00:00:07 Warehouse management and the concept of smoothing.
00:01:20 Challenges in warehouse management, including manpower costs and equipment utilization.
00:03:53 Achieving daily warehouse operation smoothing through network level optimization.
00:05:22 Economies of scale and the impact of diverging from optimal warehouse regime.
00:07:00 The limitations of classical supply chain planning and the potential for warehouse operation optimization.
00:08:00 Traditional SKU time series analysis and its limitations.
00:11:00 Synchronization and oscillation issues in supply chains.
00:13:35 Rethinking the paradigm to solve the problem.
00:14:25 The quantitative supply chain perspective.
00:15:22 Decision-making using the most profitable unit across the network.
00:17:42 Comparing SKU time series perspective and the prioritized list of actions.
00:19:14 Truncating the list to achieve optimal warehouse flow.
00:20:25 The economic value and diminishing returns of prioritized units.
00:23:58 The importance of smoothing in supply chains and its challenges.
00:25:26 Reframing the problem and changing perspective for better solutions.
00:26:24 Difficulties in finding a perfect time series forecast.
00:27:33 Conclusion and reflections on classical approaches.
In the interview, Kieran Chandler and Joannes Vermorel discuss the challenges of managing modern-day warehouses. Vermorel emphasizes the importance of smoothing the day-to-day operations of the warehouse, which means having the same amount of manpower from one day to the next. He also discusses the challenges of synchronization in supply chain networks and the need to prioritize decisions based on economic returns. Vermorel argues that classical approaches to supply chain optimization are flawed and that changing one’s perspective on the problem is necessary to find effective solutions. The interview ends with Chandler thanking Vermorel for his time.
In this interview, Kieran Chandler discusses with Joannes Vermorel, the founder of Lokad, a software company that specializes in supply chain optimization, the challenges of managing modern-day warehouses. Despite the rise in automation, warehouses still heavily rely on manpower, particularly during peak seasons. Vermorel explains that modern warehouses are designed for optimal regimes, just like the engine of a car. The problem is that the market demands may not be aligned with the peak productivity status of the warehouse, and diverging from the optimal regime can saturate the equipment or not use it to its full potential. Most warehouses are automated to varying degrees, but staffing is still a major challenge. Contractors are needed during peak seasons, but they have less training and lower productivity, and regular employees must still be paid regardless of the workload. Vermorel emphasizes the importance of smoothing the day-to-day operations of the warehouse, which means having the same amount of manpower from one day to the next. He distinguishes this from streamlining, which involves investing in new equipment to improve overall productivity. Vermorel suggests that a steady stream of operations with the same workload every day would be ideal to avoid having to accommodate source variations at all.
Vermorel explained that achieving optimal warehouse operations on a daily basis requires looking beyond the warehouse itself and optimizing the entire supply chain. This can be accomplished through network-level optimization. Chandler then asked Vermorel about the challenges of increasing manpower and the associated costs. Vermorel explained that the cost of additional personnel does not increase linearly because flexibility comes at a premium. Warehouses can achieve economies of scale by being relatively inflexible. Classical approaches to supply chain optimization rely on forecasting the workload of warehouses and adjusting staffing levels accordingly. Vermorel argued that this approach is flawed because it does not provide any degree of control over warehouse operations. He believes that the classical approach emphasizes skews and time series, which are simply mobilization perspectives on the warehouse. Vermorel emphasized the importance of looking beyond the warehouse and optimizing the entire supply chain to achieve daily operational optimization.
Vermorel discussed the challenges that arise with synchronization in supply chain networks. He explained that synchronization can create oscillations in the supply chain, leading to inefficient flow patterns. He used the example of a product that experiences a stock-out from a supplier, causing all stores to gradually run out of stock. Once the supplier is back in stock, there is a wave of products that flows across the supply chain to replenish all the stores at once. This leads to residual oscillation, which is problematic for supply chain management. Vermorel suggested rejecting the paradigm that created the problem in the first place and looking at the problem from a different angle to find a solution. He emphasized the importance of smoothing out promotions and pre-warning warehouses to have stock ready. Vermorel’s unique perspective on supply chain optimization highlights the need to think critically about the issues at hand in order to find effective solutions.
