00:00:00 Introduction: hidden costs of service levels
00:04:15 Definition: service level as stockout probability
00:08:30 Service level is a low-resolution proxy
00:12:30 Challenging 0% and 100% extremes mathematically
00:16:00 Aerospace: flight-hours per dollar
00:20:00 High service rates produce exploded AOG risk
00:24:30 Reality first; math should formalize reality
00:28:00 Customer expectations anchored to dates and price
00:32:00 Service level terminology and measurement issues
00:36:30 KPI: maximize rate of return
00:40:30 Remove service-level obsession; be subtractive
00:44:00 Fill rate is an inadequate substitute
00:48:30 Retail: customers expect matching bulk availability
00:52:30 Configurability: customers care about few options
00:57:30 Maximizing service level can hurt seasonal inventory
01:02:00 Final takeaway: hidden costs are massive opportunity loss
Summary
Supply-chain “service level” is a misleading, narrow probability (no stockout in a cycle) that ignores profits, customers, and real constraints. It encourages unrealistic targets—99% across many parts can still fail—and hides trade-offs. Models assume stationary demand and SKU-level metrics, while customers buy outcomes (dates, coherent assortments, configured goods). Practitioners often ignore targets because literal compliance is ruinous. Instead of ritual metrics, measure rate of return: allocate capital by economic trade-offs and let financial reality, not probabilistic dogma, drive decisions.
Extended Summary
Much of what passes for “science” in supply chains is a flight from reality into vocabulary. “Service level” sounds reassuring, like “quality of service.” In practice it is a narrow probability—no stockout during a replenishment cycle—treated as a compass for decisions. It is a compass that points north only by accident. The hidden cost is not merely theoretical; it is the massive opportunity cost of steering by a metric that is indifferent to profits, indifferent to customers, and indifferent to the real-world structure of constraints.
There are no solutions; there are only trade-offs. Yet service levels invite the unconstrained vision: pick 99%—or 99.9%—and imagine the world will comply. In aerospace, one maintenance event may need a thousand specific parts. If each is at 99% “service,” the chance all are available today is closer to a coin flip than to certainty. Meanwhile, the CFO, warehouse space, and lead times still exist. Costs explode long before service levels meet the fantasy. At the other extreme, 0% and 100% can both be rational in reality: 0% for phase-out SKUs, 100% (functionally) for one-dollar tape that can ground a $100 million aircraft. What looks nonsensical in the model often makes perfect sense in the economics.
Much of the confusion comes from starting with formalism and then bending reality to fit. The math assumes stationary demand—no beginning, no end—when nearly every product is born and dies. It quantifies a SKU-level probability while customers buy outcomes, not probabilities. Amazon sets a date; the customer accepts or rejects the offer. MROs use standard exchanges and substitutes; fashion retailers face seasons, not steady states. In a DIY store, a shopper wants six identical light switches today; the “service” is coherence across units, not a percentage in a spreadsheet. With configurable goods—bikes or workstations—customers specify a handful of must-haves and outsource the rest to sensible defaults that are in stock. That’s how expectations are structured in the real world.
Practitioners know this. They quietly ignore their own service-level targets because following them literally would be ruinous. The metric is useful in classrooms because it is easy to grade. In business, good intentions measured badly invite bad decisions done earnestly.
If you want one number, use rate of return. Supply chains convert dollars into atoms and atoms back into dollars. The scoreboard is how much more you get back per dollar invested—fewer AOGs per dollar, faster cycles per dollar, better assortments per dollar. Start from customers’ and suppliers’ expectations, model the trade-offs, and allocate capital accordingly. The “replacement” for service levels is not another talisman; it is subtraction. Remove the illusion. Let economic reality, not a probabilistic ritual, govern the decisions.
Full Transcript
Conor Doherty: This is Supply Chain Breakdown and today we will be breaking down the hidden cost of service levels. Now, it really doesn’t matter what vertical you’re operating in. If you are making decisions based on service levels, well, this is a conversation for you. Who am I? Conor, Communications Director at Lokad. Who is to my left? You know that Lokad’s founder, Joannes Vermorel.
Now, before we get started, let us know down below what vertical are you most interested to hear about today. Could be retail, could be manufacturing, could be aerospace, whatever it is, let us know. We’ll try to tailor our discussion to that vertical. Also, get your questions into the live chat. We will answer those in about 20 minutes. And with that, Joannes, let’s get started. So, set the table. We’re talking about service levels. We’ll get into the hidden costs in a minute, but first definitions.
I think a definition you and I would agree on when we talk about service levels: “the probability of not hitting a stockout during a replenishment cycle.” Now, if you broadly agree with that, where exactly is the problem and where is the hidden cost of this?
Joannes Vermorel: In essence, this is a proxy. It’s not exactly doing anything that would kind of make sense business-wise. Not quite. And it is, I would say, the hidden cost is mostly that this proxy is framing the problem in ways that are deeply nonsensical.
Conor Doherty: So when you say proxy, you mean something that’s indicating but not perfectly.
Joannes Vermorel: Yeah. It’s, you know, it’s directionally kind of getting toward the idea of pleasing the customer but again it’s very non sequitur. You cannot jump from this probability definition to satisfaction of customer. There is no connection. It’s extremely elusive at best.
And thus that’s why I say it’s a really a super low resolution proxy. And who that occasionally gives you like a broken compass, you know, it occasionally gives you something that approximate the north, but most of the time it doesn’t even give you the north. So it’s, and that’s why you are, I would say, leaving so much money on the table when you are relying on service level. It’s, I would say, an extremely halfway broken compass that occasionally points out the correct direction, but fundamentally it’s so bad you need to have something better.
You see, and there are so much better alternatives. That’s why I say there is tons of hidden cost because it’s mostly opportunity cost, but at some point, you know, it is very real if your competitor, instead of having a broken compass, they have a compass that actually works, they will be able to journey toward the destination they want much more directly than you if you’re just, you know, fooling around just because your compass is like most of the time completely broken and occasionally a little bit right.
Conor Doherty: Okay. Well, just to be fair here, there are people who would say, “Well, look, the immediate challenge is—well, Joannes, if you’re kind of implying, maybe not explicitly, but implying that the concept of service levels is not really that economic a perspective.” And someone might say, “Well, look, I can easily just prove that. I can falsify that by saying, well, if I have 0% service level, I sell nothing. Therefore, I make no money.”
