by new ones, and that was done in collaboration with them to not overload the warehouse.
Meaning in terms of capacity, they just didn’t have like 40% more people to take care of these big changes. So what we did is streamline the rhythm at which we would be sending stock to the stores so that we give a bit more room to breathe for a warehouse to execute.
I would say the second angle would be more the usual supply chain metrics, you know, the ones that supply chain directors and COOs follow closely on a weekly basis.
I’m talking about service level. One key element was that after we went live for the pilot phase for this replenishment from warehouse to stores, we were able to measure, and we’re still measuring these days, that the service level of the top mover increased 5% at all of the stores.
So it means you’re literally getting more out of the same amount of stock because you’re timing it better or you’re investing relatively the good balance of stock at your warehouse so that you can spread it across the network, and this leads to generating more sales.
So at the end of the day, it was like a very big period of growth for Tokic. So it’s based on estimations, but the idea is that we beat by 28% the most optimistic stock value simulations that the Tokic team was doing.
So it means that we ended up with less stock than expected, and this actually helped generate about 10% more sales.
Fabian Hoehner: How do we get there? So what are the micro things that we’re doing differently in order to get to this point?
Elliot Langella: Two main ingredients: probabilistic inventory assessment and ROI estimate.
Fabian Hoehner: Okay, I’m going to need examples here.
Elliot Langella: Yeah, so probabilistic inventory is about sketching out all of the possible demand scenarios with probabilities. So let’s say it’s not one single number per day, like you’re going to sell two tomorrow. It’s going to be between one and three units for that given SKU at that given store, but with some probability weights.
Based on that, you know, when as a supply chain director you want to have this 95% service level at your store, it’s actually possible for Lokad to exactly estimate how much stock you need to get to this 95% because we have the probability weights and we can compound them up until 95%.
Fabian Hoehner: So would it be fair to say a more accurate representation of the future?
Elliot Langella: More accurate representation, and I would say more informed for the second layer that I refer to, that is this ROI estimation. Now that you have different scenarios of sales for which you know you have 10%, 5%, 0.01% of chances of selling, then you can also guess what will be the payback of taking this decision to send these additional units to one given store.
If this unit has 1% chances of selling, it means it’s not likely going to generate a lot of margin back for the company. So you must be paying inventory cost, logistic cost to take that decision, and it’s not going to return much.
The other way around, if you see that problem from the warehouse and you have a limited pile of stock standing at the warehouse, you obviously don’t want to send this additional one unit to the store that’s going to sell for 1% chances.
You rather want to put it to another store, maybe a bit further in the network, that will have 10% or 20% chances of selling that unit. So it’s also about arbitration in terms of scarcity, and more generally speaking, it’s a good framework for the Tokic team to simulate and try to assess how much more stock they would need, let’s say in the long term, on an entire year based on this ROI assessment.
Fabian Hoehner: So where it does make sense for me is the probability of sales that you put your stock where it has the highest probability of selling. But are there other economic components going into that when you say ROI driven?
Elliot Langella: So there’s definitely the margin that you’re going to generate. There’s also the inventory cost, so logistic cost, how much does it cost you to pick and pack at the warehouse, put it on the truck, get the truck driving, store it at the store.
Fabian Hoehner: So bottom line, something smart that is representing reality where you’re not sending one part to a store that is, I don’t know, half a day away, that you’re going to bundle it up and only then that it makes sense. So that’s what you can consider with this ROI driven logic.
Elliot Langella: We also include financials. I mean, it’s a big conversation these days, you know, with inflation that has been growing up and now that’s maybe stabilizing, who knows. Capital cost is important. Also, there’s cost of opportunities, you know, if you’re investing in that inventory from that supplier, well, maybe you don’t have any more budgets to go ahead and search for additional deals with other suppliers and extend your catalog next year.
So it’s about arbitrating the best way to use your capital. And I would say that there’s a last component to this ROI estimation that’s interesting to discuss because we were discussing about the return sales earlier. It’s this what we call stock out penalty.
It’s more some kind of karma point type of approach. If you’re out of stock at a store, you’re losing your customer trust and they’re starting to think that you’re not a one-stop shop anymore. So there’s value besides the financials, besides the margin, to just be in stock at the right moment at the right place because this is going to create repeat purchases from your customers.
And also because one product can attract the sales of other ones. We don’t want to be losing sales on your bread and butter, the thing that people come to your shop and know they will find because if they come and find it, maybe they’ll buy something else on the side, oils, wipers, things like that, that you’re not really sure you need, but you just take them in case.
Fabian Hoehner: Okay, and all of that goes within the same formula, into the same algorithm to take a decision at the end of the day?
Elliot Langella: Yeah.
Fabian Hoehner: Okay, so what does that leave the managers with? What are their decisions that they are taking?
Elliot Langella: Strategical assessment. So basically delegating the mundane in computation to Lokad’s AI and, you know, just focusing on what would be the decision of going with this scenario. Is it something we can do as a company? Is it the direction we want to take?
Can we afford this additional intake of stock? Is it going to generate additional value for the customers? Is it what we want to focus on as a company, or do we want to focus on assortment, focus on other areas of the business?
Fabian Hoehner: So less mundane stuff and more, well, the strategic, impactful.
Okay, yeah, well, that’s pretty clear on where we are today in the project. Where are we going? Is there, well, have we done everything, or is there something left to do?
Elliot Langella: So early 2025, we will be finishing to roll out the projects from Croatian supply chain to Croatian and Slovenian supply chain compounded together. So Tokic has been growing on the Slovenian market, and so what now they’re eyeing towards is an in-between state where they will be purchasing from different entry points, so two different warehouses that made them cross fluxes to send back stores from one country to the other.
This is obviously making things a bit more complex from a supply chain standpoint, but it’s easier to get there when you’re already working with Lokad on the, let’s say, the first parameter that has been running well for several years.
I would say the other challenges also in general that we can help our customers with is everything related to managing the complexity that creates this compatibility of parts. I mean, supply chains in automotive supply chain, this is key. Sometimes to enter one customer need, you have 10 different parts. Which one should be part of your assortment at this given store, at the other stores? This is big challenges.
Other things is pricing, always critical. There’s even for Tokic high competition online. Now you can buy parts from e-commerce, so this should be part of your picture as a retailer to know how to position yourself given the value proposal that you have with these reverse sales that we talked about with the highest availability that we can provide to Tokic at their stores.
This is obviously easier for car people and repair shops to come and buy from Tokic, but you know, there’s always a balance between the price they’re willing to pay and the convenience that this is providing us.
Fabian Hoehner: So the goal is to get more and more into the same numeric decision-making process. Well, that sounds like you have quite some work ahead of you. So, well, let’s catch up in 2025 and see where things have been going. Thanks for watching, and we’ll see you next time.
