00:00:00 Introduction: the hidden cost of fixed lead times
00:00:57 The contradiction: everyone knows lead times vary, but companies plan as if they do not
00:01:52 Why supply chain theory and software largely ignore lead time variability
00:04:01 Executive impact: misallocated capital, excess inventory, and shortages
00:08:28 Anecdote: when one delayed order becomes the new “standard” lead time
00:10:36 Fixed lead times as a naive forecasting algorithm
00:12:29 Perishable goods example: where the money leaks out
00:17:02 What companies can and cannot control in lead time variability
00:19:06 How fixed lead times amplify other supply chain shortcuts
00:20:49 Warehouse inbound capacity: why smoothing deliveries matters
00:24:10 Why mundane daily variation hurts more than rare global shocks
00:29:00 Supplier evaluation and the need for probabilistic lead time modeling
00:33:52 Why OTIF reports and supplier scorecards are not enough
00:35:19 Macro crises vs. self-inflicted daily decision errors
00:38:55 Amazon and the mastery of mundane variability at scale
00:42:13 Audience question: historical averages and buffers
00:45:18 Chemical manufacturing example: expensive inputs vs. critical cheap inputs
00:49:12 Pharma question: modeling lead time variability in practice
00:56:28 Why historical lead time gaps alone do not trigger good decisions
00:58:23 Aside: Tetris in Excel and the limits of spreadsheets
00:59:01 Who should own the loss function?
01:03:22 Do short lead time businesses face the same issue?
01:07:25 Do we still need safety stock?
01:10:23 Practical next steps: robotizing supply chain decisions
01:14:12 Closing remarks
Summary
The problem is not that lead times vary. Everyone knows they do. The problem is that companies plan as if they do not, then wonder why capital is wasted, inventories are wrong, and staff spend their days firefighting. Fixed lead times, safety stock, and buffers are not economic reasoning. They are bureaucratic shortcuts. The real damage comes less from spectacular crises than from countless small, self-inflicted errors repeated every day. The solution is to model uncertainty explicitly and make decisions by expected economic return, not by habit, averages, or administrative convenience.
Full Transcript
Conor Doherty: This is Supply Chain Breakdown, and today we will be breaking down the hidden cost of fixed lead times. It’s good to be back. You know who I am, Conor, marketing director here at Lokad, to my left, and as always, Lokad’s founder and the completely unflappable Joannes Vermorel.
Now, some quick questions for the audience before we get started. Question one: do you think lead times vary? Now, before you type, “Of course, Conor, you idiot, of course they vary,” follow-up question: if you said yes, do you plan as if lead times are fixed? That is, of course, the tension that we’re going to be talking about today.
Now, on that note, we like this to be interactive, so get your questions and comments in as soon as possible and I’ll try to fold those into the discussion. And with that, Joannes, let’s get started.
So, we’re going to skip past the obvious question, “Do lead times vary?” That is trivially obvious. Again, I ran a very simple poll yesterday on LinkedIn, and if I eliminate all the joke answers from people who were just trying to be trolls, and it was funny, well over 90% said, “Yeah, of course lead times vary.”
Now, the question, or the tension, as I just alluded to, is what people say and then what companies do. So, if you grab 1,000 practitioners, if you grab people off the street and say, “Do you think lead times vary?” “Yeah, yeah, of course, Conor, trivially obvious that they do.” But take those same practitioners, put them into a company, and all of a sudden the decision-making policies and processes just ignore that reality, or at the very least push it really far to the side.
So my first question, Joannes, is: how do you explain the contradiction between what people say they believe and then what companies actually do?
Joannes Vermorel: There is a whole series of reasons that pertain to it. Some are very basic, such as the mainstream supply chain theory, a little bit by accident, doesn’t really acknowledge that variance, but it is what it is.
So you take the theory books, 1,000 pages, that give you, I would say, a very extensive rundown on what is considered supply chain theory. It’s mostly absent. As a consequence, when you look at the vast majority of supply chain planning products, forecasting products on the market, lead time analysis and lead time modeling or forecasting is also absent. So that cascades into the practitioner. They may be interested in modeling the lead times, but the tools that are available don’t do that.
Then there is the fact that, also for management, very frequently the intuition is, “Oh, that’s a problem for our supplier. We should just…” It’s like a defect. This needs to be addressed definitively by the suppliers, and then we’re good. So you see, why should you actually deal with something that can be addressed and then it’s gone forever? Obviously people also know that it’s not going to be gone forever, but just thinking of it as something like a problem that belongs to the suppliers is a way to not really pay much attention to it.
So you end up with a situation where you have a vacuum on that front from the theory, a vacuum from the software, and on top of that you also have the fact that the sort of implicit take of the organization is, “Well, that’s a problem for suppliers to solve. If they solve it, then we don’t have anything to do,” which technically is true, but it never happens.
Conor Doherty: One of the things we’ve discussed before, again, the hierarchy of the different sources of uncertainty and their importance when it comes to supply chain decision-making, you’ve said previously, and I think most people would probably agree, demand is probably number one. But you’ve argued, we’ve argued multiple times, a very close number two is uncertainty in lead times.
And my point here is, again, there seems to be a lack of investment, lack of, on a company level, not on the individual, on a company level, a general sort of lack of interest, attention, software, and focus on lead times, even though most people do agree that they are critically important.
So my question: what is the immediate executive-level dangerous impact of ignoring this vital source of uncertainty?
Joannes Vermorel: The thing is that if you ignore your own lead times, you’re going to allocate way too much capital to your inventory or way too little. Again, it’s a massive driver of your business. It is extremely consequential.
If you think that your lead time is seven days, but in fact it’s one month, it’s going to create a whole cascade of problems through your entire chain of operations. It’s as simple as that. It’s a massive problem.
And the interesting thing is that people understand that not delivering clients on time and in full is a big concern, and clients leave. But then you have to understand that it is exactly the same thing for your suppliers. They try to do their best, and yet it is complicated. They have to deal with a lot of unexpected events, and thus they are not always on time and in full. And that’s why you have your lead times that vary.
Because it is systematic, those variations are not just a once-in-a-decade thing. If you have dozens or hundreds of suppliers, you will have every single day a certain percentage of everything that you’ve ordered that will be late, and probably not the quantities that you wanted. That’s just part of doing business, and ignoring that is just running one-eyed on the business.