Joannes Vermorel: No. Again, you see, if you cherry-pick numbers like the extremes, we are back to the broken clock fallacy where yes, twice a day it’s going to display the correct time, and yet the clock is completely broken, doesn’t work. So here what you’re saying is that first let’s challenge that, you know, people say, “Oh look, it’s not that broken. 0%, you know, it tells me something,” but it isn’t extreme. It’s deliberately.
But is it even true? You know, we look at the extreme, is it even true? And the answer is no. If you have products that have like finite life—so they don’t, you see…
Conor Doherty: Yes, I mean, let’s assume…
Joannes Vermorel: I mean first, people don’t understand the power of mathematical assumption. When you say that you have a stationary time series, stationary demand, you say there is no beginning and no end. A billion years in the past, it was already there. A billion years in the future, it’s still there. That’s what stationary actually means from a mathematical perspective. This is an extravagant assumption. It means that the things that you’re looking at were supposed to be already there before civilization even emerged. And okay, this is crazy mathematics. And people don’t realize that mathematics let you do a lot of, I would say, extremely crazy assumptions.
So back—so here we are assuming that things have no end. It’s crazy. Every single product that you can market has an end on the market. There are a few, a tiny tiny tiny few products that have been sold as such for decades and will probably be sold as such for the decades to come. Yes. Channel 5, Champagne, a tiny few things, but really those things we can probably count them in the world. The amount of products is probably a few hundreds and that’s it. And in terms of economic mass, it’s like 0.01% of everything.
So those are like the extreme exceptions. And so back to 0%. Yeah. If your product is going to be phased out, 0% service level is what makes sense at some point, and there will be a transition where you want to bring that to zero because you are phasing out the thing. So clearly if you do not make an extravagant assumption, which is this SKU, product, whatever is going to be around till the end of time, then 0% service level becomes a valid proposition.
Conor Doherty: And conversely, what about 100%?
Joannes Vermorel: In effect, you can bring your inventory level to a position where numerically you cannot differentiate between the actual service level and 100%. For example, if your lead time is one week and you have 10 years’ worth of inventory, you are, you know, you are so overstocked that it’s not even—you’re not really computing. It’s not even in the range of things that you can really compute the probability of being out of stock.
Does it make sense to have that? Yes, it does. That’s exactly what companies in aviation are doing. For example, screws or tape or basic things that you need to maintain an aircraft because if you are about to do a maintenance operation for a 100 million dollar airplane, you don’t want just to be out of stock of duct tape that is worth like $1 per roll. It’s nonsense.
So obviously you will be massively overstocked on that and it’s just fine. It is just fine. Well, actually, interesting—in the chat, actually, some people have said that aerospace is an example to focus on, and actually my next question touches on this because of course the concept of service level and making decisions based on service level will vary considerably depending on what vertical you’re in.
Conor Doherty: So again, if you’re a supermarket and you have thousands of customers versus you’re an MRO and you have contractual penalties for missing a deadline, etc., service level varies.
Joannes Vermorel: Yes. And that’s a problem. That’s a problem. It’s because it is fundamentally a mathematical definition that doesn’t make sense business-wise. So every single time you—so again, math lets you define stuff that are completely made up. You know, it’s just the beauty of math. You can decide that you have two-dimensional numbers, you’re going to call it the complex plane, and it works. Four-dimensional, you have quaternions. It works. But then it’s not commutative anymore and it’s strange, and again it works.
Okay, in the realm of math, tons of things are just incredible and possible. It doesn’t mean that they are wise or relevant business-wise. So what I’m saying is that, you see, it’s an attitude. It’s an attitude. When I look at a supply chain, I will start by saying, okay, the atoms, the things that are happening, the expectations from the clients, the expectations from the suppliers, the expectations from the partners—all of that is my reality. The thing and the expectations, the base reality. The math is just a way to organize all this mess because, you know, all those expectations are kind of super messy, all those stuff that I have—physical units, products, raw materials, and whatnot.
All of that is such a mess. I’m using the math to organize all of that. And what I’m saying here is that you need to first look at reality and then decide what is the appropriate formalism. It goes in this direction. Reality first and then the formalism.
And here, the main problem that I have with service level is that we are trying to do the exact opposite. We start from a mathematical formalism that is really really not very good. You know, fundamentally it is more like something that is—I would say the only valid use, in my view, of service level is to create exercises for students. That’s it. That’s it. Yeah, if you want to make like semi-trivial mathematical puzzles where you will be able to grade students, I would say service levels are kind of okay for this purpose, but no more.
So, you see, it is fundamentally something that the supply chain community embraced because it was taught. Why was it taught? It was fundamentally because it was easy to grade. But when you start to actually look at the reality of supply chains, you will very quickly realize that so many things are so nonsensical.
Conor Doherty: Okay. Well, let’s get into that. So again, I’ve written down two points here and I think one describes our approach and the other describes a different approach. So in supply chain, for us, we would say decisions are meant to maximize return on investment.
Joannes Vermorel: Yes.
Conor Doherty: And on the contrary, I think for some—I’m not going to say all companies—but for some companies decisions might be to hit a service level target. Explain somewhat concretely, if possible, maybe with an example in aerospace about, like, how do those two different ideas—like, how do they translate in reality? Why does hitting service level not equal more money?
Joannes Vermorel: Okay, so let’s have a look. Let’s—Aerospace. Very good. Yeah.
Point of maintenance. The point is to keep the aircraft flying.
Conor Doherty: Absolutely. Avoid AOG 100%.
Joannes Vermorel: So, you need an aircraft repaired all the time, safety standards and everything. So, you need a lot of spare parts, spare materials, resources, assets, etc. Your goal is to keep your airplanes flying and now you want to do that with the minimum amount of dollars per flight hours. So, you see, you have an investment and ultimately what you’re buying is flight hours. That’s what you’re buying.
