00:00:00 ABC分析の隠れたコスト
00:04:19 販売量による SKU の ABC segment
00:08:38 注意の優先順位と粗い service target
00:12:57 ABC は任意の class と bureaucracy を生む
00:17:16 computer は bucket ではなく高解像度 ranking を可能にする
00:21:35 bucket は情報を大きく圧縮し失わせる
00:25:54 audience Q&A 開始
00:30:13 経営層は dollar で定量化された影響を好む
00:34:32 ABC ではなく単純な economic ROI を使う
00:38:51 市場は modern granular methods を有利にする
00:43:10 criticality は sales classification を上回る
00:47:29 pharma には product-specific criticality と rule が必要
00:51:48 ABC は cosmetic reporting veneer として残るかもしれない
00:56:07 診断 segment はよいが compression は避ける
01:00:26 高解像度 data を保ち、decision 時だけ畳み込む
01:04:45 ABC bucket ではなく Pareto curve を描く
01:09:04 最終判断：ABC は decision engine に不適
概要
ABC分析は古く粗い近道で、SKU を bucket に入れて重要な経済情報を落とします。現代の計算機なら item を連続的に ranking し、在庫判断を holding cost、stockout penalty、rush repair、margin impact などの金額で評価できます。ABC は legacy report や診断として扱い、意思決定 rule にはしないべきです。
詳細概要
ABC分析は tidy に見えますが、任意の bucket と情報喪失を生みます。A、B、C の境界は作り物であり、多数の SKU を少数 class に押し込むと、利益と損失を決める個別性が消えます。
経営層が必要とするのは ABC ではなく、margin、working capital、carrying cost、stockout penalty、dead inventory を金額で表したものです。航空宇宙では C 部品が aircraft を止め、retail では小さな欠品が basket 全体を失わせます。必要なのは modern computing による granular で economic な decision engine です。
全文書き起こし
Conor Doherty: Is supply chain breakdown. Welcome back, and today we will be breaking down the hidden costs of our very old friend ABC analysis. Now you know who I am. I’m Conor, marketing director here at Lokad. To my left as always, the insouciant Lokad founder Joannes Vermorel.
Now before we get started, let us know down below. Do you use ABC analysis? Do you think it has limitations? Does it have a place in the year of 2026, the age of AI? Get your questions in. I will be manning the chat. I actually have two consoles open in front of me, one with the questions, one with the live stream.
Our producer’s on vacation, so you’re talking directly to me today. Any questions you have, I will post to Joannes in about 20 minutes. Now, on that note, given the nature of this topic, it’s evergreen. Everyone cares about it. I already have at least seven or eight questions that have been sent in to me. So we’re going to try and set the table, cover some of the main ground, and then get to the audience questions as quickly as possible.
So, yeah, if you want your question answered today, try to get it in promptly. If not, anything we don’t answer today, we will cover in a post or we’ll tag you with a response comment. So, anyway, on that note, Joannes, ABC analysis. Now, I know you’re a huge fan of it going back eight maybe 10 years. I think you founded Lokad on the basis that this is your favorite metric. Is that fair to say?
Joannes Vermorel: No. Short answer, no.
Conor Doherty: Short answer, no. Thank you for joining us and we’ll see you. No, go ahead. What are your thoughts?
Joannes Vermorel: ABC analysis, which has nothing to do by the way with activity-based costing in supply chain. ABC really refers to the letters of the alphabet, A, B, C, D and whatever. The method just consists of a segmentation of your products or SKUs based on recent sales volume or flow volume.
That’s pretty much it. So you create an A class, that’s the most, I would say, most active items or SKUs, a B class slightly less, C class, you see etc., etc., and companies who use the system will have like let’s say up to half a dozen of classes.
Why? So the why, this answer has essentially, there are two reasons why companies do that. The first one is to prioritize the attention of the supply chain practitioners. So fundamentally what you’re saying is that if something is rotating very fast, there is a lot of business going on, you want your people to spend more time on those SKUs so that they are never out of sight. You see, that’s one idea.
So that’s one class of concern. I would say it’s the bulk of the concern. Then you have a second reason which is you can get some kind of a slightly, I would say you can have a proxy of quality of service that is slightly more refined than just having like average service level where you would say we have the average service level per ABC class.
So it’s not like massively refined, but instead of just lumping everything into one number, you will expand that into half a dozen of numbers reflecting your classes. And that can give you a sense of where you stand service-wise, but beware, we’re talking service-level-wise on each one of those classes. That is, I would say, the reason why it’s used and that’s pretty much the extent of the usefulness of ABC analysis.
Conor Doherty: Well, that sort of brings us to the first kind of fork in the road because when you talk to people about ABC analysis, there will be people, or let’s say just companies to make it simpler, who say, “Yeah, we use it as a reporting metric at an executive level.” And then there will be others who say, “Yeah, we use it for that plus we use it to genuinely set sort of shifting service levels for what we decide to be A class, B class, and C class SKUs.”
And again, you can add as many classifications as you wish. So there are two points there, two nodes to cover. First is its utility or its usefulness as just a reporting snapshot mechanism and then there is the decision-making engine that some people might use it as. In turn, what are your thoughts there?
Joannes Vermorel: So the bigger concern that I have with ABC analysis is that this thing, very frequently when it is used, it is not one metric or one subset of metrics among hundreds. No, it is usually put on a pedestal and as a consequence the whole organization is kind of shaped around this idea. And that’s really, I would say, if I have one concern with ABC analysis, it’s that it lends itself to completely pointless bureaucratic undertakings from my perspective.
You see, and again, most of the other metrics, even plenty of other, I would say any kind of just plain statistical description of the flow, the same thing could in principle happen, but what I observe is it does not. I mean ABC has this special unfortunate quality that it will really shape the organization. So you will see, for example, we need to have planners that specialize in A or B or C classes, or we have our planners that we need to have a mix of A, B and C.
And then we need to have our targets expressed in A, B and C and we need to have our financial budgets expressed in A, B and C. You see, it goes way beyond. And the problem with all of that is that it’s completely made up because the boundaries between those A, B, and C are completely arbitrary. And what I reject is that if you assume that your supply chain doesn’t have computers, then this super low-resolution discretization kind of makes sense because anything else will be just way too heavy.