The sort of penalty would be very, very consequential. If your factory stops producing, it’s very consequential. If you can’t deliver, if you can’t fulfill your own promises for your own customers, it’s very impacting. If you have a massive excess of inventory that you don’t really need because, in fact, the supplier says contractually, “We will say the lead time is two weeks,” but in reality they deliver in two days, you just carry tons of inventory for nothing. So again, it is very consequential.
One extra reason it doesn’t get the interest it should get is that when people play the political game of S&OP with a tug of war on the forecast, what is on the line is the budget of every department. The sales department wants to have the targets as low as possible so that they can exceed expectations. So they are going to sandbag like crazy the demand forecast targets. On the contrary, production wants to do the exact opposite. Why? Because then they have more production capacity and more budget for manufacturing, production, or warehousing.
So you have those tug-of-war dynamics, and that makes everybody super focused, because those, I would say, classic point forecasts on the demand side are consequential for the budget allocation. But for lead times, not so much. Lead times are extremely consequential, but lead times are not very consequential for the specific political games that are being played in the company. Thus, that’s one of the reasons why it doesn’t get all the interest it should get.
Conor Doherty: Right. Well, there was, there was and still is quite a bit of interest in this topic. And I do have a lot of anecdotes that I have promised I would anonymize. But there are certain anecdotes I want to pitch to you later and get your opinion.
But just to properly lay the groundwork, and I think this is a key point for later, when we talk about fixed lead times, we’re essentially talking about averages, taking the average and basing your decisions around that, or the contractual lead time.
Joannes Vermorel: Contractual, you say, “The contract says seven days, then we put seven days,” because very frequently companies have seen many, many supply chain practices where literally lead times are not even measured. Occasionally a supplier is really, really bad and there will be a series of angry phone calls saying, “Oh no, you have to do better.” But mostly supplier lead times are just constants in the systems.
And where do those figures originate from? Very frequently they literally originate from the contractual agreement with the supplier. This contractual agreement might be a decade old. So it’s interesting. Sometimes you have figures that are desperately out of date, and that’s what it is.
Conor Doherty: Well actually, on that note, you set it up nicely. So this is, again, an anonymized example, but because you were talking about, is it the average, is it contractual, another example from a follower of Lokad. Excuse me, I’m going to read this verbatim and just get your response to this. Again, it explains how this person, I know, believes, “Yeah, lead times do vary,” but again, once you go to work, what are the policies that, on a department level, are enforced?
So imagine this is their situation: “Imagine the normal lead time for a SKU-supplier pair is 14 days. If we place an order and the supplier says, ‘Sorry guys, this time it’ll be 100 days due to reasons X, Y, and Z,’ then the system assumes, going forward, that 100 days is the new standard for lead times. This then stays in place for the next order because the system thinks the fundamental supplier behavior has changed. Obviously this leads to huge overstock.”
Now again, that’s one anecdotal example, but the whole point is: how radically dissimilar is that, or how bad an example is that compared to how people typically handle lead times?
Joannes Vermorel: For me, it’s the sort of dysfunctional situation that I see very, very frequently among many companies when it comes to lead times. This sort of super bad policy is like an incredibly naive lead time forecasting algorithm. You see, when you say, “I just pick the last lead time and I assume that to be the baseline onward,” that is literally a forecasting algorithm.
Except that it is not understood and implemented software-wise as a forecasting algorithm, so there is no backtesting, there is nothing that would actually put some pressure to improve this thing. Lead times are treated like a purely clerical element of the many supply chain workflows.
You can think of it from a software perspective like this: a supplier has an address, it has a postal code, so there is a text box where you can enter a postal code, and if you enter a negative number, this text box will say it doesn’t look like a valid postal code. Lead times are a little bit like that. Enter a lead time expressed in days, and if you enter a negative number it would say, “Probably not good.” And that’s pretty much it.
It’s treated as something that is just metadata, like the detail of the address of the supplier. There is a very shallow sort of modeling take on lead times. And again here, the vast majority of supply chain software literally model the sort of very, very shallow take that the supply chain literature, the academic one, has on lead times. I think that really boils down to that.
Conor Doherty: Well, I want to push on and get to some concrete examples. Again, I want to be very clear, sanitized, because I do want to tease out very, very clearly and in plain financial terms where the hidden costs arise. Because often these things are not directly visible. There’s no neat entry in a ledger saying, “Oh well, this came from that,” or “This waste came from this decision.”
But let’s take an example of a company with perishable goods traveling across the globe with a nominal lead time of ten weeks. Now obviously that could be eight weeks, ten weeks, it could be twenty-four weeks, it could be half a year depending on what happens. Okay, that would be an extreme, but the whole point is any number of variables could interrupt that.
Now where does the money leak out in that decision-making process? Where are the hidden costs there? Because it’s not just, “Oh well, my lead times vary.” Therefore, it’s that sort of catalyst that produces consequences.
Joannes Vermorel: So if we’re talking about perishables and food, and you import overseas, depending on the time of the year, the demand and price for those products vary dramatically. There are plenty of products where the cost in France, and I suspect it is the same in many countries, from having this produce at the right season or the wrong season, the price is like times four. It is really not subtle. The price at which you will be selling your products is very time-sensitive.
Now if you place an order and you expect those goods to land, I don’t know, September 1st, and you expect that you will make a profit because you think, “Okay, what are the usual prices that I would expect for selling that, the expected demand volume also,” because demand is also seasonal, not just supply, you can say, “If it lands September 1st, it is a very profitable option.” Now if it lands November 1st, you don’t have any reason to think that the same quantity landing much later will be sold at the same price point, meeting the same level of demand.
So what can happen is that not only do you not serve the customers that were expecting what you were supposed to have on display in September, so you frustrate all those customers, that’s a massive source of cost, but then you end up with a whole shipment that came way late. Not only do you have lower volumes for demand, but also you may have generally lower prices for the item at this period.
That can be a combo that is quite devastating, because we are talking about frustrating customers, losing sales, potentially losing market share if they go to a competitor, and then you end up with a big stock that turns halfway into dead inventory, or you have a massive write-off. Not necessarily a write-off in quantity, but you have to give away a major discount because it’s not the right time.