And now you have a limited amount of dollars and you’re going to think, okay, how much uninterrupted flight hours am I getting per dollar invested? In fact, we have now such a reliability for aircraft that we are not exactly looking exactly the way I present, which is more flight hours per dollar. You would say how many interruptions can I prevent because, again, why am I looking at not the flight hours directly but the interruption? It’s because aircraft are already very very reliable. So it’s—The point is they’re now so reliable that fundamentally you’re looking at what remains of unreliability, and thus your investment becomes how, for this amount of dollars, how much unscheduled interruption of my aircraft flyings all the time can I buy? That is a question.
So fundamentally—and again that’s an economic perspective where I start, you see—economics doesn’t, it just gives you a frame of mind to look at reality. So we say, okay, what is the goal? We want to keep the aircraft flying. And then the economic perspective says, okay, here is how you should think about this goal, and you should think per dollar invested how much, I would say, AOG can I remove from my operations. That’s the economic mindset. So reality first, economic mindset, and that will tell you how you should actually quantitatively look at the problem.
Okay, very good. Now, that is what Lokad advocates—you know, reality first, economic framing—and boom, you have your formalization that emerges. And that’s going to be AOG saved, avoided per year per dollar invested. That would be the key question, and that would be the rate of return, you know, something very very close to that.
Okay, now the alternative with service level. Service level tells you a very different story. It just says pick your targets. It doesn’t specify how those targets are being picked. So you would just assume, well, we’re just going to pick something high. Let’s say 99%. Okay. Okay. 99%. Why not? And you’re going to apply that to all the parts. How many parts do we have for an aircraft? Well, an aircraft, the number of parts that you need for maintenance is about 100,000. The aircraft is itself like 300,000 distinct parts, but not all those parts are really intervening into the maintenance. So let’s say we are talking for 100,000.
Now, most repair operations do not need 100,000. They would need maybe a fraction that would need 10,000 depending on the grade of the checks at that time. Yeah. Exactly. So I would say a vanilla maintenance operation, let’s say to be safe, we are talking the order of magnitude 10,000 distinct parts.
Now the audience can do the math. If you say that you have 10,000 distinct parts and they have like 99% service level, which means that each one of them has a 1% chance of not being there, and your aircraft will need out of those 10,000 potential parts, it will need maybe a thousand of them. So in fact, it’s not 10,000. It will need to cherry-pick 1,000 semi-randomly out of those 10,000. So we are talking of a thousand parts and each one has a 1% chance of not being there. Just do the math. You will realize that your actual amount of AOG explodes. You see, AOG: aircraft on ground. So you apply something—you follow the intuition which is mathematical, very high service level—and you follow it till the end and you say, oh, my AOGs just explode.
So, okay, the mathematician would say, “Oh, it’s because 99% is too low. You need to pick something higher.” Okay, but then if you follow this logic—so you say, okay, let’s do it with 99%. And now you have a problem. It’s your chief financial officer that says we don’t have the money.
Conor Doherty: Yeah. We don’t have infinity money.
Joannes Vermorel: Exactly. I mean the funny thing is that sky is the limit, you know. Yes, in aviation. But at some point, even when the sky is the limit, you know, you’re facing a limit. And so we are back to—we see that 99 doesn’t work. So we want 99.9. But then finance says, sorry guys, no infinite money, it doesn’t work, and the warehouse is not even big enough.
Conor Doherty: But exactly.
Joannes Vermorel: So, okay, so now you’re stuck. Now what do you do? Because you are facing a situation where your service levels just collapse into absurdity. And it is not, I would say—this is not an anecdote. I’ve seen over and over the pattern that replicates across all industries are the following because that’s where Lokad started.
So we start—Lokad—15 years ago we were starting and the company said we want, you know, we have the service level and we just want something that mathematically would be compliant because we observed that the observed service level do not match the prescribed service level because you have this distinction between what your system says like this is what I state—I want 99%—and then there is what I get, which might not be 99%. So there is the discrepancy between what you project and what you observe, and people were saying, “Well, Lokad, can you have, you know, the mathematical instruments to make sure that we get exactly that?” And we delivered. We delivered. And it was only catastrophes.
So, that was a long time ago. It was 2008, 2009. So effectively what we did was we reduced the gap between the non-compliance gap between the stated intended service level and the observed, and the more we were closing in on that gap, the more things were getting crazier and dysfunctional for our clients. And in fact it was very very puzzling because they say this service level is what we want. And when we provably deliver that, all hell breaks loose.
And I was—and then I started to pay attention to people. I was looking at what supply chain practitioners were doing, and in fact they were absolutely not following their own service level. So it was very interesting. I was asking a practitioner, say, “Okay, you say 95% service level,” and the guy would say, “Yeah, that’s my target.” But then I look at what you order, it has nothing to do with this target. He says, “Ah, yeah. Ah yes, it has nothing to do with this target, yeah, but it’s still the right quantity.” And I was like, I agree.
But this 95—I say, “Ah, you know, morally, morally it’s what I want, but I’m just doing something else because it’s what works.” I say, and then—and then we have, okay. Was it like a rare occurrence? You see, this sort of situation where I have like the moral statement and what I do—what works? And the answer was, about 90% of the time there was a complete, complete disconnect.
So, fundamentally I had this very bizarro situation where people say we want those service levels, they really think they do, and they do something completely different and it kind of works. And I was thinking, hold on, just remove the service level and do what works. They say, “Uh, I like my service level.” Why? “It’s like poetry, you know? You have something that makes something look good, but it has like no consequence whatsoever on what you’re doing.”
Conor Doherty: Well, this is again to build on that, and I just—I want to be very concrete because again, there’s a couple of sectors we need to touch on and we will push forward. But on the idea of aerospace and again to be concrete, you talked earlier about again the probabilities and the numbers of parts. And I recall because again I wrote a paper on this—Lokad produced a paper on tooling in aerospace—and one of the examples was, if you took, I don’t know, for example, like a hydraulic pump for example or just any piece of equipment that requires, let’s say, 100 parts to be either manufactured or repaired—let’s just say that—and everything has to be done in a certain order because again you can’t start at the end and work backwards; you work, disassemble, and then fix, etc.
If you needed 100 parts to complete that process and you had 99% service level on all of those parts and you need all of them to be available in order to do that today, the probability that you will have all 100 parts available even with 99% service level is a lot less than 100%.