Vermorel discussed his company’s approach to supply chain optimization, which focuses on using a quantitative perspective to prioritize possible futures and decisions based on economic returns. This approach does not use SKUs or time series, but rather individual units of products, each with unique economic returns. The goal is to find the most profitable unit and then investigate which unit would be the second most profitable, and so on down the list. Vermorel explained that while the number of potential decisions may seem infinite, in reality, it is finite and limited by the number of units in the warehouse. Vermorel claimed that by prioritizing the list to the optimal regime of the warehouse, the solution for supply chain optimization can be obtained for free. This approach was contrasted with the traditional SKU plus time series perspective, which did not have the same options available to the warehouse manager. Vermorel argued that with modern computing power, the process of optimization is feasible, although not necessarily easy.
Vermorel explains that many warehouses have a prioritized list of actions, but they lack granularity in decision-making. The classical supply chain pattern only provides replenishment orders, leaving it up to the warehouse to decide which order to prioritize. In contrast, Lokad’s prioritized list allows every unit to have an economic reward, which helps warehouses make better decisions. Vermorel explains that the first units that need to be expedited are the most critical because they prevent highly probable stockouts. However, as the warehouse sends more units down the list, the economic rewards diminish rapidly. Vermorel notes that in a regular supply chain network, the first 10,000 units sent by the warehouse can have an economic reward 100 times greater than the last 10,000 units. Vermorel suggests that by analyzing the supply chain network as a whole, warehouses can determine how much workforce to apply on a daily basis and prioritize their actions more effectively.
Vermorel argued that balancing the equilibrium of a network is crucial for a supply chain director. Smoothing, the process of removing irregularities in data to obtain a clearer trend, is essential for ensuring that supply chains run smoothly. Vermorel emphasized that smoothing has always been in the background for decades but lacked a solution from the classical perspective. He suggested that reframing the problem and taking a quantity supply chain manifesto perspective can help to prioritize decisions according to economic rewards, leading to a solution for free. Vermorel concluded that changing the padding can result in a solution that is relatively easy to implement.
Vermorel explains that it’s easy to identify supply chain problems, but changing the way you approach the problem is difficult because it requires reimagining the entire supply chain network from a different perspective. It took him and his team several years to find a solution to the problem. Initially, they believed that a better time series forecast was all they needed to solve the problem, but they soon realized that a perfect forecast doesn’t exist. However, changing your perspective on the problem eliminates the need for a perfect forecast. Vermorel suggests that many people are burying their heads in the sand and pretending these problems don’t exist by using classical approaches. The interview ends with Chandler thanking Vermorel for his time and signing off.
Kieran Chandler: Today we’re going to discuss how warehouses can prioritize decisions, and in particular, the concept of smoothing. So, Joannes, what is it about the characteristics of modern-day warehouses that make them quite so difficult to manage?
Joannes Vermorel: A modern warehouse is pretty much designed for an optimal regime, just like the engine of your car. You get economies of scale because you have your regular employees, and if you need to run beyond peak productivity, you end up needing contractors. You usually need to pay them more by the hour, and they happen to have less training, resulting in lower productivity. On the other side of the equation, if you have fewer people, you have full-time employees that you pay no matter what. So, even if you decide to go with a slower regime, you still end up paying for all those hours at the end of the month or quarter.
And that’s just looking at the staffing element. The reality is that nowadays, most warehouses are automated to some varying degrees with conveyor belts, robot pickers, and plenty of other things. Those investments are typically calibrated for certain flow of goods. If you diverge from that, either you saturate your equipment, or you don’t use it to its full potential.