This is the sort of chaos that unfolds. In addition to that, if your organization is not natively prepared to deal with lead time variations, it means that it will create a lot of last-minute firefighting. If you have nothing to account for those variations, if it’s not part of the modernization, it’s not part of your automation, then it means that a lot of clerks will have to step in and click buttons and adjust settings and reconfigure things.
Instead of having the supply chain teams being very proactive at anticipating the future and making the right calls for the right allocations of capital, you end up with people who are firefighting mundane, symmetrical problems that are just glitches caused by those completely unsurprising variations, and yet they are disrupting your operations.
Conor Doherty: I’m going to come back to that point because that’s a perfect point to push back a little bit, not necessarily steel-manning the opposition, but it’s pointing out what are the limits of your own control. So for example, when we’re talking about importing perishable food across the world, you’re not really in control of all of that. You personally can’t make the boat faster.
So there is, we do have to disentangle a little bit here about what remains within your control and what are the, to quote Shakespeare, the slings and arrows, the things that are just beyond your control. So if you could just tease that apart a little bit.
Joannes Vermorel: Again, that’s the reason why I say you cannot expect that those lead time problems will be solved, because they are not under your control. Typically they are not even completely under the control of your supplier. It is something where not only do lead times vary, but they will keep varying indefinitely. It’s just not going to be solved anytime soon.
Now, what I’m saying is that those variations are very consequential when it comes to the return on investment that you can expect for every allocation of capital. So if you decide to place a purchase order to a distant supplier and you have these lead times that vary a lot in between, that needs to be properly reflected in your economic calculation for the rate of return. Otherwise you might be ordering way too much or way too little and you don’t even know.
So I’m saying that lead times are very consequential from an economic perspective and thus need to be accounted for, because taking that into account should modify your allocation. That means that when you take into account the variations, sometimes it will shrink your orders, sometimes it will expand your orders. It really depends on which way it goes from the economic perspective.
Conor Doherty: We recently discussed this, I think we released the video last week, but there was a section where we were talking about shortcuts in supply chain. And I think it builds on the point you just made, which was that fixed lead times are an understandably convenient shortcut. But one of the hidden costs there is that when lead times do vary, which most people, if the quasi-totality of people will agree they do, they have the uncanny effect of magnifying the financial impact of the other mental shortcuts that typically go with fixed lead times.
So for example, let’s say we’re talking about importing perishable goods. Often the decision-making stack, let’s just use that term, might be, “I’ve got service levels, I’ve got my ABC classification, I’ve got static lead times. When the things arrive, whenever they do arrive, I will use static MRP rules, or ERP rules, to allocate accordingly.” So I have all these sort of mental shortcuts that are stacking up.
Now when the lead time does vary, let’s say it’s twice as long as I planned, the impact of all those other decisions, like kind of arbitrarily setting or bureaucratically setting a service level, suddenly hits much more financially badly. So I’m just curious, in this new context, do you still agree with that take or do you want to add anything to it?
Joannes Vermorel: Yes. If you are blind to something so consequential, then you have a lot of stupid problems that emerge that are just extremely costly for nothing, because they would have been so easy to prevent otherwise.
Just consider a case where many times you have a warehouse and, on this warehouse, the inbound capacity is limited. So imagine it’s full trucks doing hours-long unloading into the warehouse. Let’s say, for the sake of simplicity, that this warehouse can only get ten inbound semis per day. After that, there are no slots available. You just can’t squeeze another truck. It just does not fit. You are at full max inbound capacity.
Now, the thing is that those trucks are from your suppliers, and when those future trucks arrive depends on the lead time. If you construct a plan where you have hotspots, where on many days you are really maxing out your capacity, meaning that you have many days in your planning where you are already at ten trucks, and many days where you have only five, it means that if there is a late delivery, those ten trucks that are supposed to arrive on a given day can become eleven. Why? Because there is a late truck that just pops up, and now you’re stuck.
And in the real world, when these sorts of things happen, the driver has no other option but to drive back. That can be immensely costly. It’s not even an option to wait because the truck physically itself might already be committed for another shipment. So it can be a massive headache where you have a truck and a truck driver and you don’t know what to do with that, and that creates a whole cascade of problems that now needs to be solved.
If you had acted proactively, you would have said, “Okay, I prefer you to deliver two days later,” so a little bit longer lead time, “so that I keep my loads at, let’s say, eight trucks a day, so that when, inevitably, a truck will be late, I will not hit immediately my capacity.” So you see, it’s just about smoothing a little bit the deliveries. It is very, very simple conceptually. Let’s smooth a little bit that.
But if you want to smooth your inbound flow for a warehouse, that means that occasionally you are explicitly asking, counterintuitively, for a slightly longer lead time. Because usually you can pressure your supplier for shorter lead time, but the cost typically explodes. So usually the economically viable answer would say, “Well, instead of trying to deliver in seven days, what about if exceptionally it was eight days?” And then the supplier is like, “Okay, no problem. If you give it one more day in this direction, there is no problem.”
That’s the sort of thing where, suddenly, if you do that and you smooth the flow, you avoid incidents and then you avoid tons of firefighting that end up being extremely disruptive and distracting for the teams and prevent any actual super-productive work from being done.
Conor Doherty: A key point here, and it is something which again was posed, this question was posed, from someone who generally agrees with us but just wanted you to elucidate the point a bit better. Because internally to this person’s company, people often think that lead times only really vary, or at least only become financially really damaging, when extreme events occur. So for example, the Strait of Hormuz is closed, or COVID. Like these enormous system-wide or systemic events occur. That’s where we get hurt by lead times varying.
No, companies get hurt all the time by very mundane variations. That’s number one. Please explain, because that’s not necessarily clear to everyone, hence the question.
Joannes Vermorel: Again, the default situation of companies that are not using Lokad is that almost 100% of the people are essentially doing busy work. It is terrible. It’s absolutely terrible.
What do I call busy work? If you invest one hour doing something that is not almost 100% accretive, capitalistic. Are you turning your supply chain into a productive asset that is completely automated, and every human hour that you spend makes this asset better? So it’s a money-making machine, and every hour that is invested is to make the machine better.