Joannes Vermorel: Yes, much closer to like 35, 36%. And that is just maths. That is just probability theory that is indisputable. Now imagine that you have many things, many schedules that are contingent upon a certain number of parts being available at a certain time. And also that’s only touching on the physical parts. We’re not talking about the people. We’re not talking about the skills of the technicians who—someone might be sick, etc., etc. So just taking that perspective as an example in aerospace with high service levels—and that’s why I’ve seen—that’s the numbers, that’s just maths.
Exactly. And I’ve seen—and as a consequence of this, which is very blunt—people do their stuff very differently in ways that violently and exhaustively contradict the service level per theory.
So, you see, for me it is very much like the way I look at service level, it’s very much like the Roman armies that were, you know, doing all sorts of weird rituals. You sacrifice a chicken before the battle, you know, you do all—that’s one way to frame it. And people say, “Oh yeah, but you know we have won all our battles and every single battle we sacrifice a chicken, so sacrificing chicken before the battle is so important.” I mean, I guess so, if you think so, for moral, maybe.
But you see, it’s the sort of things that—it is, for me, that we are touching on something because supply chain is pre-scientific. There are a lot of things that are fundamentally not very different from mysticism, you know. It is just tradition and there is a mystique of the service. But when you start thinking of, okay, let’s try to use reason, let’s try to assess that—or insert falsifiability—to test what I’m actually—falsifiability is like the gold standard of science. You don’t even need to go there, you know. You see, falsifiability was invented beginning of the 20th century, Popper.
Conor Doherty: We don’t need to get off topic.
Joannes Vermorel: We don’t even need to go into such a refined instrument. Even 17th century people with the Enlightenment and people who were starting to think, okay, can I just—like Descartes—can I just use my reason to see if it connects? And if it’s like a hot mess of things that makes no sense whatsoever, then maybe I can challenge that. And that’s what I’m saying is that 17th-century common sense is already sufficient to change service level. You don’t need to go into the refined 20th-century instrument of science.
Conor Doherty: Okay. Just as a side note here, Alexi behind the scenes, our producer, can you please drop into the chat—now, I believe the supply chain lecture is Miami. It’s Miami MRO. You did a lecture many many years ago where you decomposed all the problems in aerospace. You took an example using an airline called Miami. Please drop that lecture in. And can you also, Alexi, drop the link to the tooling in aerospace white paper just to substantiate what we were talking about here because we do need to push on.
But what I want to push on to—and again underline the idea of hidden cost—you’ve made the argument before that people confuse the idea of high service level with high quality of service, and that this is where the cost comes in. Like, if you have high—a high service level but crappy quality of service, you’re hurting your business. Expand.
Joannes Vermorel: So first, that is the problem of bad verbiage. You know, the vocabulary is misleading. It’s just like safety stocks is completely unsafe. Business intelligence: it’s not about business and it is not intelligent either, etc. So we are overrun by bad terminology. Service level—I mean first, when you refer to a level, you know, stock level, everything level typically refers to a quantity as an integer—stock level. If you want to refer to a percentage, you would say rate. That is the proper terminology. So first, you see that it is that broken. So first, what we are calling service level should be more like service rate because things at a rate are percentages; levels are absolute integers.
Conor Doherty: Let’s make it concrete though.
Joannes Vermorel: No, I mean, it’s important because, you see, if the words that you use make no sense whatsoever, why do you think—I mean, let’s pause for a second—the terminology is completely bogus. Why would you ever expect that something will emerge out of that? You know, again, it is crazy. If I tell you your supply chain works because the god Apollo does something every day for you—you know, I just made the word Apollo; I don’t even know what this god would be actually doing—it’s just entirely made up. It’s pure mystification.
Why do you think it will work? So, I mean, we have to go back to—if you have a concept that is so badly designed it’s not even properly named, hold on. I mean, you see, the proof is on the people who are pushing the bad concept. So now, what are—okay, so we have level that should be rate. Quality. Service. Let’s—yeah, let’s discuss about service.
Is this service—has anything—so when you say that, again, we need to look at reality. Okay. Reality, what does it say? The reality says you have people, potentially a lot of people, who are your customers; they have expectations. Okay. Those expectations are very very fuzzy. Why do you think that your percentage that you just compute has anything to do with what they expect? I say again non sequitur. These things are absolutely not connected. It has nothing to do between the expectation and what they say. Let’s take it very simple.
I am shopping on Amazon. I order something now. How is my expectation—how does it unfold? Well, Amazon tells me, “Mr. Vermorel, you will be delivered on that day.” So I have a date. So now, the question is, technically what I want is the thing to be delivered on that date. They pick the date. Amazon picks the date for me. So fundamentally they are setting the expectation for me. If they think that the date they are picking is not feasible, then they just pick another date. And for me the management of expectation takes the form of you’re telling me “this is a date,” I agree or I do not agree.
You see, where do you see those percentages and whatnot? This is not how the problem is framed. The problem is framed by the retailer who says, “You want this product? I’m telling you this is the date and the price. Do you agree or not agree?” That’s the way the expectation is framed. Your service level is not even close to—you—you’re not even living in the same universe. We’re not even talking of the same thing. The customer is not even there. The date is not even there. The etc., etc. And that’s the problem that I have with service level, is that it’s something that is fundamentally a byproduct of a mathematical perspective.
And the only upside is it’s nice to generate puzzles for students. It is not something that emerges from looking at reality. And I say we need to have a look at reality—how the problem is actually framed in specific situations, and then you will realize when you—because you would say, “Oh, you just cherry-picked Amazon.” I say you can cherry-pick Amazon: service level doesn’t work. We cherry-picked aviation: it doesn’t work. We cherry-picked fashion: it doesn’t work. Maths—I mean, it’s like you can cherry-pick like all the verticals you can think of, and every single time it doesn’t work.
And not like, “Oh, it’s almost right but not quite.” No, no. It’s like, no, no, it doesn’t work at all. It’s like super violent. Again, I’m not talking of a subtle mismatch where you can just tweak the thing and it will work. No. It’s like violently disconnected, completely irrelevant, must be classified as “don’t care, irrelevant.” And if you can pick 20 verticals and you say, “Let’s investigate what is the quality of the match between what reality tells me and what the math lets me—if I recycle this intuition of service level—what is the quality of that equation?” And I say it’s like extremely crappy, 20 out of 20, then you should just say this is a bad concept that needs to be discarded.