But as soon as you have spreadsheets, it doesn’t make sense. You can, for example, if you want to prioritize your SKUs, you can sort them from whatever is your criteria, most important to least important. You don’t need to create buckets of a granularity of five classes. It adds nothing.
Conor Doherty: In fact, again, so this is why I wanted to separate the two because some people again would argue somewhat reasonably that if you do want to get a temperature check of how things are performing, you do want to know, well broadly and directionally speaking, where’s the bulk of the economic action occurring? What are the best-selling items or SKUs that I have?
Joannes Vermorel: So, I mean, first it doesn’t answer this question. You know, let’s be clear. ABC does not answer any question. That’s the thing again. Let’s take a basic situation. You have like 100,000 SKUs.
Conor Doherty: Yes.
Joannes Vermorel: And five classes.
That’s, I mean, so literally the math says that maybe your A class is going to be 2,000 items, and in that you’re going to have everything and the rest. That’s already a lot. So you see, by definition, if you try to project 100,000 items into five classes, that’s going to be immensely lossy information-wise. So it’s not going to give you any accurate picture of anything because those classes are themselves super, super large.
And again, that’s going to be, let’s say, 2,000 items. That’s going to be 10,000 B, 20,000 C, and all the rest D. You’re mixing enormous amounts of things. Those segments are not clean. You will have everything and the rest.
And that’s why I object to that. It’s a construct. It’s not real. It’s completely made up. The boundaries between your classes are just cherry-picked. You know, it’s whatever. And it creates, and then the problem is that then the organizations themselves have a tendency to engineer processes, workflows, guidelines, even literally the organigram, the hierarchy of the people, are going to be reflecting that and I say, what I am saying is that this is complete nonsense.
It’s, again, like any descriptive statistics, you can have that with hundreds, but anything that would put that on a pedestal, I really reject categorically. It is a mistake and this mistake will turn out to be very costly because it’s distraction and also because if you see value in ABC, that means that you have a very, very, I would say, important critical misunderstanding of what a computer would bring to the table.
So you see, it’s more like a litmus test: are you fundamentally misusing computers for your supply chain? And if you’re misusing them, then we need to address this big problem and then you will see once down the road, once we have addressed this massive concern, then it will become self-evident that ABC is just, it’s just mostly a nothing burger.
Conor Doherty: Okay. So again, just to push back a little bit there because again, if you say an ABC analysis provides no insight, but someone would say, well look, realistically in terms of the performance of my company, yeah about approximately 80% of my revenue is coming from about 20% of my catalog. Now, what decision you make with that is slightly different. Are you questioning that that has any relevancy whatsoever? Like that insight has no value.
Joannes Vermorel: First, you are picking, you see the point is I’m saying you’re picking the 20%. I say yes, excellent. Why not 25? Why not 30? Why not plot the Pareto curve? And I say plot the Pareto curve. Excel will let you do that in five minutes. You can have the complete Pareto curve.
And the Pareto curve is very interesting. It will essentially give you a thing where if you grow in terms of percentage of SKUs or products it will converge to 100% very quickly and it’s very visual and you have all the information. And you can cherry pick a few points along the curve just to make it more insightful.
I’m just saying that if you have the curve plus a few points that are cherry-picked for readability of the graph, you’re making it very obvious that those points are cherry-picked. This is not, this is, you see, that’s where ABC is not used like that. If it was just used like, again, you have a curve and you cherry pick a few points, I would say yeah, no problem, it’s just like any of the hundreds of other potential metrics.
The problem is that it’s put on a pedestal, that suddenly this A class is going to, it gets a name, it gets processes, it will get dedicated software logic, you see, it’s going to be a lot of stuff. And I have frequently visited prospects where if I were to assess the amount of software code that was induced by those completely made-up categorizations, we are talking of stuff that is in the tens of thousands of lines of code.
And this is very important for something that is completely made up. And then also it is the way it gets used completely abuses what you can do with those classes. Because again, if you rank all your products from the most sold to the least sold, you will see that you have a curve. Yes, it makes complete sense.
But if you try to segment that, the stuff that you’re putting together doesn’t belong together. They might reflect completely different classes of needs. You’re not representing the dependencies that you have. So you see, the problem with this segmentation is that again, because it is made up, it is fundamentally extremely arbitrary. It doesn’t match anything that is tangible for your flow. It doesn’t match anything that is tangible even in the mind of your customers.
Conor Doherty: There’s one point here. Okay, I agree. I should say there was an unusual amount of activity here, so I’m spinning a few plates here. Yes. But there is a point that occurred to me while you were saying that which is, I want to address this at a foundational level because I think at a foundational level the actual assumption of ABC analysis is more or less something that I think you would actually challenge or agree with to a degree, which is not every SKU, because that’s the unit of analysis here, not every SKU is equally deserving of attention. Would you agree with that?
Joannes Vermorel: In your catalog, so you have, mostly no. Mostly no.
Conor Doherty: That’s a more foundational point to actually address.
Joannes Vermorel: I mean again, that’s computers. Why? Because yes, in theory, more attention in the computer world would say you need to spend more CPU on a SKU rather than another. The reality is that it’s not the way it works in modern enterprise software. I would say this is not how it operates.
You cannot really meaningfully get, I would say, a speed up for less important SKUs. I mean you will get it to some extent. For example, if you have fewer data points, that would mean that you have lighter data structures that are attached to a SKU. Yes, to some extent. But it’s not really the same thing as for a human because fundamentally computers are not starved for attention.
You know, a computer can rescan your 100 million SKUs 100 times a day. And it’s not, and you can do that, 100 million SKUs, that would be a bit large. We’re talking of Walmart sort of scale, and you can do that with the processing power of a smartphone essentially. So really, really, we have like a hyper-abundance of computing resources. So this is just not a bottleneck.
And again it’s very different from human supervision where if you have like a thousand SKUs and you have one guy in eight hours it’s just extremely difficult to survey those 10,000 SKUs in eight hours. So the person will need to prioritize a lot otherwise the stuff that is super important will never be even seen.
But again that’s a concern that is like predating the computer age. It’s not really relevant once you have computers. You see, the logic you need is to process and make economic decisions that maximize rate of return for all your SKUs, and for every SKU you consider all the relevant information. And again if you think in terms of how much information do I have per SKU the answer is typically per SKU you will have tens of kilobytes of information, even if data is sparse.