Is it the way it is done? Absolutely not. The vast, vast majority, it is essentially firefighting, just doing tweaking all the time. Everybody involved is treated as consumable. Every single day, the machine, the company, the flow requires, let’s say, 50 man-days to just operate. So every day, for a large company, let’s say a billion-dollar company in annual revenue, the flow needs 50 man-days of white-collar effort to just keep flowing.
So we are talking about people who are just nudging things in Excel spreadsheets, appeasing the god of the ERP, and this sort of stuff. This is completely non-capitalistic. For me, this is just busy work. You’re doing work that should not even be happening in the first place. And the fact that you’re not dealing with lead times properly guarantees that this amount of firefighting will be large and constant.
That’s where I say there is a massive cost: when you have all this firefighting, it means that there will be less mental intensity to actually call the right shots for all the resource allocations. Because again, supply chain is an economic game where you want to maximize the rate of return of every single allocation of capital that you make through the flow. That is a game, and we are talking about thousands and thousands of decisions every single day for any sizable company.
If everybody is distracted doing things that are not accretive, it means that there is a massive opportunity cost just because you are very distracted. Think of it like this: you’re playing chess and you want to win. Now imagine that at the same time you have to do some kind of firefighting on the side. So you’re looking at the game to try to play the good move, but you’re interrupted constantly by a freaking situation that unfolds and you need to do things.
Just imagine something where you have cooking going on and stuff is burning and you need to adjust things all the time, and you have to split your focus all the time. Are you going to be very good at chess? No. You will be completely distracted, and you will be doing instinctive moves at best. They will not be critically bad, but obviously this is a situation where you will keep doing very blunt mistakes. You will not be at the top of your game.
You see, this is the cost: if you have this firefighting going on, then people are distracted and the primary decisions get made poorly.
Conor Doherty: Well, right at the start of the episode, you mentioned again that the internal perspective of most companies is that lead times, that’s a supplier issue. And you have just described how minor deviations, or daily trivial deviations, “it was a day late” or “two days late,” these things have financial impacts.
That naturally leads to a kind of uncomfortable or potentially political discussion about what is the responsibility here in terms of suppliers and how robustly should you be monitoring these. How does modeling lead times probabilistically influence the way that you evaluate your own suppliers, essentially? And how does that influence the dynamic between company and supplier?
Joannes Vermorel: First, until you have a probabilistic modeling of the lead time, so you can really, if you have that, then you can actually start modeling the economic impact of those variations. And you can do A/B testing in your model, with or without the variation. That will give you a baseline of how much is there on the table.
That’s the starting point, because until you have this economic assessment, you can’t even really start a negotiated discussion with the supplier. Because if you attack the problem like, “Supplier, we want you to be perfectly reliable,” then the supplier says, “At any cost?” What is the trade-off here?
There are many businesses where they will have absolute reliability, but at completely extravagant cost. For example, for a personal item that I sourced recently, if you want, in Paris, to buy a new lock for your door, you can either buy it from the lock manufacturer and it will take like three weeks to be delivered, or you can have also a specialized provider that will actually deliver that within one hour.
Imagine somebody tried to break into your apartment, the lock has literally been broken, and now you have a door without a lock. I just discovered there are people who can literally, within the hour, deliver you any lock. And the thing is that they have crazy coverage, but the price point is five times the normal lock. It was like a €200 lock that would, at emergency price, cost €1,000.
So yes, you can have crazy good lead times, but the price is also crazy high. That’s why I say that if you cannot assess what is the economic impact of the betterment of the lead time, you can’t have a reasoned discussion with your supplier.
And maybe sometimes you could even say to your suppliers, and we have seen that, by the way, for aviation: what if you were to say to your supplier, “We are not actually in a super hurry for this order. We need to have it, but eventually. Could you lower the price if we give you two more months to deliver the thing?” And it turned out that sometimes the answer was absolutely yes.
So you see, the negotiation doesn’t necessarily go one way, into “we want to shorten the lead time.” Maybe you can actually have a much better price if you accept to delay the lead time, because maybe the supplier has a massive problem right now and they will concede a sizable discount if you actually relieve them of the pressure on some of the things that give them very intense price points.
Again, you can only do that if you start modeling probabilistically your lead time so that you can assess the impact of variations, not just the impact of a lead time of seven days versus nine days, but literally the impact of the odds of having negative outcomes. So a lead time that is going to be much longer than usual. What is the impact of having that if it’s like 1% of the deliveries that are very overdue, or if it’s like 5%?
Conor Doherty: If you could contextualize, again it’s off the top of your head of course, but contextualize the difference between what you just described in terms of the economic impact, and what I know, we both know, a lot of people are already doing. Someone might respond to you, “Okay, Joannes, I already have an on-time-in-full report. I have supplier scorecards. I have my contract terms. There are penalties in place.” There are already sort of financial metrics or rules in place. How much better is what you’re describing compared to what I already have?
Joannes Vermorel: Enormously better. Again, just the baseline is: assess what people are doing in your organization and really think that supply chain is a productive asset that should run unattended. Unattended means zero people. Zero. Not 50. No, zero. That means that you could normally remove the entire supply chain team and, for three weeks, the company will run just fine. Not if there is a massive worldwide crisis unfolding during those three weeks, but if it’s just business as usual, it should run just fine, unattended, strictly unattended.
On the contrary, what I observe is that for most companies, a billion dollars of annual revenue, their supply chain would not run a single day without investing dozens of man-days every day to do essentially firefighting, micromanagement, and busy work. All of that, and the fact that lead times are not modeled, are one of the enormous contributors to that, because it creates a lot of surprise situations that then need someone to manually step in and spend hours rectifying a situation that was actually very preventable if it had been modeled correctly in the first place.
Conor Doherty: Just to draw a line under this before I push on, because again I want to be sure that we just hit a very key point there. Because the initial question that started that little chapter was: most companies believe that the real financial pain, or the real devastation, is when COVID occurs, when a canal is blocked. But it’s the accumulation of daily suboptimal decisions that compounds.
Joannes Vermorel: The thing is that, again, that’s my take: supply chain should be seen as an economic game that is played competitively. If a world disaster strikes, it strikes everybody. The price of energy is going up for you and your competitors and your clients and your partners. So is it really a problem for a given company? In essence, for society as a whole, when the price of energy goes up, everybody gets poorer because our capacity to purchase stuff gets lower. But for any individual company, does it change anything? Not really. What you buy gets more expensive, so you raise your price and you forward the cost. It doesn’t really change anything fundamental.
