Supply Chain Debate On the Demand Driven Adaptive Enterprise Model (with Carol Ptak) - Ep 165

00:00:00 Introduction by Conor Doherty
00:00:35 Debate format explanation
00:02:59 Joannes Vermorel’s opening remarks
00:09:52 Carol Ptak’s opening remarks
00:17:07 Joannes Vermorel’s rebuttal
00:22:13 Carol Ptak’s rebuttal
00:27:17 Joannes Vermorel’s concluding remarks
00:29:19 Carol Ptak’s concluding remarks
00:31:24 Questions from the audience
00:32:10 Challenges of decision making
00:34:56 Thoughts on theory behind DDMRP
00:37:51 Demand Driven approach during the COVID
00:40:52 Lokad’s take on handling disruptions
00:42:17 DDAE and probabilistic forecasting
00:49:14 DDMRP comparison with MRP
00:56:40 Minimal technology for optimization
00:58:44 DDMRP implementations in large retail networks
01:00:02 Meaning of the flow in DDMRP
01:01:09 Adaptability at a system level
01:03:35 Can case studies be compared
01:07:46 Managing uncertainty upon uncertainty
01:12:26 Main critique of the DDMRP model
01:19:19 When DDMRP is not enough
01:24:47 Perspective on push vs pull
01:26:46 Safety stock and high variability
01:29:46 Why Demand Driven approach isn’t more widespread
01:35:01 The end of the debate

Full Transcript

Conor Doherty: Welcome to a very special episode of LokadTV. Today, I have the pleasure of hosting a live and hopefully friendly debate between Carol Ptak and Joannes Vermorel. Carol is a partner at the Demand Driven Institute and a visiting professor and distinguished executive in residence at Pacific Lutheran University. Meanwhile, Joannes, to my right, is the founder and CEO of Lokad. He is an engineer with Corps des Mines France and taught software engineering at École Normale Supérieure for six years.

Now, I will quickly try to go through the parameters of the debate. First, the topic: “Is the Demand Driven Adaptive Enterprise model capable of addressing the challenges of real-world supply chain decision-making?” Carol will argue in support, and Joannes against. First, there will be opening remarks of seven minutes as agreed ahead of time. Joannes will speak first, followed by Carol. Then, each speaker will have a five-minute rebuttal. Following this, each speaker will have a two-minute concluding remark. At that point, I will pose some questions, hopefully entirely sourced from the audience. Feel free to submit your questions at any time in the live chat. Oh, and at the end, they will have a free exchange, which is what everyone’s here for, really.

Now, in preparation for the debate, both speakers have agreed to the following definition, and I quote: “The DDAE model is a management tool for sensing market changes, adapting to complex and volatile environments, and enabling market-driven innovation strategies. Its three primary components are the demand-driven operating model, demand-driven sales and operations planning, and adaptive sales and operations planning.” Now, to be fair, that is a long definition, so for this very reason, we have inserted a link to an open Google Document in the live chat. If you click that, you will be taken to an open Google Document in which you’ll find detailed definitions for all of those terms, and complete bios for the speakers.

Now, during the debate section, I will strictly time both speakers. The only interruption will be a gentle nudge to remind you when you are running out of time. I do also recommend that you both time each other on your devices. Speakers, we’re almost done. Speakers are to remain completely silent during the prepared section of the debate. If you start interrupting each other during your prepared remarks, you will be muted, and you have been warned of that ahead of time. And lastly, if you enjoy what we do here, you like the supply chain debates, I encourage you to subscribe to Lokad’s YouTube channel and follow us on LinkedIn.

And with that brazen self-promotion out of the way, I ask you both: Is the Demand Driven Adaptive Enterprise model capable of addressing the challenges of real-world supply chain decision-making? Arguing against, Joannes.