The ABC class is nothing but a shallow projection of the sales history. It tells you almost nothing. We are talking of one or two bits, not even bytes, bits of information. It’s like, again it is really, really nothing.
Conor Doherty: Just to give one last, because I have my own question, but just to give one last echo of challenges that other people have channeled through me. How do you respond to the idea that yes, you’re advocating, and most people would agree with, a financially ranked perspective on decisions? Some people would say, but that’s already what I’m doing at a lower level with ABC analysis. I’m looking at what I have available and I’m ranking by financial performance. You agree or disagree with this? I’m not saying it’s perfect. I’m saying do you really see the value?
Joannes Vermorel: If you’re doing that, why do you just not discard ABC? If you’re doing an economic prioritization, just remove ABC and it will work better.
ABC is a segmentation. Just think of it. You have an image. No, no. You have to understand it’s a lossy process. So you lose information. Just think you have an image and it’s 1,000 pixels by 1,000 pixels. So it’s a very nice high-resolution image, and now I tell you I want to transform this image into a 4x4 pixels image. It’s going to be incredibly low resolution. I mean you don’t see anything.
And you see that’s the thing, that if you just do it, and please the audience, just have a look: take an image, anything that you find on the internet, and convert it to an 8 by 8 pixel image and that will give you a sense of the information that you lose when you go for a low-dimensional projection of something that was high-dimensional. I mean everything is lost, literally you just have like a few squares, everything is lost.
And I think that ABC just gives you an illusion that there is still something, but in fact you have lost the quasi totality of the information that you started from. And so what I’m saying is that in an age of computers you need to preserve the high-resolution information that you had at the beginning. Why do you want to throw it away? Why are you so eager to super-compress your image into an 8 by 8 pixel resolution or something? It just doesn’t make sense.
If you had to do all the calculations by hand, yes. But even if you don’t have something like Lokad, you have spreadsheets and even a basic spreadsheet can do tens of millions of operations per second and that’s not even pushing it. So it’s really pointless.
Conor Doherty: All right, I’m going to push forward. I’ve now given at least fair lip service to the initial challenges. There are others that I’ll get to later, but one of the things you just said, and this is me now, my questions here. You talk about it’s a lossy process and you said it compresses. You have to understand theory of information, how much information do you have in and out. And here you have to understand that you are losing information.
You don’t, it’s again think of it, you have an image, you compress it, you can’t recover the original image. The information has been lost. Well the point that I’m going to get at is one of the pieces of one of the most critical pieces of information if you want to go down the optimization route is understanding the interdependencies within your catalog. So again, put to one side whether or not you’re talking about, well should it be 20% 25% that contribute, whatever. It’s when you make discrete parameters between classes of items in a traditional ABC analysis, do you retain the insight to see, well, what are the interactions between let’s say C classes and A classes?
Joannes Vermorel: Nothing because again your volume is completely arbitrary. So it will put a frontier in your assortment that is quasi random and that’s also something that is super baffling. Again, pick any vertical. You see, the point is that you should not reason about supply chain. One should never reason in the abstract. We always need to look at a series of verticals to see how things unfold.
And that’s how we can see that there is madness. So let’s pick any vertical. If we are looking at a plane, any part that is missing will create an AOG, you know, an AOG problem, aircraft on the ground. So you have dependencies because the aircraft is, there is a maintenance operation, and if any of the parts is missing the aircraft is stuck.
So obviously your ABC is pointless because if you need a thousand parts and one of them is missing your aircraft is stuck on the ground. So, and yes it might be a C part that is missing. Too bad, the aircraft is still grounded. So, okay, it’s kind of nonsense here.
Conor Doherty: Well, most people would add criticality to that, but then you would say correctly then you’re moving beyond the limitations of the thing that you were initially using and you’re adding again, you’re adding.
Joannes Vermorel: Criticality is super important. That’s a perspective for aviation. But you see, criticality is not made up. It’s literally there are things, there are parts where literally aircraft engineers have said, “If this part is missing the aircraft can fly. If this part is missing it will not fly.”
And you see, so it’s not like an opinion. It is a fundamental distinction. And then you have parts that are consumable, which means that once they are dismounted from the aircraft they must be thrown away, disposed. There are parts that are repairable. It means that once they are dismounted from the aircraft, they must be sent to the shop for a potential repair. Again, that’s not arbitrary.
You see, that’s the difference, are you touching something that is real, tangible, fundamental, or are you just making things up? Again, let’s pick another vertical: luxury. You have a store, you’re selling let’s say leather bags, and it turns out that 90% of your sales, give or take, are just brown and black leather bags. Why? Because it’s what sells.
But if in your store you only have brown and black, the store looks sad and unappealing. So you need to have a few bags that will be let’s say yellow, bright yellow, bright red, something that is, yes, your clients are never going to buy those, but it makes the store look good. Again, if you think with ABC you will never see these sort of things. And I can go on and on and on. Every single vertical will see all those edge cases. They always exist. They are always massive.
And what I’m saying is that if you have something that is, pick any vertical, systematically contradicted by reality, it’s just a bad idea. Don’t do it. And again, why you should not do it? Because there are alternatives that are better and simpler and more profitable. So you see, it’s a sort of thing for me that is just methods that have been preserved from a pre-computer era and they stop. The thing is that those methods like ABC stopped making sense probably like four decades and a half ago.
So you see, it was already not making sense anymore in the 90s, and here we are, again three decades later, and it’s still being used.
Conor Doherty: There’s another example. Again, we’ve mentioned it before, so I’m not going to spend too long on it, but it’s one of my favorite terms that we use here, the basket perspective. The idea that certain things, if they’re absent, let’s say a C-class item typically bought in conjunction with an A and a B-class. I walk in, the C is not there. Well, then I go somewhere else because I wanted A, B, and C. I wanted a basket. I wanted Aperol, Prosecco, and sparkling water. You don’t have Prosecco. Well, then now I can’t make a spritz, so I’m going somewhere else.