That’s why I think you really need to think in terms of market share. Am I doing something for my supply chain that makes me more competitive? When you think about those world events, they are extremely difficult to predict, the way they unfold. For example, the situation with Iran was extremely chaotic. Nobody could have really predicted what the U.S. administration would do exactly, when, and how it would unfold. Probably if, on January 1st this year, nobody was making any reliable prediction.
And then when it unfolds, it impacts everybody across the board. So I would say it is very distracting, and it’s not going to move the needle much in who wins market share. For example, did this thing change the market share between Apple and Samsung? Not really. It is very consequential, but fundamentally it doesn’t really.
The sort of mundane problem that I described, this is the sort of thing where every company has its own chaos-generation machines, where you have so much stuff that is self-inflicted. And because it’s self-inflicted, there is a great chance that your competitors are getting ahead just because they do less of that.
So that’s why this is for you to fix, as compared to the situation with Hormuz, where apparently it is not your business. There is nothing really that you can do, and that will be the same for your competitors. That’s why I say focus on what you have the most leverage over. And here, for this sort of busy work and constant firefighting, it is very self-inflicted, because if you take a massive assumption such as “my lead times don’t vary,” and your plans are based on that, and then it creates a massive amount of friction and overhead and firefighting, then it’s on you.
It’s really your decision that created all this friction, all this firefighting. That could have been completely prevented. And many of your competitors are already getting ahead by just not having these self-inflicted wounds repeatedly occurring.
Conor Doherty: With better words, you took the idea that I was actually going to summarize with, but I’m going to do it anyway, because the terms I was going to use were macro and micro. The macro is what you just described. That’s a system-wide devastation that impacts absolutely everybody. The micro is what you call the mundane. It’s the day-to-day. It’s what do your suppliers do. Could you have made a better decision with your order, with your supplier, your terms? Could you have known more about the likelihood of a delay there? That’s what you have more control over.
And that’s really, when you think of it, the magic for Amazon, and that’s why Amazon keeps crushing all their competitors.
Joannes Vermorel: You see, Jeff Bezos, when he was leading, didn’t have a crystal ball to know, “We need to do exactly that, keep this thing in stock. This article should be really at this price point.” The idea at Amazon was mastering the chaos at scale to get all those micro-decisions done right. “Yeah, we have very tight monitoring. We have products where the deliveries seem to be unreasonable for this postal code. We don’t know why. We don’t know how. We can’t fix it right now. But let’s make sure that the promise that we display is not overestimating our capacity to deliver this thing on time.”
It’s just a basic thing. And maybe this thing has to be tuned postal code by postal code because it’s much more difficult to deliver a parcel probably in busy New York as opposed to doing that in quiet Arkansas or something.
Same thing for the prices. They constantly adjust the prices of the products that they sell. It was a lot of things where, in terms of mundane adjustment, just doing the mundane things right at scale. And the thing is that if you can do it at scale and it’s very automated, then you can even do it, like Amazon does, at hyperscale. Because if your recipes are automated and can manage a million SKUs automatically, then they can manage, just like Amazon, 300 million SKUs automatically.
It becomes a little bit like an IT challenge to scale out such a calculation. But that’s a very mundane part. The difficult part was to have the thinking that supply chain should automate all the mundane.
Amazon does that for everything. For example, if you ask for a refund, they will have clever heuristics that say, “Oh, you’re such a good client. You’ve been paying so much for the last ten years. You never request refunds. Most likely, okay, done immediately, no questions asked, because you are obviously not abusing this mechanism.”
Because imagine the alternative. The alternative is it creates an alert, somebody has to step in, do a manual approval, they might actually do the wrong thing, etc. So I think that’s why I’m a big proponent of just mastering the mundane variability for all things, lead times among other things, just because it is very actionable.
And in practice, this is where most companies lose a lot on things they could have completely prevented. And it’s really, really self-inflicted for the most part.
Conor Doherty: Well, your words are resonating with the audience. Myself Torres is essentially saying, “Yeah, great insight. Important to watch out for the lead time economic impact.” And on that point, I’m actually going to push forward, because there are audience questions and also ones that were sent before.
And basically everything I was going to ask you is going to be folded into this anyway. So I’ll start with one that I think will set you off, but we’ll cover a bit of ground that we haven’t exactly covered earlier, which is, and I quote here, so don’t get mad at me: “We already track lead times and use the historical average, sometimes with a buffer. What is the difference between that and what you’re proposing, Joannes?”
Joannes Vermorel: Okay. So, you see, the problem: a buffer is a capital allocation policy without an actual economic calculation. That’s what it is. You just say, “We keep lead time plus seven days of stock.” That means that you’re going to allocate money into buying stuff without doing the economic calculation: is it profitable, and how profitable is it? Does this thing compete for storage space in my warehouse, and what is the opportunity cost versus purchasing something else?
When people tell me that, what you have is a heuristic for decision-making. And what kind of decisions? Capital allocations. Because if you tell me “a buffer,” that’s about allocating capital whenever you purchase. And I’m just saying that until you do an economic calculation, you have no clue if this policy that you’ve set up is any good.
Maybe your buffer is exactly right and it’s exactly at the point that maximizes the rate of return, but maybe not. What I’m saying is that if you don’t do an economic calculation, what are the odds that you maximize the rate of return? I would say very, very low, very, very low. And what are the odds that sometimes you do something that is dramatically bad in terms of economic calculation? Probably not as low as you would like.
So that’s my take. That’s really where the difference is. Modeling the lead times paves the way for a probabilistic assessment, which in turn paves the way for an assessment of the risk-adjusted rate of return that takes into account all the possible futures. Then you take the expected rate of return on that.
Conor Doherty: Alright, once I’m convinced. This next question, I mean this could be an episode in and of itself. You don’t know the question ahead of time, so this is a chemical manufacturing context, sanitized because this was sent ahead of time. Just give me your thoughts on this.
“Imagine improved optimization software,” parentheses, “with lead time forecasting, obviously,” “identifies two risks. One is a high-value material that is expensive to hold. The other is a low-value input, but if it arrives late, it blocks production and destabilizes our plans. So how should planners decide the best overall action in this context, like change parameters, build stock, expedite, negotiate, etc., etc., etc.?”