And guess what? There is like an infinite number of mathematical concepts that are attractive because you can make puzzles so that you can grade students, and they’re completely useless. You know, I say that as a mathematician. It’s not because there is an infinite amount of mathematical stuff that you need to embrace, especially when they are profoundly disconnected from reality. Your formalism must be reality-driven.
Conor Doherty: Yeah. On that note though, again, someone can easily say—because you said reality-driven. Someone can say, “Well, the reality is,” and I’m sure you’ve heard some variation of this idea. “Service levels are the implicit contract that we have with our customers. We violate that, we lose our customers.”
Joannes Vermorel: Show—show me any, you know, show me any client in the street. Let’s take people in the street and say, do you have a percentage-based target for your expectations? I mean, just ask people: what do you mean a retailer is delivering a good service for you? And how many people would actually reply to you, “Oh, you know what? I really like KFU because for Nesqu they have this 98.5% service level—it’s so great as a store. I don’t like lair because they have only 97.”
Sorry, it’s—you see—we’re not picking on lair. Again, it is mental. Nobody thinks like that. Nobody thinks like that.
Conor Doherty: Our customers don’t. Yeah.
Joannes Vermorel: And businesses don’t either because again, if we go back to airplanes, people think they want their airplanes flying. It has nothing to do with the individual availability of the parts. Especially since, for example, in aviation there are also other things that we did not mention—substitute. There are many parts.
Conor Doherty: I literally just wrote the word “substitute.”
Joannes Vermorel: Yeah, there are many parts that are compatible. So when you think and you have substitutes, then all your KPI—like percentages—make even less sense.
And that’s why I say this is a fiction. This is a fiction. And this is mysticism. Again, if you look through history, people were extremely convinced of stuff; when you look back it looks like impossibly strange and weird and dysfunctional. And to think that the 21st century would be the first century where we realize that there are things that are completely nonsensical that are still lingering around—I mean, what are the odds that we are living in the first century where suddenly we don’t have broken ideas anymore? Yes, we do, and service level is one of them.
Conor Doherty: Responding to some comments—well, the thing is we’ve touched—we’re kind of mixing—not mixing—but like we’re touching on retail, we’re touching on aerospace. Again, a point that has already privately been sent to me is again, realistically there is a difference between citing, let’s say, retail where your lair—you have in a week 10,000 customers. If you don’t have the product somebody wants—okay, I lost—and they leave, I lost one client, but it’s replaceable. Like the damage is not that catastrophic. That’s very different if you’re working in aerospace and you have serious financial contractual penalties for losing or for missing a deadline, and then, “Oh, I’ve just lost 50% of my business because I’ve violated contractual penalties with this client.” So again—
Joannes Vermorel: But again, service level is a one-dimensional super naive approach. It’s not the way the world works. That’s not—in aviation, for example, those small companies who are serving as MRO, specialized MRO, for example, they have—if a part is not there, they can always send the component itself. You know, for example, people send you a compressor to repair. You don’t have the screw—so you’re missing one screw for a compressor—so you send another compressor.
And you will do what is called a standard exchange, which is you sending me an old compressor that has already many flight hours and flight cycles, and you want me to change one screw. I don’t have the screw, but what I have is a fully replacement that is brand new. I send you the brand new thing and I charge you for the difference.
Now the clients, they would say what we want—and it’s completely accepted to do what is called standard exchanges. Now the client says when you do that, you are charging more.
But by the way, the thing sometimes happens in reverse. So they send you a brand new compressor that needs just a screw, and you send an old compressor that is serviceable. And so your provider, the MRO, can end up paying the client because the difference in the standard exchange price is—what they’re sending back is worth less than what they receive from the client. So again, then the question becomes how much frictions and complications do you create by having all those standard exchanges? That’s going to be a discussion with a client, and it will depend on what do they expect.
And it is again—what I’m saying is that you need to look at the reality. What is the reality of the flow? What are those atoms that are moving, you know, through your supply chains? What are we talking about? And then you have to clarify how the expectations themselves are structured. And I can tell you expectations are almost never—never, that’s really exceptional—never expressed at the SKU level or at the part number level. You see, it’s just the expectations of the client, the expectation of the supplier, the expectation of your partner, of your transporter—they are not expressed at the part number level. They are not expressed at the SKU level. But your service level is. You see, you have such a gap. You’re not even apple and oranges. You’re not even at the right resolution.
And that’s why I say service level is a bad, bad proxy because it is a very very simplistic, blunt mathematical instrument that is for me very much of the toy category. It’s a toy. It’s a mathematical toy. It’s good to grade students. That’s all there is that is good about service level.
Conor Doherty: Well, I’m going to paraphrase a DM that I’ve just received there, but it’s—well, if service level isn’t the KPI, what is the KPI?
Joannes Vermorel: Rate of return.
Conor Doherty: Let me read it because I just want to be fair because someone actually wrote this. If service level isn’t the KPI, what is the KPI? Please give me one number (or one dashboard) that helps me judge whether my supply chain decisions create value.
Joannes Vermorel: Good. I have one number for you. Rate of return. Everything boils down to rate of return. Everything. That’s what I put in this book. Okay. I don’t need, like, the score as from ASCM, you know. I don’t need 250 or 300 metrics. No, no. Everything collapses in the end: rate of return.
The game of supply chain is I take my hard-earned dollars, I convert them into atoms—you know, physical stuff. I do things: transport, transform, distribute, market, etc. And then I get rid of those atoms to get dollars back. And what I want at the end of the day is more dollars at the end of the day—after doing that—than what I had initially when I started. So you see, you’re playing a game where you convert your dollars into stuff and then stuff into dollars, and you want this game to be profitable. So all your decisions are resource allocation. You allocate resources, and thus you’re doing that to maximize your rate of return. That is the one central metric and everything in the end boils down to rate of return.