Bottom line is that you have an optimal regime or sometimes configure your warehouse to have maybe two optimal regimes, one for the low season and one for the peak season. But you cannot have something that operates optimally at any regime. You have to pick certain conditions, and the problem is that what the market demands may not be aligned with the peak productivity status of your warehouse.
Kieran Chandler: Okay, and when we first discussed today’s topic, smoothing warehouse operations, you were very particular about the word “smoothing.” Why is this word quite so important compared to something perhaps like “streamlining”?
Joannes Vermorel: I’m referring to the idea that the day-to-day operations of the warehouse can be smooth, meaning the amount of manpower that you need is pretty much the same from one day to the next. I’m not referring to streamlining, which would be investing in a new conveyor belt to improve the overall productivity of your warehouse. Once that investment is done, you’re back to the initial situation, which is you want to smooth your daily operations from one day to the next. This is because variations require more or less staff, and it’s very difficult to accommodate. Typically, because it’s hard to accommodate, the best way would be not to have to accommodate those variations at all. So you can really have just a steady stream of operations with a consistent workload every day.
Kieran Chandler: In a single day, the question is, how do you achieve that? How do you achieve daily operations optimization?
Joannes Vermorel: If you’re stuck by just looking at what is happening within the warehouse, you cannot. But as soon as you start looking at the supply chain as well, then it becomes possible by doing network-level optimization to smooth the daily operation of any single warehouse.
Kieran Chandler: Let’s look at some of the challenges that are involved. You mentioned a little bit about manpower, and it’s interesting that as manpower increases, you’re using more staff and the costs don’t increase linearly. Why is that?
Joannes Vermorel: The cost doesn’t increase linearly because contractors cost more and they have less training. If you had your usual team of 50 people and you need to bring in 20 extra people, maybe those 20 extra people will cost you as much as your original 50 people. Contractors charge more by the hour for their flexibility. So, the adjustability of the market is that flexibility is an option, but it’s an expensive option. By being relatively inflexible, you can lower your costs. In a warehouse, you end up with economies and diseconomies of scale whenever you diverge from your optimal regime.
Kieran Chandler: Let’s talk about the current approach. How well do classical approaches work when it comes to working in this ideal regime?
Joannes Vermorel: It’s very interesting. From a classical supply chain perspective, this problem doesn’t even exist. When you look at the theories, it’s puzzling. I’m referring to the belief that there is something called a SKU and a time series. I believe that neither SKUs nor time series are really real; they are just a mobilization perspective on your warehouse. The classical supply chain planning emphasizes SKUs and time series. It turns out that when you take this perspective, the workload of your warehouse is not something you have any control over. You end up with people trying to forecast the workload of warehouses so that they can adjust the staff. I would argue that this is a completely wrong viewpoint on the problem. It’s interesting because, from the classic perspective, the idea of optimizing warehouse operations and smoothing them is not available as an option because there is nothing in the planning that gives you any degree of control over that.
Kieran Chandler: So, it’s very interesting because any warehouse managers I’ve met realize that they have these massive economies of scale when the warehouse diverged from the regime. So obviously smoothing the flows is of prime interest. I mean, it’s like an obvious move. But when people step back and say, how are we going to do that, they realize they don’t have any levers. There is nothing in the organization, the software, etcetera, that gives them any control. And the problem is that at the core, it’s a particular way in which there is no control is not even possible.
Joannes Vermorel: Okay, but surely that kind of warehouse manager is just reacting to demand that’s happening upstream. And so, how much control can they actually have because they’re just reacting to what ultimately customers are buying?
Kieran Chandler: But are they?
Joannes Vermorel: And usually, you see that’s the sort of things where people jump to the conclusion, oh yes, I mean, we have a diamond demand, whatever system, um, yeah, it’s very rational. It’s driven by sales and says history and whatnot. But is it? You see, because again, let’s go back to the padding. The padding is you have SKUs and time series. Where is the network? The network does not exist in your padding. And I would say I would argue that usually the networks, the supply chain network, doesn’t exist as such in any pieces of the software that is actually running the supply chain. And it may sound puzzling because people say, but we have all the transactions. We have all the stock movements. So, certainly, the network exists. I would say no, it doesn’t. And actually, when you look at how, I would say, the vast, vast majority of um of supply chain software are implemented.