Joannes Vermorel: We are back to the economic calculation. That’s why I say this sort of financial perspective, where I say, “Look, we need to assess the rate of return on investments,” is not a lunacy from the part of Lokad. You have many options that compete. How do you allocate your budget? How do you allocate your resources?
The answer is: which one do you prioritize, and by how much? Do you spread your investments? Do you go all in on something and ignore the other things? How do you tune this sort of situation? “I have a very expensive component that is very expensive to hold, but I can supply in short terms.” There is probably, like with everything, diminishing economic returns. So how sharply decreasing? Well, it depends on the specifics of your situation.
What I’m saying is that the discipline of doing the economic calculation is what will tell you what should be the trade-off, how much you should have. You can think of all the possible potential decisions. So I can hold four days’ worth of inventory for these products, five days, six days, seven days, etc. Each alternate policy comes with its rate of return, and that’s going to be an economic calculation.
And same thing for the thing that is cheap but potentially very critical. In this case, for a chemical manufacturer, that would be the exact same thing as duct tape in aviation. To repair an aircraft, you need duct tape. It’s super cheap, but you frequently need it. And if you don’t have that, you can be stuck with an aircraft-grounding incident for something that just costs cents.
So yes, but again, you need to do an economic calculation because maybe this product that is very cheap is bulky, and maybe you were not taking into account the fact that it takes up a huge amount of space in general, storage space. And maybe at some point you just run out of space to store it, and that competes with other stuff that you would need to store there.
Again, the economic calculation is the idea of converting all those factors into euros or dollars so that you know how to decide.
Conor Doherty: The only thing I would add to that is, because you didn’t explicitly say it, but I know it’s in your head, that this is not a set-and-forget situation. It’s not like you do it once, “Right, I’m done.” The process you’re describing is iterative. It’s revisited. At least in the aerospace example you gave, that would be every day it’s revisited.
Joannes Vermorel: Yes, but again we want to keep the work accretive. So you have this unattended process, which is essentially a piece of software that generates all the decisions, allocations of resources, to get the flow flowing. It’s all about flow.
And then the people, when they work to revise that, they work to improve those numerical recipes. They work to improve those unattended decisions, but in a systematic fashion, so that if you deliver an improvement, this improvement makes the decision of today better, but also the decision of the day after, and the day after, and the day after, etc. That’s how you get something that is accretive.
Conor Doherty: Alright. Thank you. This next question, it’s similar, but I’m going to tweak it just a little bit so we don’t repeat the exact same thing, and it’s in the pharma context. But it does touch on a slightly different point, which is more the nuts and bolts.
Quote: “I’m particularly interested in understanding how organizations can better model and forecast lead time variability, and how those insights can be translated into more effective inventory and service-level decisions.”
They give an example, but it’s very similar. It’s just a pharma-context example of what you described in manufacturing. But this is more literally: random variables, the tech involved. In terms of nuts and bolts, at a high level, what is actually required to do this? Because up till now it is just theory.
Joannes Vermorel: Yes. I mean, what you need is to have the machinery for probabilistic modeling. There is a series of lectures, and there is even a lecture on probabilistic modeling of lead times. Essentially, think of it as a specialized machine-learning component that can turn your historical observations into probability distributions.
That can be done in ways that are very straightforward. In this lecture, I give an example where, if you have the appropriate programming language, we are talking about like ten lines of code. Again, you can refine, but you can have something that is production-grade and extremely tight.
Now, once you have that, you have those probabilistic outlooks. What you need is the machinery that will be able to combine this sort of probabilistic projection that you have for lead times with the probabilistic projection that you have for the other sources of uncertainty: the demand, future prices, potential returns, etc.
And it’s not just the lead times that you probably want to model probabilistically. It’s also the quantities that you want to receive, because maybe in pharma the problem can be that your supplier is not only late, but they don’t even deliver in full. So again, the same probabilistic take should account for that, how much you will get. It’s not just the probability distribution for the delays, it’s also the probability distribution for what you will actually get when it finally arrives. And usually this thing is lower than what you were expecting.
Okay, now, so you have all this machinery that will give you all the possible futures. You need to have a little bit of specialized machinery because, again, Excel won’t cut it. That’s the big limitation of Excel. As soon as you want to enter this probabilistic world, it will not fit the sort of spreadsheet model where you have cells that hold numbers. You need something that is more suitable for that. So it’s just going to be an environment like Lokad, but it could be something else. It’s just not going to be Excel.
Now you have all the possible futures, but you need to have something that will repeatedly ask the question, for all the possible decisions, what is the economic rate of return that they will observe? Here, this probabilistic modeling is mostly a matter of machine learning. Then you have an element of economic modeling, which is literally: if this thing happened, and this happened, and this happened, what is the outcome for the price?
So here you adopt a very post facto sort of thinking. If this happened and this happened and this happened, the cost is that. That’s just a deterministic calculation. That’s literally your back-of-the-envelope modeling of the economic factors.
Now that you have the probabilistic take, the economic model, you need to have a stochastic optimization, a stochastic solver. What does the stochastic solver do? It explores all this space of potential decisions, of options, just to find the one that maximizes the rate of return.
So you see, it is like the classic solvers, but the classic solvers expect deterministic input. Here, the particularity, that’s why you need a stochastic solver, is that your inputs are probabilistic in nature.
The way Lokad approaches that is that the probabilistic modeling is bespoke, the economic recipe is kind of bespoke, and the stochastic solver is also kind of bespoke. But at every stage we have a technology that makes the rollout of the bespoke recipe very tight in terms of number of lines of code and very straightforward, very robust.
So we cannot dodge the programmatic aspect. It will be present for each one of those three stages. But if you have the right programming languages, the right programming paradigms, it can be made very, very tight so that your software asset that represents the core of your supply chain doesn’t devolve into a monster with hundreds of thousands of lines of code that nobody really understands, that nobody can really maintain.
Conor Doherty: So we use computers, essentially. That’s, in a sentence.
Joannes Vermorel: Yes, but spreadsheets are also computers. So it’s using smart computers. No, the big point is not bespoke. It’s just using, again, computers. A spreadsheet is a programming paradigm. It’s a way to represent the data and it’s a way to model the logic on top of your data.