Joannes Vermorel: Ladies and gentlemen, esteemed colleagues, and fellow supply chain enthusiasts, it’s a pleasure to be here to discuss the Demand Driven Adaptive Enterprise model and its capacity to address real-world decision-making challenges. For this very purpose, Carol suggested three books to me: “Demand Driven Material Requirements Planning” of 2016, “Demand Driven Adaptive Enterprise” of 2018, and finally, “Adaptive Sales and Operations Planning” of 2022.

That’s 886 pages in total, but don’t worry, you only need to read about a third of it. The rest is like a Netflix series that can’t stop recapping previous episodes, as those books overlap extensively. I will spare you all and discard them as a single, overly repetitive work. As someone deeply invested in supply chain, I approached the demand-driven paradigm with high hopes. After all, who wouldn’t be excited about a framework promising to revolutionize our industry? However, after subjecting myself to nearly a thousand pages, I’m not convinced.

First, the trivialities. Page 43 of “Adaptive Enterprise,” and I quote: “If executives want to fulfill their mission, they must understand where to start.” Well, yes. Page 163: “Consistent definition, consistently adhering to the same principles.” I suppose it’s consistent to define consistent for those who might have skipped through primary school. The illustrations, presumably intended to help the reader, aren’t any better. On page 150, we have a table of numbers labeled “Data”, a bar chart labeled “Graph”, and a piece of text labeled, wait for it, “Text”. Thank goodness they clarify that. I was about to call the bar chart modern art. It’s as if the authors fear that we might not recognize these basic concepts, but perhaps they are performing a public service for those failed by their primary school.

Now, if the easy parts are insultingly simple, what about the hard parts? Maybe the true value of demand-driven lies in between, buried between the clichés. Let’s examine the equations. And yes, they include equations, or at least they label them as such. On pages 17, 25, 28, and 29 of “Adaptive Enterprise,” we encounter what the authors call equations. But these equations are just a random assortment of Greek letters and fraction bars. They aren’t equations by any stretch of the imagination. As someone who has also played with Microsoft Word’s equation editor, I get the temptation, but considering they’re trying to teach better supply chain decision-making, perhaps providing some actual mathematical formulas would be more useful.

Conversely, from pages 99 to 105, we endure an excruciatingly tedious explanation where the authors, in plain English, tell us: “Add this, subtract that, and multiply this.” It’s like reading a cooking recipe for mathematical operations. Half a dozen pages could be condensed into just a few lines of basic formulas. But perhaps doing so would reveal that the Demand Driven Adaptive Enterprise’s underlying math lacks the sophistication of a middle school textbook. Not exactly what you would expect from a work claiming to be part of, and I quote, “the emerging science of complex adaptive systems.”

To be fair, there is one genuine equation in those three books. Just one. And no, it’s not the so-called net flow equation on page 150 of the DDMRP book, which despite its grandiose name, is merely a definition. The lone equation is found in “Adaptive S&OP” on page 156. It is the Taguchi capability index. This formula is a straight cut-and-paste from Wikipedia, but hey, it’s still an equation. Unfortunately, it’s an equation from mechanical engineering for dimensional tolerances, and it’s usually regarded as completely unrelated to supply chain. It appears randomly in the middle of a discussion on S&OP performance objectives.

Now, I wouldn’t suggest that the authors are trying to confuse readers with irrelevant equations. Perhaps they just got lost in a sea of copy and paste. As we delve deeper between the clichés and the pseudo-equations, we find numerous calls to action. Now, calls to action are great. Companies do need to act. On page 44 of “Adaptive Enterprise,” we are treated to a series of recommendations that suggest that people should be trained to think systematically, people should have a common language, a common systemic language to think and work, and we must enable people to understand the connections between departments, resources, and people.

Ladies and gentlemen, what a brilliant program. As a CEO myself, I would be overjoyed if my 60 employees could achieve that. And mind you, at Lokad, we hire elite engineering talent, and even for us, what Carol suggests is ludicrously difficult. I can only imagine how well this would work in a company that employs thousands of employees, where the main connection they understand is after-work drinks on Friday nights. So naturally, I anticipated the book’s guidance on how to rewire my employees’ minds, teach them a new language, and have them comprehend the intricacies of every department. But after dropping this bombshell, the books promptly move on to the next chapter, providing exactly zero guidance on how to accomplish these lofty goals.