Joannes Vermorel: Yes. And again, this thing happens for pretty much all sorts of businesses. It’s extremely rare that you don’t have dependencies between your products because if you’re a retailer chances are that people come with a certain idea of what you’re selling. If you’re a producer, people expect a certain scope of products. Just like, for example, you’re selling the iPhone and you’re selling the air plugs for the iPhone and you’re selling the printer and ink cartridges or loads of things.
So it’s really, again for me, the problem with this ABC analysis is that because it creates, it has a bureaucratic momentum of its own almost invariably. It distracts the company from the things that should really matter for the flow and instead people are just, I’ve seen so many companies where people are constantly spending time and mental energy on stupid problems such as should we add another class, should we change or tweak the percentages, how can we stabilize.
Because then they start looking at, oh we have like from one quarter to the next we have 40% of the SKUs that move from one class to next. So it’s like a lot of nonsense. So they need to invent stuff to mitigate this problem. But you see, you end up having like a whole set of processes, software, workflows that are just induced by this initial direction of putting the ABC class on the pedestal. And if you just remove that all your problems are gone.
Conor Doherty: All right. Well, Joannes, as you know, I’m convinced. But what I’m going to do is, because there are just too many questions now, I’m going to immediately switch over.
We have an old audience of people who are proponents of, just so people understand, there’s live questions, that are open like comments, there’s DMs, and I also have questions from earlier. And we’ve already gone for 30 minutes. So I’m just going to go straight to the audience because that’s why we’re here. We’re here at your discretion.
I’m going to start with Nicholas’s question because it actually summarizes something I was going to say later and it kind of amalgamates a couple of concerns from other people. So, Nicholas, friend of the channel: how can practitioners persuade executives not to accept ABC analysis as the default consulting answer given that ABC is one-dimensional and can misguide service, inventory, production, and shipping decisions?
Joannes Vermorel: So first there is this misconception that the top management is interested in ABC. I’ve never seen that. For me, ABC analysis is something that is only of interest to supply chain middle management. You know, it doesn’t interest people at the very bottom because people at the very bottom, they have a list of SKUs and they need to manage all of them.
You know, it doesn’t matter if your boss decided that you would get 200 SKUs that are A, 200 SKUs that are B, whatever. At the end, we expect you to have solid, reasonable production decisions or replenishment decisions for your entire scope. Even if supposedly the C items are less important, in practice, people will complain a lot if you just neglect those. So, okay.
So, people at the very bottom who are dealing with the disaggregated decisions, for them they don’t really care. It doesn’t reflect the resolution at which they operate anyway. And the people at the very top, they don’t care either because in fact what they care is macroeconomics, you know.
Okay. I’ve never seen really a board super interested in the service level per ABC and whatever. The question that they would be interested in is give me in dollars how much money is left on the table through poor quality of service. You know, give me this number and then tell me if I add 10 more million dollars into, let’s say, working capital for more inventory and more staff, how this picture would change.
That’s the top management sort of questions. Again, they don’t really care about ABC. What they want is something that would give them closer to dollars. And very frequently, the frustration of top management is that when they talk to middle managers in supply chain, they only get percentages.
That’s something that I very frequently mention in this channel, is that the mainstream supply chain theory is only about percentages. So it is extremely difficult for the top management on average, for most companies, to extract anything that would be dollar-related from their supply chain department because the supply chain department gives them just a bunch of completely useless percentage-based numbers.
So ABC, by focusing a little bit on the mass, gives to the management some kind of a way to get closer to something that would be expressed in dollars. But what I say is, if you directly give them things expressed in dollars, they just love it and they don’t care in the slightest about ABC.
Because you see, if you tell them, instead of having your ABC classes with your percentage and service level, they don’t care, they don’t care, they don’t care. If you tell them, here is the bulk of the projected margin that we are going to generate, here is the working capital, and here is the projected dead inventory cost that we will suffer because again the long tail is what it is, and if you extend the long tail to capture let’s say 10 more millions of sales, here is the delta in dollars that you will get in terms of dead stock, etc.
So you see, if you just translate, instead of giving half a dozen numbers that would be ABC with percentages, just give half a dozen numbers directly in dollars decomposing the problems along the dimensions that make sense, where it’s like dollars of margin, dollars of carrying cost, dollars of working capital, dollars of negative impact for missing the quality of service. I can promise you I’ve never met any C-level executive who complains that the numbers are too financial.
That, you know, “Oh, you give me a rate of return. No, no, no, no, I would prefer to just have a percentage. Thank you.” I’ve never seen that. It’s always, if you give real financial money quantified in dollars or euros, for them it’s the best. And the problem is that usually they’re completely frustrated because their teams only come up with tons of percentages that are completely disconnected from the bottom line, the profitability of the company.
So the P&L essentially, the P&L view is completely lost. And so the top management is trying to negotiate with their own middle management team so that they can get something that would at least be bridgeable with a P&L somehow. But it’s not. Again, the lesson is, no, your CEO doesn’t care about your ABC, really. And I’m sorry if there are supply chain people who really think that the C-levels care about ABC analysis. My experience is that really they don’t.
Conor Doherty: All right, this one was submitted ahead of time, but it does build on what you said. I will read, for time I’m going to cut down a little bit. Begins: I don’t have a very specific question about ABC that has not already been discussed. The obvious alternative I’m sure he would say would be something closer to ROI or economic optimization, but in practice that can be harder to implement.
So my question is, what is the simple but not wrong abstraction that executives can use for reporting, governance and incentives? Slightly different there, because even if you’re going to rank by ROI you still do have reporting. Like, what is the discretization? Like what are the times at which you report? What does that look like operationally, I guess, is what this question is driving at.
Joannes Vermorel: I mean, yes, to go into dollars or euros is slightly more complicated. Yes, because ABC analysis is literally something that, formula-wise, it’s just sorting your items and just taking slices. So if I were to say in terms of number of lines of code in Python, that’s what, three lines? I’m generous. You can make it a one-liner.
So this is super trivial. I mean the bar, I mean let’s be real, the bar for 21st century supply chain should not be three lines of code. “Oh gosh, it’s too much.” I mean we are talking of a company that operates a supply chain. By default we’re talking of $50 million plus annual revenue. I mean that’s what we’re talking about. This is not like a mom and pop tiny shop around the corner. This is a sizable business.