Kieran Chandler: What does that mean?
Joannes Vermorel: That you have a SKU. It means that basically, you have replenishment going on for the stores. But replenishments are done according to some kind of, I would say, glorified min-max policy. So min-max, what is a min-max policy just mean that you have a trigger condition that’s going to be your mean stock level. So basically, when you reach a certain level of stock, you trigger replenishment. That’s your, so basically a min-max is basically first you have a condition to meet to met and then once you are meeting this condition, you’re going to trigger a certain quantity to be replenished. This quantity can be just one, you know, kanban style. Um, you sell one, you replenish one, but sometimes because there are like lot multipliers because you have packs because you have like economic order quantity because you have many other constraints, you will shoot for a slightly bigger quantity. And thus from the warehousing perspective, you end up with every single day with tons of downward SKUs that end up meeting their trigger condition. You know that mean, and thus you end up with all those cues that say, send me more stuff, and thus you end up with a list of things that you need to be that needs to be done on the daily basis by the warehouse itself. But when you look at that, the warehouse has zero control on any of those figures.
Kieran Chandler: Okay, so what you’re sort of saying is taking that kind of SKU time series approach, you’re very much looking at things in isolation, and you’re not taking into account kind of the rest of the network and how that kind of interacts. So what Would you kind of smooth out those activities and reduce those kind of spikes?
Joannes Vermorel: Before jumping to that, you need to further analyze what is happening with skews, because otherwise it’s very hard to make sense of the solution. The first thing that happens when you have skew plus time series analysis is that you end up with plenty of synchronization within your supply chain network. You would think, “Oh, it’s very nice, I have synchronized events in my supply chain,” but no, it’s absolutely not. In terms of smoothing the flows, that’s pretty much your worst-case scenario.
Kieran Chandler: What would be an example of one of those synchronized events?
Joannes Vermorel: Let’s assume that, for example, you have a warehouse serving stores and you have a stockout from your supplier. So, one of your suppliers has a stockout, and you cannot serve a given product to the stores. The stores gradually run out of stock. Obviously, this is a bit random; some stores had more or less stock initially when the supplier went out of stock. But let’s say after a few days or a few weeks, we can reasonably assume that all the stores are out of stock, and the warehouse is out of stock as well for the product.
Then the supplier is back in stock. So what is happening is that at the minute your supplier is back in stock, you will have a big wave of products that will flow across your supply chain to basically replenish all the stores at once. And because probably your stores are similarly designed, they have plenty of similarities; it means that initially, you might have found that all the stock levels were randomly distributed across your stores. But the stockout has just synchronized the replenishment pattern for the given product across your network.
So you see, all the stores went out of stock at the same time, and all the stores are getting back in stock at the same time. And if you look at the time series of flow for the product, you will see a drop to zero, a big spike, and then chances are that you have very strong residual oscillation, like echoes of your stockout. And actually, you have many elements that provoke such oscillation. Stockouts are one, promotions are another, and you can have calendar effects. It doesn’t matter if it’s a retail network or a multi-echelon manufacturing supply chain; you have the same sort of phenomenon. So basically, you have plenty of phenomena that create synchronization and oscillation in your supply chain.
Kieran Chandler: Okay, but things like promotions are always going to exist. So, how can you go about smoothing them out? Is it more a case of the warehouses being pre-warned and having stock ready when these promotions come?
Joannes Vermorel: The thing is that if you want to solve the problem, first you have to reject the paradigm that created this problem in the first place. That’s probably the most intriguing perspective. There are frankly problems where the only way to tackle the problem is first to start looking at the problem from another angle. And sometimes, something exceedingly puzzling happens when you start looking at the problem from a different perspective.