This is the economic perspective. Now the thing is that to get the greatest rate of return, you need strategy, you need vision, you need innovation, you need plenty of things. I’m just saying that all the stuff that you need in the end will materialize through superior rate of return. Rate of return is a metric that is like keeping the score—how good you are. It just tells you, just like if you have a football match, you count how many, you know, scores have been made, and that’s the way you decide the winner. The rate of return decides the winner. It is not sufficient. You need strategy. You need vision. You need plenty of other things on top of that. But ultimately this is the ultimate score. It’s the rate of return.
And that’s, I believe, the key mistake done—one of the key mistakes done by service level. It is by construction something that is fundamentally anti-economy. If you embrace service level, then you are incompatible with an economic perspective. And that’s the problem that I have because if you’re incompatible, that says that you are incompatible with improving rate of return. And it’s bad because, you see, again you can have a different vision on how you’re going to score your goals, but if you are doing something that prevents you from scoring goals at all, it’s not going to be conducive of victory.
Ultimately, you can say any strategy goes as long as it’s not something so defective that you yourself even remove the possibility to even score goals. That’s the problem that I have with service level. If you’re following that, you remove yourself the possibility of even achieving positive rate of return. That’s very bad. And by the way, that’s why people don’t do that. That’s why I was saying it’s poetry. People have their targets and they don’t follow them. And why? Because the supply chain practitioners, when they use their common sense, they realize, oh no, no, no. Intuitively, the company needs to make money. I’m not going to do something where the company is obviously going to lose money. You know, they have this common sense in their head. They are not going to voice it out loud, but it runs in their head. Say, “Am I about to spend tons of dollars for something that is wasted? If it’s obviously the case, I don’t do it.” And that’s why they—and by the way, this is a correct reasoning—but it will contradict over and over and over the service level.
And what I say is your common sense, which is “If the company makes money out of that, it’s a good investment; if they lose money out of that, it’s a bad investment,” this—your intuition—is correct. Your service level is not correct.
Conor Doherty: Okay. Well, there are some questions, so I’m just going to push forward and ask for quick concluding remarks. So again, if the problem is that service levels are what you might call a blunt proxy—directionally correct, like directionally, casually, but yes, but like lower resolution than you are capable of doing—okay. Well, then what does the replacement look like in very simple, everyday terms? Like, what changes in terms of me coming into work tomorrow? What changes?
Joannes Vermorel: When you meet your surgeon and your surgeon tells you there is a cancer, do you ask your surgeon, “Okay, you’re going to take away the cancer, what are you going to put in place?” It’s not a good question. You don’t replace your cancer by something else. You just remove it. You know, it’s the solution. That’s the thing, is that people think that—
Conor Doherty: I’m asking an operational question—pardon the pun—an operational question.
Joannes Vermorel: The correct answer is subtractive, not additive. You don’t replace service level by anything. You just remove it. You know, for example, again, let’s go back to the Roman legion. Doing animal sacrifices before battle—US Navy. Does US Navy still do animal sacrifices before engaging the enemy? No. So, what have they done of the animal sacrifices? Did they replace those sacrifices with something else? No, they just stopped doing that.
You see, so very frequently the solution is simpler than what people think, is that when you’re doing something nonsensical, the solution is not necessarily to improve this thing; it’s just to stop doing it. So what is the solution for your service level? Just remove those service levels and you will be good. You will be good because already your team—I can already tell you because I have experience—your teams are already not using your service level. That’s a good thing. They are already—because they have so many edge cases where it’s so broken that in fact they’re already doing things that are widely different from what you think they are doing with the service level.
So fundamentally your belief that service levels are useful is misplaced, just like the belief that troops need to sacrifice an animal before engaging with the enemy is also misplaced. Same problem. You just subtract that and you have something better. Now again, if you just subtract the animal sacrifice, can you turn a Roman legion into a modern, you know, platoon of marines? No, because they are missing tons of other things. So you see, it’s just a step. You’ve just removed something that was nonsensical. If you want to turn your Roman legion into a modern army corps, you have many other things that you need to do. But you see, the subtraction of the stuff, it’s a net win. It’s a net win. You’re not losing anything here.
And so what I’m saying is that if you want to have the full picture, you can look at this book. It tells you how you should articulate rate of return, how everything falls into place. But if you don’t have the time to read the book, I can already tell you: just remove the service level and everything will be already kind of better. Just like if you remove your time series, everything will kind of be better as well. There are plenty of things that are so dysfunctional. If you remove them, it will be better. Just like cancer—if you remove cancer, you’re better.
Conor Doherty: All right. I thought you would just say rate of return and better software, but okay. Okay.
Joannes Vermorel: The problem is rate of return is something that will slightly, slightly, slightly remove a few edge cases. But you know, again that would be the equivalent—if I retake my analogy with the Roman legion—would be instead of sacrificing a cow, you sacrifice a chicken. Why is it an improvement? Because a cow is such a logistical problem to carry around. So at least if you sacrifice a chicken instead of a cow, it kind of makes the process smoother.
And that’s what you would get with fill rate. Yeah. Fill rate—it’s less a problem than—but fundamentally still a problem. So that’s where—and you’re not looking at the problem the right way. You see, that’s the sort of thing that—you need to be looking at the problem the right way, and it cannot come from the math. It needs to be reality first, and then you formalize in this direction. Not formalism. You do not start with a formalism and then you say how I’m going to bend reality so that it fits. That’s just the wrong direction. That’s not the right intellectual direction. You need to do it the other way around.
Conor Doherty: All right. Well, I’m going to push on. I will close with a comment and then I’ll ask the questions. But someone who works in aerospace—I’m not going to say the name, they wanted to be anonymous—“I think service levels is like EBITDA—thinks it’s an appropriate measure, but in reality it only provides a very simplified view of the situation.”
Joannes Vermorel: Yes, I very much agree. And just—and by the way, how can you test this hypothesis? Well, Buffett, one of the richest and most successful investors of all time, told you, “I never invested in any company based on EBITDA,” because for me he was literally saying EBITDA is vanity. It is unreal. It is glorifying the ego of the management. It is not real. And well, you can say, “Oh, it’s just an opinion.” Yes, but it’s an opinion from someone who became like super super rich on having the right opinions. So, you see—modicum of trust here.