The guy who is running the accounting of the company is doing very complicated things. So you see, I’m just saying that this is not a reasonable argument. That would be like saying, you know, driving a truck requires a driving license that is more demanding than a car. “We can’t do that.” I mean yes you can. Definitively you can.
You see, there is this idea that if something is like, I mean again, ABC analysis is really primary school level. I’m pretty sure that if I take a child eight years old and give me two hours, this child will, with an Excel spreadsheet, manage to do an ABC analysis. This is this level. It’s not reasonable to say that that is the final frontier for 21st century companies who operate at the scale of tens of millions of dollars. This is nonsense.
So what I’m saying is, back to economic analysis. Yes, it is more complicated, but it can be. More complicated than something simple, yes, something hyper simplistic. Yeah, exactly. That’s bad. Is it complicated in the absolute? That’s what I mean. Oh, so it’s a relative statement.
I mean, again, you can just do back-of-the-envelope economic calculations. I mean you can approximate, doesn’t matter. I mean if you’ve ever talked to C-levels they would not criticize you for doing approximations. This is not like double-entry accounting where if you get something wrong by one cent then you have an imbalance in your bookkeeping.
Here what matters is to be directionally correct. If you have, for example, let’s say the quality of service, you would need to find some heuristic to convert that into dollars. If you’re, you know, if you are only like 20, 30% inaccurate and instead of being like 10,000% inaccurate, that’s good. Nobody expects for, if you quantify quality of service in dollars, to be 1% accurate because ultimately this indicator lives in the head of the customers.
But nevertheless, you want to make something that is a grounded assumption and then give a number that is kind of making sense. And when you compare your figures, working capital, gross margin, stockout penalty or, you know, quality of service penalty generated by you not being on time and in full, you have figures that kind of make sense.
And again the idea is that for the higher-ups it’s just to give them something that is back-of-the-envelope level that is a little bit actionable at a strategic level, and it’s still very simple. I mean if you do back-of-the-envelope calculations we are not talking of three lines of Python. We are maybe talking of 15 lines of Python, and when I say Python it would be 15 lines in a spreadsheet too.
It’s not really that complicated. That’s why for me the bar is unreasonable. You cannot just have a bar that says I need to do a fixed limited number of addition, subtraction, multiplication and this number needs to be less than five. It’s really unreasonable.
Conor Doherty: Well, this actually lends itself nicely to the follow-up which is from Galad. Again, good to see you again, it’s been a while. It actually follows on, I guess, to the sort of the barrier to entry to what you’re describing. So he says, how do you bridge the gap between the analytical methods that are technically standard, for example let’s say ABC analysis, though you might dispute the use of analytical, doesn’t matter, standard terminology, how do we bridge the gap between that and organizations that might lack the statistical literacy that you’re talking about here?
Joannes Vermorel: First, it’s not about statistics. That’s the problem. Forget the statistics. This is not about statistical sophistication. It’s about are you doing something that is pointless bureaucratic work, or something that is actionable business-wise. I’m saying ABC is firmly in the camp of pointless bureaucratic nonsense.
Conor Doherty: But we’re being asked about the other camp now. So we don’t need to retread that ground.
Joannes Vermorel: So okay. So now, you see, those standard methods, it’s just again, this is nonsense. I mean yes, it is widespread, but just like alchemy was widespread before you had chemistry. But it didn’t make alchemy better. It just means that it was accepted nonsense. So my take is that just do the, you just have to start expressing your figures in dollars. Just do that and even loosely is sufficient to get started.
And again, what is the stock costing you think? And how much, my stock, how much working capital is at stake? And how much margin do I generate for this flow? And if deliveries are late, how much does that hurt you? Yes, again, those basic questions. If you get them, again it’s about getting them approximately right.
And then I can guarantee you all the people in the yard above you will be super pleased because literally, for them, I reiterate, the top management, my experience has been, for them it is immense pain to have this mainstream supply chain middle management that just keeps them fed with numbers that are completely unactionable, that are just walls of percentages that don’t translate into anything of significance business-wise. So just make it significant business-wise.
And by the way, that also at a very basic level explains why the pay is higher in sales compared to supply chain. I can guarantee you for the audience, just do this experiment: see how sales people report their stuff to the C-level and see how supply chain people report their stuff to the C-level. You will see salespeople, they don’t say, “Oh, we know on our A class of clients, we had an 87th outreach success and then on the B class of clients, it was only 67.”
No, no, no, no, no. The sales guy would say, “You know what? For last quarter, for the big guys, we increased revenue by 10 million. You know, that’s 10 million of quarterly revenue. We locked that in.” And it will be immediately comprehensible. Everybody says, “Oh, yes. Yeah, incredible.” And, “On this revenue, our projected gross margin is like $1.2 million.” Okay. Excellent.
You see, just look at what the sales people are doing when they communicate on their figures and you will see they maybe have some segmentation of saying big accounts, small accounts, midsize accounts, etc., etc. But when it comes to convey the message and negotiate their bonus and package, they would just speak in euros and dollars.
And what I’m saying is supply chain should be speaking the same language because again, inventory costs a real amount of money. You need to produce the right stuff. Not having the stuff at hand costs real money, etc., etc. So my suggestion is just do this back-of-the-envelope and just look at what the sales, if you’re afraid of the sophistication of economic modeling, just look at what sales people are doing. You will see that it’s not, if you think that sales people are using advanced math, I think you’re completely delusional. They are using extremely, extremely basic math. It’s okay. It’s okay.
And that’s what I’m saying about this ABC analysis is that fundamentally, for me this is a broken paradigm that semi-accidentally became popular. It’s not good.
Conor Doherty: I think we need a paradigmatic shift, do you?
Joannes Vermorel: Yes, exactly, that’s a grand word, but if people just mirror what sales people are doing in terms of communication inside their company, they will be fine.
Conor Doherty: All right, well I do want to push on because there were some other questions that actually touch on specific verticals. So this is from earlier: “Aerospace guy here.” You spoke earlier about aerospace. We have many low-volume spare parts that look like C items on paper, but some are operationally critical. How does your approach, what you’re advocating, handle that situation without creating endless exceptions?
Joannes Vermorel: But that’s the point: ABC creates endless exceptions.
Conor Doherty: And he’s saying how do you handle that?