Kieran Chandler: So, if we approach the problem from another angle, we might find a solution that is quite literally free. It’s very puzzling, but it does happen occasionally. Let’s first try to see how we can look at the problem from a different perspective and then investigate if we can get the solution for free by any chance.
Joannes Vermorel: The different angle is basically something that Lokad has outlined in its quantitative supply chain manifesto years ago. It’s all about possible futures or decisions, all of them prioritized by economic returns. So, if we start looking at the problem from this quantitative supply chain perspective, we don’t have SKUs anymore, nor do we have time series. What we have is units of products, each one of them being unique, like a snowflake. We don’t try to pack all those things into SKUs. The question we ask ourselves is, given all those possible futures where the market will ask for different things, which unit should the warehouse decide to ship downstream to one location? We can’t know them because there is irreducible uncertainty. We can look at the unit that is maximizing the economic returns. So, there will be one unit that is the most profitable across the entire network. If we had only one unit to send downstream, that would be it. Conceptually, once I’ve decided that my number one unit would be this one unit, I can investigate which would be my number two unit, the second most profitable one. We can investigate all locations, all units, all the economic returns, and figure out what is the second most profitable unit. We can walk down this list of decisions.
Kieran Chandler: Okay, so you have an infinite list of decisions, and they’re all prioritized. How do you decide where to end the list?
Joannes Vermorel: First, you will get to the bottom because, at some point, your warehouse will be empty. You have just a finite number of units sitting in your warehouse, so the number of decisions is potentially conceptually infinite, but the reality is that it’s very much finite. It’s limited by the sheer number of units in your warehouse. You might think, “Oh, but my warehouse has millions of units.” Yes, but computers are powerful nowadays, and even if you have to enumerate millions of units, it’s no big deal. A single core on a computer does billions of operations per second. So, I’m not saying that it’s easy, but it’s certainly very feasible. Going back to the problem, you can truncate this prioritized list to the optimal regime of the warehouse. If you do that, you get the solution for free. You have a warehouse where you’ve just decided that what you would be sending today would be exactly matching your optimal regime in terms of
Kieran Chandler: So Joannes, can you tell us about the challenges that warehouses face when it comes to supply chain optimization?
Joannes Vermorel: Yes, of course. So when we were looking at the problem from the SKU plus time series perspective, the options did not exist. The warehouse manager had no prioritized list of action. It was just a list of replenishment. The only option for the warehouse was just to ignore those quantities and maybe ignore some stores. So not serving certain stores or products, but they didn’t have anything granular to decide exactly what to do. They just had replenishment orders that say ship 50 units towards this location and then you have another order that says ship 25 units to this other location. The classical supply chain planning doesn’t say which order is more important. It’s not even a question being asked. Thus, because it’s not a question, there is no answer. And if you look from the Lokad way of looking at the problem, which is different, particularly where you have this prioritized list where every single unit gets its economic reward, suddenly, if you want to smooth the flow of the wells, it’s a given because it’s just a matter of truncating the list of what you want to expedite to the optimal regime of the wells.
Kieran Chandler: That makes sense. How are they basically deciding how much workforce to apply on a daily basis?
Joannes Vermorel: So here we have to go back to the supply chain network as a whole. The interesting analysis that we have done many times with those clients is that when you prioritize your units to be expedited, you know where to look from. When you look at the first unit, second unit per unit extra, you will see that you have a strong glow of diminishing returns. So that means that the first units that you need to send are typically have enormous economic value. Why? Because basically, you’re preventing highly probable stockouts. So, you know, it’s the same sort of pattern whether it’s a multi-edge loan supply chain in FMCG or a retail network. Basically, the first unit that you send is critical because otherwise, downstream clients or a manufacturing unit or an assembling unit is going to face a stockout. So those have very significant impact. But as you walk through the list, very rapidly, you’re just increasing your safety net. Every single unit that you send down the list, I mean, it’s you’re actually gaining. Your economic rewards diminish very rapidly. That means that and what is interesting is that typically, for a regular supply chain network, there is no such thing as a regular supply chain network, but let’s say typically what you the situation that you end up with is that let’s say your supply chain network, there are 100,000 units flowing through the network every single day. So that means that let’s say your house somehow ends up expediting on average, 100,000 units every single day. Well, if you look at the first.