And yes, for me it’s exactly that. EBITDA is a fiction. Does it—you see, because it’s agreed upon and it has a certain style, it does convey a little bit of information. You know, I’m not saying that service level, intended or measured, carries no information. It does. It does. If you go back to the book, there is a chapter on information. So, it does carry some information.
But if you think about it, it’s just a very poor signal. That’s what I’m saying, is that—and again, I think Warren Buffett would tell you if you have the choice between investing in a company where you know nothing and investing in a company where you just know a bit, Warren Buffett would say it’s like a very foolish thing to invest with so little information. But if you give me no choice—it’s like I’m sent to the gulag if I don’t comply—then I will tell you then, yes, give me the EBITDA. It’s better than nothing, but you see, it’s still very very poor.
Okay. And that’s the sort of things that you need to think about. It is a very very crappy signal. There are so many things that are so much better.
Conor Doherty: So these are—I’ll push on because there are two semi-technical questions here. So one is from Terry Alexio, friend of the channel. He submitted this earlier but we were just talking—obviously talking about specifically service levels in retail. How does this apply to a distribution business with multiple vendors and items? I guess like how do you map what you’re describing onto that very specific situation? Because we were talking about aerospace a lot.
Joannes Vermorel: We have to start: what are your customers expecting? Okay. You know, again it starts with—you need to have empathy for the customer. So let’s take again—and here we have to be very mindful of abusive generalization. So let’s start making it very concrete. Next door from my home there is a DIY—do-it-yourself—store. Okay. So we are not talking of a store in abstraction. We are talking of a DIY store.
Okay. Now, they have a segment where they are selling, for example, light switches. Okay. Good. The problem is as follows. My expectation if I am changing the light switches in my apartment is I’m not going to buy one because most likely there are two rooms, three rooms, so we are talking of five, six light switches.
Am I—so my satisfaction is I don’t want four light switches; it’s not going to be enough. You know, let’s say for example I need six. Can I trust that I can go in an AO store and find the seventh and eighth that I actually need? No. Because—and I don’t want to have the light switches that have different colors and style. So, my expectation is that if I buy a light switch, I can have many units—just as much as I need—and they are all the same pattern. Am I very, very demanding? Not really. I mean some people might prefer something that is whitish. Some others would say I prefer to have something that would be blackish.
It’s fine there. But as long as it’s all the same, it’s already a win. So you see, the expectation is light switch. I need to have this many units so that I can buy all of it at once and they are all the same. That’s—so you see that’s the way the expectation from the customer is structured.
If we’re talking about in the same store about a hammer, do I need to have like five units of the same hammer? No. And am I okay if there is only one hammer? Yes. Because when I’m buying a hammer, what I want is just one hammer. It’s super rare that you would think I need four hammers and they have to be all the same. It just does not happen.
So you see, the expectations come with certain structure and we need to look at the fine print. And what I’m saying is that anything that says “I can skip the thinking, I can bypass the thinking” is going to let a lot of money on the table. That’s all that I’m saying. You see, you cannot with impunity literally ignore basic expectations from your customers and think that you will not be punished by the market in terms of dollars or euros. You will be. You will be. Again, that’s what I’m saying. And so if we go to retail, I just say you need to think of what is the expectation of your customers. You need to have empathy and think like them. And by the way, at some point you will even survey your customers and develop instruments so that you can understand better exactly what they are about.
And this process will take demand efforts and you will end up with some kind of model emergence that will reflect what is my capacity to satisfy customers. And I can promise you one thing: it will not look like the service level. It will be something very different—maybe simple, but very different.
Conor Doherty: Simple manufacturing?
Joannes Vermorel: Yes. Okay. Yes. Again, manufacturing for example—that depends on so many things. But sports goods, and sports goods—you, for example, it’s again if you’re Nike, what does that mean to meet the expectations of your clients? You know, you really have to think about it.
If you were thinking that everything that matters is shoe size, the amount of products by Nike would be something like 20, you know—like a certain number of sizes for men, a certain number of sizes for women—and boom, that would be it. So it turned out that they have, top of my head, something like 50,000 distinct products per collection. So obviously the expectations of the customers are way more than just the size.
So there is depth, and so you need to go into the detail of what are we talking of. What are people truly expecting? And guess what—it’s going to be so obvious. Sports—there would be different sports. So they expect the quality of the offering sport per sport. You’re going to be running—it’s not going to be the same shoes than if you do running or if you do basketball or if you do football or if you do blah blah blah. So, again, the expectations have structure, and the service level is an illusion to tell you you can bypass all of that. It’s unimportant. I say no. The expectations of your customers are the reality. This is a base reality that your supply chain operates in. You need to have empathy, sympathy for that.
Conor Doherty: Okay. All right. Push on. Thank you. Next question is from Michael, who’s a follower of Lokad. Good to see you. So this actually is a question specifically about a client, so I’m going to slightly reformulate it, but I will get to the heart of it. So for context, we recently posted about our work with Trek Bikes. They’re great. And the core, the nucleus, of that project was configurability—so their Project One bikes. Actually, we spoke to Dan Charack a couple of years ago. We recorded an interview and again, configurability for high-end bikes.
So the question overall is: how do you manage the stock of parts for highly configurable products? And I remember that’s a client—so high level for that.
Joannes Vermorel: High level—I mean first, you have to think that especially for high-end products like Project One for Trek, every single part has many alternatives. Many complex—we are talking of a selection where, through the combinatorics, you’re going to take let’s say two dozens of selections, and every selection you have sometimes like 50 options.
And it turns out that if you start to talk with enthusiasts and people who really like this sort of bikes, they will tell you when I take a custom bike, there is like out of the two dozen options that I can choose, I only care about five. The rest I will happily follow the default.
So let me rephrase. You have two dozen options. People have a strong opinion on five of them, and the rest they will happily follow the default, granted that your defaults are making sense.
So now, what does that mean in terms of satisfaction? Well, for your default, does it make sense to pick a part that is already out of stock? So you see, the client says, “For this handlebar I don’t care, pick whatever—something good,” because obviously if you’re buying a bike that is straight, you want something good.
Conor Doherty: Yeah, and Project One is expensive, just for the record. It’s a luxury product essentially.