So if your instruments are not fit for the task, you will struggle immensely. For example, it is absolutely possible to implement Tetris, the video game, in Excel. But it’s super not convenient. It’s way easier to do it in Python. But if you really are up for the challenge, it is absolutely feasible to implement Tetris in Excel.
So again, let’s take the right tool for the right purpose. And what I’m saying is that as soon as you have probabilities entering the picture, unfortunately spreadsheets have to be dismissed because they are just not going to be a good fit for this class of objects.
Conor Doherty: We’ve been going for an hour. I still have the energy to keep going and answer the questions. I assume you do as well, so we’ll push on.
I hope I’m pronouncing this correctly. So this is a couple of comments that I’ve compressed into a question from Sophian, I think. Forgive me if I’m mispronouncing. So a couple of comments. I’m compressing them into a question, which is: considering companies do track lead times, or most do, why aren’t historical lead time gaps enough to trigger action once they start creating economic impact? So is the real blocker tools, theory, or lack of ownership for lead times?
Joannes Vermorel: I mean, first, the problem is that companies track lead times, that is true. But again, the whole software stack has a workflow attitude to it. So they see the past lead times that are recorded, but there is no modeling layer and there is nothing really to exploit those probabilistic models that could result from this modeling.
Varying lead times modeled probabilistically are useless unless you can combine that with all the other sources of uncertainty. Just modeling the variations of lead times in isolation is useless because it’s not actionable. So you need to have the end-to-end recipe that goes up to the resource allocation.
So the reason why companies usually don’t do anything about that is because even if they were to actually do a modeling of the lead time variation, all the rest is still missing. And when I say “all the rest,” I mean the probabilistic modeling of the other sources of uncertainty, the economic recipe, and then the stochastic solver. The package is still completely missing.
Thus, companies just don’t really start, because on its own, in isolation, modeling the lead time is not conducive to really anything unless you have the rest of the package.
Conor Doherty: Side note, I did just receive a DM. “Did Joannes just say Tetris can be programmed in Excel? What the hell was that?” You’re being serious, just for the record. Tetris.
Joannes Vermorel: Tetris, yeah, absolutely.
Conor Doherty: So you didn’t misspeak, just to be clear.
Joannes Vermorel: No. I mean, I’m sure there are sources online. Excel is a Turing-complete programming language. It can do fantastic things, crazy things, and things that you can do, that you will find plenty of examples of online.
Conor Doherty: Alright. Well, let the record show, when people DM me, I do ask. Alright, this next question is from Jonathan. The comment, excuse me, comment, but give me your thoughts on it.
“Unattended might be the easy part. I don’t think the accumulation of suboptimal decisions is a tech problem. It’s that nobody owns the loss function. Every team optimizes its own metric. A bigger challenge would be to get a big organization to agree on a single economic objective before automating anything.”
So?
Joannes Vermorel: Yes, mostly yes, but also I would push a little bit back on some nuances within that.
First, the problem with supply chain is that it is a bureaucratic effort by design. I had, a long time ago, pushed that point. What do I mean by that? I mean that this sort of planning and resource-allocation game has to be specialized within the companies. The problem is that you’re growing people who are not connected with the clients or the production. It is very much a bureaucratic effort inside the company to do this decision-making process.
And so there is a danger with all bureaucratic undertakings: they tend to grow. Keeping them under control is very difficult. So whenever people suggest, “We need to create a new team,” I say, “Up, up, up, up, up.” Most companies have way too many people in supply chain. Again, just to give you a scale, at Lokad we have supply chain scientists where one person is managing €1 billion or $1 billion worth of inventory. One highly talented, skilled person, not interns, to be clear. But that’s the sort of productivity we’re talking about. And usually, when we do that, on the client side they have hundreds of people. So way, way, way too much.
So I would say, before adding yet another layer, we need to think subtractive rather than additive. What can we do to remove people as opposed to add even more people?
Now, when you say the loss function has to be owned by someone or an entity or something, that’s where I slightly disagree. My take is that we need to decompose this loss function into business drivers, economic drivers, and each economic driver already has a clear owner. The cost of a stockout should be owned by sales. They are clearly the ones that can say, if we lose, if we are not on time and in full with this client, it costs us that much. They are the ones that should be able to own that. The cost of money should definitely be owned by finance. They are the ones that know how much it costs to borrow from the bank, what financing options are available, etc.
So what I’m showing here is that you should not take the loss function as one object. You should decompose that piecewise into economic factors. Most of the economic factors are already owned implicitly by divisions. So let’s make this ownership explicit. We don’t need to add any extra division.
Then at the end of the day, supply chain is just responsible, has ownership, for putting it together. But again, it’s a very, very modest layer of ownership because it’s just about bringing those ten economic factors together into a formula that makes sense. You are not steering the whole formula, you’re just steering the packaging of the formula, taking those economic drivers together.
Then supply chain will actually provide feedback to the various departments if those economic drivers are bogus, not aligned, or generate unintentional detrimental consequences. So I would not create a new entity. If I shorten it: no new entity. Supply chain is enough. Supply chain owns the packaging of the loss function, but not its components. The components are actually owned by the already existing departments.
Because this ownership is very light, we are talking about something like a few hours per year, even for a large company, this work of packaging the thing. So it’s very, very thin. It doesn’t require a new team or a new setup. The existing organization will be just fine.
Conor Doherty: Actually, speaking of detrimental consequences, it leads to the next question, which is a short one, but it does highlight something we haven’t actually touched on yet. And again, quote, this was sent privately to me ahead of time. This person is very familiar with Lokad, by the way.
“So everything that we’ve said about lead times seems very relevant for imports or long lead times. Does what you’re saying matter as much for short lead-time businesses? So for example, if my lead times are typically two to five days, not ten to fifteen weeks, to what degree does everything still hold?”
Joannes Vermorel: Depends on the economics. Yes, depends on the economics.
But usually the thing is that when your lead times are very short, it means that down the road your customers also expect very short lead times. Things tend to cascade a lot more. If you have ten-week lead times, then your clients probably don’t expect things overnight, and you have a lot of leeway if you’re importers. The sort of dynamics are not the same.
For example, if you are in a very, very tight supply chain where you expect fresh foods to be delivered every day to have your stores stocked, a one-day delay can be massively more impacting than a one-day delay on something that you ordered ten weeks ago from China.