Joannes Vermorel: So us, the way we look is we think, what is the decision at stake? So the decision, let’s take a simple case, would be: do I keep in stock 0, 1, 2, 3, four parts for my unscheduled repairs? Because if they are scheduled, it’s slightly different. So let’s talk about the unscheduled repairs.
Okay. Now, if I have my probabilistic view of the future I can assess, okay, I take the option of having just one in stock. What is the cost projected for one year and what will be the cost when the projected demand exceeds this one unit in stock? And we have asymmetries here because having the one unit in stock has a certain cost and then not having more than one has also a cost.
This can be computed roughly. It’s difficult, but you have, again, you can use, for example, some of our clients, they would just use an indicator, say an A320 being stuck for one day: €250,000 per day. Okay, that’s, you know, quarter million, that’s rough, but it’s in the order of magnitude. Obviously if your A320 is stuck in Dubai on December 31st and the only rooms for your passengers that are left are five-star hotels, your cost will be ranging in the millions.
Okay. So yes, you have, but you see, you need to just do a basic analysis, a basic assumption so that at least you get something that is not completely nonsense. You see, you have those asymmetries which is, okay, on one side what are the costs and what is the upside, not having those penalties, and you can do that for every single SKU because it’s similar sort of analysis.
And then what you will get is just the rate of return for every unit. So you will have the rate of return for the first unit, for the second unit, for the third unit, etc., etc. It’s not like super advanced math. I mean, you can make it more advanced if you want to have a very fancy probabilistic modeling. You can make it very fancy economically if you want to start factoring the fact that if you buy parts in bulk you can get a discount, you know, with price breaks.
Or that it’s aerospace: if you ask for parts many months in advance you will get a better price from the OEM than if you pay for the super fast delivery, etc. But again, those are refined adjustments on the numerical recipe. You can start with something that is very, very simple, where it’s just a basic, you know, Poisson estimation for the future demand and just a few parameters like free: cost of the part, cost of not having the part, and carrying cost, and that’s it.
Again, the most elementary modeling has like four parameters. It’s economic modeling. It is still very, very simple.
Conor Doherty: All right. I’m going to push forward because we’ve covered aerospace a couple of times now. I already know what the answer will be, but you’ll get to apply it to a new vertical. So this is from Pujit, I hope I’m pronouncing that correctly. Is ABC analysis the right method for use with regards medical products or is there any other method that it could be combined with to improve it? So again, just a pharmaceutical perspective.
Joannes Vermorel: As a firm, just remove ABC. Everything will be better. Whatever you’re doing, just remove ABC, it will be better. It’s very strange, but you see, for me, that’s the question. I’m sorry to go on a tangent, but it’s literally as if people were asking, “But how do we improve a fax machine?” And you could improve a fax machine, yeah, you probably shouldn’t.
You can improve a fax machine, but fundamentally it’s broken. Just do emails and don’t try to get fancy. Because yes, you can make a fax machine scan faster, have a higher quality impression, and you can have standards for this, for that, but go faster? You are stuck with something that is fundamentally a dead end. So just, just—
Conor Doherty: The upside is limited, I think would be a better way to put it, a more charitable way to put it. You’re stuck in a modality that has limited upside.
Joannes Vermorel: So again, if we are with, that would be like drug-related, you know, pharmaceuticals. The real criteria are, first: is it, are we talking of life-saving drugs or just something that is convenient? Again, we need to start going into the fine print of the products. If it’s like, let’s say, I have a patented anti-cancer treatment. I am the only company on earth who can meaningfully improve the survival rate of a class of patients for a certain class of cancer, then not having the product in stock is going to be super bad because it’s literally you are failing on your promise to improve the odds of survival of those people.
Now on the other end of the spectrum, if essentially you’re selling ointments for convenience and it’s a little bit undifferentiated and there are like 500 other companies who are doing kind of the same thing and you’re very cost-driven, it’s a completely different sort of problem. So you see, criticality is important again, to come back to patients.
And then this thing can be layered with, for example, you might have also legal commitments that country by country, it varies, government, but for example you might have signed an agreement with a legislator that you need to have a specific quality of service even if you are essentially selling paracetamol. So again, we need to look into the specifics.
What I’m just saying is that all those specifics just do not match ABC. They are very important, but none of them will be captured by ABC analysis. So what I’m saying is that if you’re in the pharmaceutical industries then we need to have a look at the things that are real, tangible, and specific and relevant to your industries, not a made-up mathematical criteria.
You see, that’s the problem, ABC is just a mathematical criteria. There is nothing real about the domain in this thing, in terms of the impact or consequence.
Conor Doherty: Yeah. I mean, just again, I can say that the A class is like all the items that represent 40% of the flow or 35.
Joannes Vermorel: Yeah.
Conor Doherty: Or 36, you know.
Joannes Vermorel: Yeah. That’s the subjectivity, just the committee-based nature of it.
Joannes Vermorel: Yeah. It’s really, really, but you see it’s subjectivity of the worst kind because I can go, for example, in luxury, and in luxury people would say, “Oh, this product, let’s say, we believe that this piece is actually what we call a masterpiece.” Okay, it’s kind of subjective because obviously there is not a very clear agreed-upon delimitation of what counts as a masterpiece and what counts as, you know, a regular piece.
But nevertheless it reflects something that is very, very tangible such as a masterpiece is typically done by a renowned craftsman, it has a distinctive style, it is done in incredibly short series, potentially it’s unique, etc., etc. You see, it’s not completely, even if it doesn’t have a definitive subjective definition, the subjectivity is bounded in a way. Here it’s just completely, it’s like bureaucratic subjectivity. You have a box to tick. It makes no sense. But whatever.
Conor Doherty: Actually, as a follow-up to that, and I should say I am slightly tweaking the ending of this question, and I’ll signal where I have tweaked it, but I’m putting, so you’re betraying the audience. I am putting some words in Galad’s mouth but I will indicate where that is. So his question was, it sounds what you’re describing is kind of like path dependency. So today supply chain executives are guys in their 50s or older that built their careers on the kind of outdated methods you’re describing.
Now is the real friction getting them aligned with more modern quantitative analysis—and my ending—or do you think that this will naturally resolve itself as, you know, decades pass and younger generations who are more sympathetic assume those roles?