So to sum up, we’ve got almost a thousand pages that swing between the blindingly obvious, the utterly trivial, the mathematically nonsensical, and the wildly impractical. Demand-driven boasts leading a revolution in supply chain management. It’s ironic that the only thing it has revolutionized is my disappointment in the current state of supply chain literature.

Conor Doherty: Joannes, you have 21 seconds left.

Joannes Vermorel: I’m good.

Conor Doherty: You’re good? Well, on that note, Joannes, thank you for your opening remarks. Carol, I turn to you for your opening seven minutes, please.

Carol Ptak: Oh, thank you very much. Well, that was amusing at best. I didn’t realize I was coming on for a book report and for a page-by-page critique. So to discard that, I was really hoping that our debate was going to be about the Demand Driven Adaptive Enterprise model, not about a book report and pages cited. Just to get that out of the way, those three books were written for three very separate, different markets. I did not expect any person out there to read all thousand pages. I just thought, with Joannes’ scientific mind, he might enjoy understanding both the operational, tactical, and strategic view of this supply chain.

So let’s get into what the Demand Driven Adaptive Enterprise model really is and why it is revolutionary. DDAE is really based on complex adaptive system science and understanding that supply chains are not chains. Supply chains have never been chains. We named it incorrectly when we named it, and it’s because those of us that were involved with the naming of supply chains, myself included, we came out of operational capability where we were used to and accustomed to using optimization algorithms to understand where our bottlenecks were and how it is that we could maximize the output of the overall process based on the maximization of the bottleneck.

So when we first named supply chain, we said, “Well, okay, I’m going to take my operations and I’m going to connect it to my customer and my customer’s customer and my supplier and my supplier’s supplier, and voila, there we have a supply chain.” We were very wrong. Supply chains are not chains, never have been. They are complex adaptive systems, and complex adaptive systems run on a very different science than a chain does. A chain is a linear system. Complex adaptive systems are not linear. They’re networks. There’s lots of nodes, there’s lots of connections, and unfortunately, academics love to cut the connections so that they can study the nodes in detail and then believe that we can add it all back together again and understand the whole. When in fact, the minute that the connections are cut, then we’ve lost the context for the whole.

So what makes DDAE different is the fact that it does understand that supply chains are indeed not chains; they’re complex adaptive systems, which means they don’t stay in any steady state too long. As soon as there is any pressure put on them, they will change and morph, and by definition, they cannot be mathematically optimized. Complex adaptive system science comes out of the idea of biology and economics, and so it’s very well understood. If anybody’s interested in a very good book on that, there was a book written called “Team of Teams” written by General Stanley McChrystal.

So how does DDAE work? Well, we understand that every company today is in a variable, volatile, uncertain, complex, and ambiguous world. So what we must be able to do is have the ability to sense changes in that marketplace very quickly and then adapt planning in production, pull from suppliers, and take care of it and do it all in real time. Now, brand new idea? No. The definition existed for demand driven back in 2001. It was coined actually when I was at PeopleSoft. We didn’t really understand how to do it until about 2006 when Chad Smith and his team started to use the concept of decoupling across the supply chain.

Because of the VUCA world, that volatile, uncertain, complex, ambiguous world that we live in, unless our response times to the marketplace are less than our customers’ tolerance time, somewhere somebody along the supply chain has got to hold inventory. So inventory is an asset. We’ve allowed inventory to be discussed as a liability, as an asset, things back and forth, but it depends on where and how much of that inventory exists. If we have the right inventory in the right place, then inventory is clearly an asset because it improves the return on investment for the company, which is the relevant metric.