Kieran Chandler: So, we’ve observed that there are 10,000 units being sent by the warehouses. What you will typically find is that the economic reward associated with those first 10,000 units out of 100,000 is literally 100 times more than the last 10,000 units. It’s not just slowly diminishing returns, but super rapidly diminishing returns, where the top 10 percent of what you send is worth 100 times more than the last 10. So, does it really make a difference if, on any given day, you decide that you’re going to expedite 90,000 units or 110,000 units compared to just expediting 100,000 units?
Joannes Vermorel: The answer is it almost makes no difference whatsoever, and that’s very puzzling because practitioners from the classical perspective might think it’s a big problem when the warehouse doesn’t have the capacity. For example, if the warehouse needs to expedite 100,000 units today and they can only expedite 90,000, they would say it’s a big issue. But why is it a big problem? The answer is because the 10,000 units that they cannot expedite are not prioritized. Potentially, out of those 10,000 units that they cannot expedite, there might be things that are absolutely urgent and critical, and because there is no prioritization taking place, nobody knows. Usually, just because they do it randomly out of necessity, odds are that 10 percent of the 10,000 units that you cannot expedite today are part of the 10 most critical things that you should be sending. But because you can’t prioritize, you end up with a big supply chain incident.
Kieran Chandler: So, when you see those diminishing returns, you only need to roughly balance the equilibrium of your network. As we start wrapping things up, for a supply chain director who’s probably watching this, why is smoothing so important? Is it something that they could easily implement or change from that classic little skew time series approach, or is it something that’s actually quite challenging and will take a long period of time to adjust to?
Joannes Vermorel: Smoothing, I would say, has always been in the background for decades. I’ve discussed it with many supply chain directors, and it’s an obvious move that is of interest because of the diseconomies of scale that occur as soon as you diverge from your optimal regime and flows.
Kieran Chandler: Obviously, from the classical perspective, this problem doesn’t have any solution. Is it possible from the classical perspective?
Joannes Vermorel: It is not. It doesn’t matter whether your technology is super crude like moving averages or super advanced like deep learning because from your perspective, this problem doesn’t have a solution. First, you need to kind of reframe the problem. The interesting thing is, when you take the problem from this “Quantitative Supply Chain Manifesto” perspective, with all possible futures and all possible decisions prioritized according to economic rewards, you get the solution for free. If you’re willing to change your perspective, it becomes something very easy. But changing the way you look at the problem is actually quite difficult. You need to reimagine the very same supply chain networks that you’ve seen your entire life and look at it from a completely different perspective. By the way, at Lokad, it took me and the teams several years to get to the point where we had found a way to look at the problem from a completely different angle, where the solution was possible. It’s not obvious, and there is nothing that says your perspective on this problem is wrong. For years, the initial years of Lokad, we were thinking we just needed a better time series forecast, and maybe if we had a better time series forecast, everything would fall into place. For example, if we had the perfect time series forecast, remember what I said about the warehouse manager who wanted a perfect demand forecast for their warehouse? If you’re thinking in terms of time series, you’re thinking, “I just need a perfect forecast, and that will solve my problem.” Yes, but the problem is this perfect forecast does not exist. Interestingly, if you change your perspective on the problem, the very need for a perfect forecast doesn’t exist anymore either.
Kieran Chandler: Okay, we’ll have to wrap it up there. It certainly sounds like maybe, with some of these classical approaches, many people are kind of burying their heads in the sand and pretending these problems don’t exist at all. So 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.