So the modeling is important essentially?
Conor Doherty: Yes, yes, yes.
Joannes Vermorel: So my take is that, again, to know if properly dealing with your lead times will be economically beneficial, without doing the whole economic calculations as a litmus test, I would say just have a look at the amount of firefighting, the ambient firefighting. How much, if you were to say everybody that is dealing with replenishment, warehousing, planning, forecasting, monitoring the prices, talking either to the suppliers to put pressure on them or talking to the clients to appease them when you’re actually late, put all those people together and think: how much of the hours of those people are actually spent building something that is accretive, capitalistic, and how much of the hours of those people are just consumed every day to keep goods flowing through the company?
Just think of the ratio. If you have more than 20% of the effort going into firefighting, then you have a massive upside. It means that people are extensively distracted by things that should remain very exceptional.
Again, is it okay to do 20% of your time firefighting, or even 50%, the next day after something like the Strait of Hormuz? Yeah, probably. But again, those exceptional situations don’t happen every day. And if they do, then they are the new normal, and you need to robotize around this new variability, which has become the new normal.
Conor Doherty: Essentially and succinctly, the financial importance of everything we’re describing is contingent upon the financial impact of delays. That’s it.
Joannes Vermorel: Exactly. It has to be modeled. If it’s negligible or nothing, then you don’t need to think about it. But it’s unlikely that that is the case. Again, you have some magical businesses, for example let’s say Rolex, where they manage scarcity. “Yeah, you want a watch? Well, that’s going to be one year.” Okay. But that’s companies that spent literally a century building their brand. There are probably, top of my head, ten companies in the world that can achieve these sorts of things. A fresh-food retailer does not have that luxury, pardon the pun.
Conor Doherty: Yes, exactly.
Well, I saved this question for last because I think you still have energy for this one. It touches on one of your favorite topics. And the person who asked this is a fan, I want to be clear about that, but they’re asking from within the context of a company where this is relevant to them, so it’s very terse.
“If we start forecasting lead times probabilistically, do we still need safety stock?”
And you can see why I saved that one for last.
Joannes Vermorel: You don’t need safety stock. Again, safety stock is a class of non-economic policies. Safety stock is a way to allocate your capital without doing an economic calculation.
The idea that you could, through something that is non-economic, hit an economic target is crazy. It’s like I have a device that displays a number. It’s supposed to be the time, but the thing has nothing to do with time measurement. Why would this thing ever give me the right time?
So if we go to that, the main critique of safety stock is that it is a non-economic instrument and thus, by design, it will never give you an economically valid answer. Except by pure chance. I can take this book and decide that this book will give me the time of day, and maybe there is a time printed on a given page, but it has nothing to do with that.
That’s the sort of folly that I have with the mainstream supply chain theory, where some stuff is completely bogus and presented as if it were correct. The idea, when you think about it carefully, that a non-economic instrument can deliver anything of economic significance is extremely weird as a proposition. It would certainly need a massive, massive justification to say by which magic this thing happens.
So my take is, to connect to the discussion of the day, that the probabilistic lead-time approach is one more element that clarifies how irrelevant and obsolete safety stocks actually are. It’s not because of proper modeling of lead times that safety stocks are obsolete, but it’s just one more nail in the coffin.
Conor Doherty: Well, my last closing thought is that at the start we talked about how most people, if you ask them, “Do lead times vary?” they’ll say yes. But then you say, “In your company, do you actually forecast them probabilistically?” most people will say no.
So assume that all the people who are listening are very, very impressed and want to actually start taking the next steps, bearing in mind they are operating within a corporate environment where there’s a lack of academically forced training that teaches people about this. What is the next viable step to actually start moving in that direction, in your opinion?
Joannes Vermorel: You need to think about robotizing the decision-making process. The thing is that lead time modeling is a means to an end. It’s about getting better decisions.
And the reason, by the way, Lokad, that was more than a decade ago, the reason why we came to that, it’s because we tackled the problem of decisions and the decisions were coming out wrong. That’s like an elimination game. “Okay, we have this problem here, we tweak the modelization, okay, decisions are still wrong, we still have cases where decisions are nonsensical.”
The criterion for Lokad is that, to go to production, we need to have 0% insanity in the decisions that we generate, which are resource allocations. Insanity would be something where the economics tell you that you’re super bad. That’s what an insane decision is. It is a decision where, no matter how you fudge the economic drivers, it is just very wrong.
And here, improper lead-time modeling popped up as one of the key factors generating insane decisions, and thus we had to do it.
But ultimately, don’t try to do lead-time modeling in isolation. It is pointless. You need to take it from the angle of, “I am generating decisions and game decisions, and here I need to assess, in my modelization, what are the blind spots? Where do I keep generating insane decisions while I should not, because my modelization is simplistic?”
Maybe you’re one of the tiny, tiny few lucky businesses where you could say fixed lead time is very incorrect, but it is sufficient to have non-insane decisions. This is not really the game at play. It can happen, rare, but it can happen.
So my take is that your modeling efforts need to be driven by those insane decisions. You want to go to production and thus eliminate the busy work and the firefighting. When you have something that generates decisions with 0% insanity, where every single decision is quite good, not optimal, I have no clue about optimality, but when every single decision can be reviewed by colleagues and they say, “This is solid. I would maybe not do it exactly like that, but if it was a colleague saying that, I would say it’s green to me either,” that’s the vibe.
You want something where every single decision gets that reaction. Bad lead-time modeling will just completely undermine that, with maybe up to 20% of your decisions that are like, “What is this decision? This is completely nuts. We need to do something differently.” Usually that’s what you want to fix. You want to have all the decisions coming out from your numerical recipes that are green, that say, “I’m roughly correct. This is sane.”
Conor Doherty: Alright. Well, I have nothing to add. I’m out of questions, and you have been going for almost 80 minutes, so I think we’re out of time. Thank you very much, as always, for your insight.
And to everyone for watching, thank you for attending, thank you for your questions, thank you for your DMs, thank you for your comments. If you want to continue the conversation, as I always say, you’re already on LinkedIn. Click on Joannes’ and my profiles. Start talking to us. We’re lovely. We love to talk about these things.
And on that note, we’ll see you in a couple of weeks, and we’ll be talking about pricing in supply chain. But I have nothing else to say except: get back to work.