Joannes Vermorel: I mean, luxury—it’s very high quality for a very good price. I think that the way they would frame it. But the thing is that if you say, “I trust Trek, the brand, to do the right choice for me and pick something that is really good because I have an opinion, but not on this specific component,” then if—if the company were to pick one part, and they can pick another part, but they pick one part that is out of stock and that slows down the delivery accidentally for the customer, the customer says, “You picked wrong. You picked wrong. I did not insist on picking this part that is out of stock. Help me and auto-pick the parts that is good and not out of stock.”
So you see that again, the expectations, when you start thinking about that, is that the expectation—if you think of a configurator—is that the client says, “You have dozens and dozens of options. In fact, I only care about a few options where I have a strong opinion, and the rest ultimately is ‘Do something good—good price for value, good something—and just sort it out for me.’” You see? Because the thing is that with a configurator, people don’t want to have necessarily infinite configuration. They have an opinion because in their case they have a few things that they want to have just right, and the rest they would say, “I trust the brand—do what’s good,” and so you realize that again, your perspective of service level is completely nonsensical because the client is literally telling you, except for those few things where really if you don’t have it it’s going to be a problem, all the rest is up to you. Do something good, and you have the leeway. I give you the leeway, so use this leeway to do something good.
You see, that’s why I say it’s very—and by the way, Trek is not the same. It’s—Dell for computers is doing exactly the thing. For example, I buy the workstation and for the workstation I am very specific on the Nvidia GPU card. I want this one because I do deep learning and whatever. So I say, “Okay, I want this one. For the rest, just do something that kind of makes sense.” Yes, you need to have a power adapter that fits those power-hungry components. You need to have this and that and that and that. But then, you see, I specify one or two components that I really care about, and for the rest I let Dell do their magic so that for the rest it is just a sensible selection. And you see, that’s again—that defines the quality of service and the expectations in ways that have nothing to do with service level. And what I’m saying is that the journey that I’m making—think like a customer, embrace their perspective, and then formalize that so that you can, in the end—the formalization is crucial so that you can have software at the end of the day that deals with it.
But it is the sort of endgame perspective. Formalization is not the starting point. It’s the endgame once you have the high-level understanding.
Conor Doherty: Alright. Well, push on. Thank you. And last comment—I’m going to read this verbatim and get your reactions—from Harold. Hello. “If you maximize service level while minimizing your total cost, you will see beautiful improvements. Return will increase automatically.” Right. Not sure—
Joannes Vermorel: Absolutely not.
Conor Doherty: Not sure if that’s a—might be a tongue-in-cheek comment. I’m not sure.
Joannes Vermorel: Absolutely not. Yeah. Yeah. I mean again, let’s go back to a fashion retailer. You maximize your service level. Fine. So you have now your—we are in April and your store is full of winter garments. It’s not going to be good. It is not. That is a direct consequence. If you max out your service level, you will have in April a store—fashion clothing—that is full of winter apparel.
Not good. Really not good. So you see, and that’s when—and you see, that’s why I say if you tell me a theory and it takes like half a minute to find a blatant contradiction, you don’t even need falsifiability the way Popper style—it’s just it’s so broken that just a thought experiment of 30 seconds just demolishes the thing. And the thing is that it was not cherry-picked. You can give me any of the 20 top, you know, usual verticals. I will come up with a counterexample on top of my head. It is not even difficult. So that’s why I say it’s very broken. It’s something that is so dysfunctional that you will find—you can imagine experimental thought situations that will just be catastrophically bad if you embrace the service level perspective.
And the reality is that even people who do say that they embrace service level—in practice they don’t. And why? It’s because supply chain practitioners are not crazy. They’re not lunatics. On top of their head, they have this economic reasoning: “Am I doing something that is going to be catastrophically costly for my company? Yes, no. If yes, then don’t do that. Is it going to be profitable for my company? Yes. Let’s do more of that.” You see, and what I’m saying is here it’s very much like common sense. And what I say is this common sense is right. If you have someone who tells me very bluntly, “You know, I look at something—when the company loses money, I do less of it. When the company earns money, I do more of it,” I say, okay, fine. Directionally you’re on something that looks like very very good. Keep doing that. Just more.
Conor Doherty: Alright. Joannes, that is all the questions, both private and public, that have been addressed. We’ve been talking for about an hour. In summation, the hidden costs of the service level perspective in concrete terms. What can people take away?
Joannes Vermorel: This is an opportunity cost that is massive. It’s massive because service level is a non-economic perspective. So it means that it’s something that is deeply uncaring for your profits and your bottom line. They just do not care by design. It is—that’s the problem with non-economic perspectives—it’s something that literally says, “I don’t care in the slightest about your profits.” And what I’m saying is that when you have something that violently disregards your rate of return, disregards the profitability of your company, usually it turns out to be very wasteful.
It’s just the way it is. You know, if you have a system that just completely disregards waste and profits and everything, you’re going to have a lot of waste and you are going to have a lot less profit than you would by just not having that. That is all what I’m saying. And so the lesson is: just subtract those service levels. Just remove them. Remove them. Have a look at the expectations from your customers. Ask yourself, “Okay, what are my customers even expecting?”
Ask the same question about your suppliers: “What are my suppliers expecting from me to give me a lower price?” Because again, if you behave badly with your customers, they leave. But if you behave badly with your suppliers, they don’t leave. They just charge you for the inconvenience. So you need to behave—to do the right thing—for your customers so that they stay, and you need to do the right thing for your suppliers so that they give you good prices. And that’s as simple as that. And thus you need to think of those—whatever indicators that you have—“Am I aligned with the long-term interest, rate-of-return style, for my business?” That’s it. It’s not that complicated, but it requires having the fortitude to eliminate an illusion that has been around for a long time.
Conor Doherty: Alright, thank you. We’ve been talking for over an hour. I’m out of questions. I think we’re out of time. As always, your insight is appreciated. Thank you very much. And to everyone else, thank you for attending, for your questions, your comments, your DMs. As always, as I say at the end of every broadcast, if you want to continue the conversation, reach out to us privately. You’re already on LinkedIn. You can see our profiles. Connect with us. We’re happy to answer your questions. And on that note, we’ll see you next week. And yeah, get back to work.