Joannes Vermorel: So clearly it will resolve itself, but not in the way that people expect. It’s just that companies carrying on with outdated methods will just go bankrupt. That’s the way market works. Market works not because people improve, learn, or retire. It’s just companies go bankrupt and that is the filter that represents, you know, that capitalism is essentially a mechanism to filter incompetent companies.
And that is very tough because it means that companies that were good places, you know, with people who were really doing their best, just go out of business. And it’s sad, but it is what happens when you’re using methods that are out-competed by your peers. So now, if you want to have a look at what the future of supply chain will look like, just look no further than Amazon. Amazon is not slicing and dicing their supply chain into ABC analysis.
They are doing things that are much more granular and that’s how they can actually be performant. It makes no sense, especially just think at the scale of Amazon where they have millions and millions of products put on sale. Why would you slice and dice all of that into four, five categories? It makes zero sense and they do not.
So I think you really need to think of the fact that, do simply what’s right for the company considering what is possible with the computing hardware and computer software that is commonly available. You see, that’s not about copying your competitors because the reality is that if you look for a long period of time, multiple decades, most of your competitors will actually go to bankruptcy themselves.
So that’s the story of capitalism again. Most companies are gone. If you go, for example, 100 years ago exactly in Europe there were probably like 500 automakers on the continent and now there are like what, 10, so independent companies. So that means that we have the vast, vast majority of those companies that have essentially gone bankrupt or gone through acquisition because they were not profitable enough to not be acquired.
So what I’m saying is that here, ABC analysis, you need to kick that out for the greater good of your company. Because it is self-evident, just do it like you would do for a fax machine. If in your company you tell yourself a story that, “Oh, we will just replace the fax machine once Bob goes to retirement,” you know, Bob may still be around for 15 years.
That’s not the proper timing.
Conor Doherty: That’s long enough to destroy the company.
Joannes Vermorel: Yes. Exactly. Again, maybe your company might have a fantastic product, might have a fantastic brand. You know, if you are like Hermès, I’m pretty sure it’s such a fantastic brand that they could keep running, if they wanted to, fax machines for the next century and still manage to turn out a profit. But it’s not a reason to do so.
So you see, yes, in the supply chain you might be saved by what is done by other departments: fantastic marketing, fantastic product design, fantastic sales team that has the perfect, you’re out-competing relationship-wise your peers by being closer to the client, etc., etc. But it’s not an excuse to be completely, I would say, drowning in obsolete stuff in terms of supply chain practices.
Conor Doherty: All right, there are two more questions to go to, and one of them is very technical. You’ll be very pleased about it. Okay, I’m going to go in order here. So this one was sent earlier: is it possible to implement the kind of decision software you’re talking about and retain ABC purely as an executive reporting layer? The question continues, I like the idea, but I can’t force the exec committee to change their worldview completely. “I have to meet them in the middle.”
Joannes Vermorel: So obviously, yes. ABC is three lines of code. So yes, obviously you can keep that around if you wish to. You know, obviously. What I’m saying is that first, I really challenge, I really challenge because I’ve never met a C-level that told me, “I am in love with ABC. I need ABC.” I’ve never seen that.
Usually the feedback is much more blunt. It’s very simple. They say, “My supply chain teams feed me with hundreds of percentage-based indicators that are completely useless. Every single one of them. I am struggling to kind of get something that I can, on my own, then convert into dollars and the least worst thing that I found was this ABC thing that they seem to be willing to produce. All the rest is, and they produce hundreds of other numbers that are just complete junk. So this is the least worst thing that I can get from them.”
Now, if we go back to this sort of situation, I can tell you the executive committee will love your dollar-driven, euro indicators. They will love them. They don’t care in the slightest about those ABC things. It’s really, again, that’s a middle-management perspective of supply chain. That is not a C-level perspective.
And then, organizationally, yes, you maintaining that in terms of software is trivial, that’s not the question. What I’m saying is that the reason why, and by the way at Lokad, for example, when we implement an account, we will typically have hundreds of metrics that we just internally implement to check dozens of things. Why? Because it’s like a way to get more insight. Supply chain scientists, they produce tons and tons and tons of descriptive statistics for their own insights and it’s just fine.
I’m just saying it is fine as long as you don’t put it on a pedestal. You see, that’s my real beef with ABC. It’s not that ABC exists, it’s that invariably when it exists—I’ve seen companies, and I’ve seen dozens of companies doing that—they will turn that into a massive bureaucratic extension where suddenly you have jobs that will mention the ABC. You have bonuses that are tied to ABC. You have finance reports that are tied to ABC. You have processes and entire pieces of software that are tied to ABC.
And that’s where it’s wrong, because for example when at Lokad we have a supply chain scientist that produces those hundreds of metrics, those metrics are tied to nothing. They are disposable. The person can just create more metrics or just remove them at will, and they frequently do because it’s just about having many numerical insights into a situation.
And those things are not put on a pedestal. They are treated like disposable insights that will have a short shelf life. You create indicators, if it helps to improve a numerical recipe you keep them. Once you’ve exhausted their potential for new insights to improve the numerical recipe, you de-prioritize them. And maybe you can actually introduce better metrics that are more insightful, and then you actually purge the old ones that are just less informative, etc., etc.
Conor Doherty: All right, I’m going to close with this question. It’s from Alejandro. Regarding, I’m just going to read it as it is: regarding segmentation. This may or may not be exclusively about ABC analysis, but another way to classify assortment. So the question is mostly about classifying things in general. What is your view on segmenting demand sparsity using other methods? For example, the Syntetos–Boylan framework: smooth, erratic, intermittent, lumpy SKUs. Is that framework also fundamentally flawed or is it still useful as a diagnostic?
Joannes Vermorel: Again, as a diagnostic, it is fine to have tons of crappy metrics. It’s just fine. You know, again, let’s consider, again I would say, as a rule, you know, a supply chain scientist working at Lokad on the supply chain situation, they will probably generate more than a thousand numbers, figures, to characterize the flow. Most of them will be super low quality, disposable. Or you’re just trying to figure out if you have bugs in your implementation. A lot of things are just redundant where you will be looking at the same situation from three different angles just to make sure that it is kind of consistent, etc., etc.
So from this perspective, if you take it from a perspective that you are experimenting with metrics, and the metrics are in flux—you know, they come, they go—because again, why do I say they are in flux, it’s because unlike decisions, this is made up. Those are numerical artifacts. So the company is not committed in maintaining these metrics. They are just ideas of the mind. So those things can be implemented, removed, purged, upgraded, phased out, etc., etc.
So if you add, for example, your indicators of lumpiness and whatnot, why not? For example, let’s say you have your probabilistic modeling, but maybe this thing has a bug. You’re not sure. Because that’s the problem, when you devise probabilistic, when you’re trying to do a forecast, is that a bug will just manifest itself by a lowered accuracy. So maybe your stuff is quite accurate, state-of-the-art in general, but because there is a bug, there is like 1% of the SKUs where you are like 20% less accurate than you should be. No, I’m just numbers being picked up.
And because it’s just 1%, it doesn’t show on the macro assessment. But still, if you have a 20% gap, even if it’s 1% of your SKUs, that sounds dumb to have that, because it’s a bug. And then to characterize this bug, if you end up having indicators of lumpiness, of skewness, of this and that, it’s just fine. It’s just fine. But you see, I treat that as something that is part of the debugging tools, not the core logic.
And as a rule of thumb, anything that is a discretization of the original data is almost invariably a losing proposition because you’re losing information. So you see, you want methods, numerically, that preserve information as much as you can for as long as you can. Just think of it, you’re working on an image, and I think the image is good as an analogy.
You will see, just ask any guy who is doing design with images, they take the full-resolution image, and yes, at some point when they have to publish the image on the website they need to scale it down because you cannot publish a 20 megabyte image on the website. That would make your website impossible to browse, slow, etc.
But if you ask a guy who is working in Photoshop, I can guarantee you they work with maximum resolution for as long as they can and they will only lower resolution at the last, last moment when they want to publish. So your methods should be just like that. You should essentially preserve maximally the information and at some point you will collapse all of that into your decision.
And yes, your decision is, I reorder, or I produce zero unit, one unit, two units. And that’s your collapse of information. And you lose all the information that led to that. It’s fine because it’s, you’re not looking at it. You haven’t lost it. It still exists. You can call it. But you see, if you as an intermediate step collapse information and then try to generate the decision based on this collapsed information, this super-compressed information, you’ve just lost so much.
And what I’m saying is that it’s a misunderstanding of how computers work because there is no benefit. Your computer is not starved for computing resources. It has plenty of computing resources. So there is no reason to do this compression that is something that a human needed to do because a human cannot go through a thousand SKUs in one hour. It’s just not— even if you’re super, super good, that would be a SKU every 3 seconds.
You can’t meaningfully have an opinion, even like a guess, at this speed. Maybe you can do one SKU every second for maybe a minute, but then after a minute your mind is toast. So you can’t sustain that for one hour. So that’s what I’m saying. But the computer, it can. It has no problem doing this.
Conor Doherty: All right. Well, I’m going to wrap things up now because we’ve been going for about 70 minutes. There are some other questions, but I think I’ll address those as a follow-up article because you actually touched on ABC-XYZ. And I know you love that more than you love ABC. You were talking about mathematical moments. We’ll get to talk about mathematical moments. You love that.
But for today as a closing thought, ABC analysis as the decision-making engine, you are categorically against it. As an initial temperature check, fine if you have to. Is that fair?
Joannes Vermorel: Yeah, exactly. But even if you have to, there are better indicators. Just plot the curve. You know, the curve is, you take all your SKUs or all your products, you rank them from the one that is highest in volume according to whatever you’re measuring—if it’s volume in units or volume in dollars—you just rank them, and then you plot the curve which says how much I am getting to 100% adding one more thing.
And then you will have a curve and you will see this curve is super steep at the beginning and then slowly converges to one. Just plot the curve. This curve gives you, in terms of information, 50 times more and it’s much more visual too. You see, because if you use this curve you will even see if you have bugs in your data. That’s the reason why at Lokad we frequently plot these curves, it’s that if this curve is not completely smooth there is a bug in the data.
So it’s a very nice sanity check to, in addition, see if there is nonsense going on in your business. And it also gives you a sense of how fast do you converge to something that is quasi 100%. And that will give you a real visual sense. Because you see, people tell me it’s visual, but I say ABC is not visual, it’s crap. If I plot a graph, it is way more visual.
So you see, that’s also the only reason why the graph was not popularized is that if we go back in 1960, it requires a sort operation. So you need to sort products from the ones selling the most to the ones selling the least. This sort operation, if you have to do it manually, it’s immense pain. That’s why ABC is winning if you have to do things manually. But for your computer, sorting millions of items, it’s like nothing, really. So just plot the curve.
Conor Doherty: Okay. Well, I don’t have any further questions. Anything else would just be circling the pond, retreading worn ground. But again, I’m sure we will come back to that, probably to talk about ABC-XYZ in a later day, because again, a lot of the DMs coming through about ABC-XYZ and I’ve reframed some of the questions to address the fact that it was ABC not ABC-XYZ.
Joannes Vermorel: But the criticism is essentially similar. You know, it’s again, think in terms of images. Instead of having an 8 by 8 image of resolution, you’re just telling me that you’re going to have 16 by 16. Ah yeah, okay, four times more pixels, still completely crappy.
Conor Doherty: We’ll get into that another day, but it does actually raise an interesting point which is, and this is more broadly to the audience, if there are topics that you do want us to cover, by all means comment them down below. We are on two platforms right now, LinkedIn and YouTube. If you’re watching this, just drop down, “Yeah, actually, you know what, I’d like to hear you about XYZ, ABC-XYZ,” or I don’t know, managing lead times, whatever.
The whole point is I get messages, excuse me, like DMs saying, “Oh, it’d be cool to cover this,” but feel free to simplify this for me. Feel free to work with me on this. Whatever it is you want to cover, we will give our thoughts. But anyway, for now, I have no more questions. Thank you very much for your patience and your insights. But we’re out of time for today.
And to everyone else, thank you very much for watching. If you’re on LinkedIn, that means you can see Joannes and me. You can see our profiles. Feel free to connect with us. If you have questions you want us to answer, things we didn’t get to do in detail today, feel free to reach out and we’ll be more than happy to do that. But with that said, we’ll see you next week and yeah, get back to work.