So how is it that we can achieve coherence across an organization to drive ROI? How do we manage that operational, tactical, and strategic relevant ranges so that the company is in coherence to achieve ROI? I can’t go out on the shop floor and ask Joe on the shop floor what did he do to increase ROI that day, but I most certainly can go out and talk to him on the shop floor and say, “What did you do to improve flow?” And again, not a new idea. We’ve known about flow for a very, very, very long time, going all the way back to the ancient Phoenicians as they had to retrofit their trading vessels to warships.

The DDAE model is built on the coherence of flow to the organization, which transforms everything in the organization. No longer are we focused on cost efficiency and optimization because we recognize that what we are managing is not a linear system; it’s a complex adaptive system. And the modern world that we’re managing it in is a volatile, uncertain, complex, and ambiguous world. MRP, for example, was conceived in the ’50s, commercialized in the ’70s when Joe Orlicky wrote his book. And what we understood at the time is that we needed to be able to do dependent planning, and so dependent planning was the real asset from MRP.

But remember, back in the ’50s and ’60s, we had 8K of memory and a couple of tape drives, and so we typically only ran the material planning maybe once a week, many companies once a month, and then we disaggregated from there. And we really thought that as technology would get faster, that things would get better. And in fact, in 2001, PeopleSoft brought out the first real-time MRP system, and the reaction from our customers was, “Please make it stop,” because they could not handle the system nervousness. The level of precision when we try to connect it across the supply chain causes such self-induced volatility and variability that the planners can’t handle it.

So the idea is, how can we at the same time react very quickly to changes in the marketplace in a volatile, variable, uncertain, complex, and ambiguous environment and be able to take advantage of today’s real-time computing? When Dr. Goldratt and I wrote the book “Necessary But Not Sufficient,” we were talking about technology because what we understood was that as technology changes, business rules have to change. And as business rules change, technology has to change. And we’re very fortunate today that we have things like machine learning and artificial intelligence that, by the way, are also based on the same science as the DDAE model.

And so that’s what makes us very innovative, is that now the business rules are now aligned with the possibility of technology, and so that we can now sense changes in the marketplace, adapt our planning and production, pull from suppliers, and take advantage of the real-time systems that we have.

Conor Doherty: Well, Carol, I gave you an extra 3 seconds there, but they were well spent. Thank you very much. Thank you. At this point, Joannes, I kick it back to you for your 5-minute rebuttal.

Joannes Vermorel: Yes, I mean, the first thing is that I can’t help but notice the contradictions, for example, on mathematics. Because when Carol is quoting modern computers, computers, as the name suggests, compute. That’s the only thing that they do. They don’t have any kind of crystal balls or anything. And in fact, in the very books, there are tons of equations. Again, I’m not saying that I found—I am describing things as equations. Things are mentioned and listed as equations by the authors themselves. And then when they deal with nonlinearity, we are again in the realm of mathematics. So this is not something that I am setting up for myself; this is what the authors are setting up for themselves.

Now, based on my critique of these books, which are pretty much the holy scriptures of demand driven paradigms, the response seems to be, although there is a lot of digressions, that the whole is more than the sum of its parts. Right, we can’t really look at the pieces. So no matter how dysfunctional the pieces are, slap them together and voila, you’ve got greatness. It’s like assembling a car from spare Toyota parts and expecting a Tesla. And guess what? We have also case studies to back that. That’s also a point that would be of interest.

On page 325 of the DDAE book, we have a retail case study using DDMRP, for example. It claims 60% of increased revenue, 40% of decreased inventory, and I quote, “the elimination of a sense of scarcity in the stores despite nearly halving the stock in the first place.” Well, if you believe that, I have a bridge in Brooklyn to sell you. But here is the kicker: we can’t verify any of those case studies. Shocking, I know. And the endorsement comes from the very vendor hawking the demand driven miracle cure. It’s like a restaurant owner writing their own five-star Yelp review: “Trust me, it’s the best sushi in town.” Sure, but the case studies are nothing more than a fancy way of saying, “Because I said so.” Not exactly compelling evidence.

Now, on the point, because there were so many tangents here, we were on factoids, the definition of complex adaptive systems, anecdotes, where does the name supply chain come from, and some trivia on ERP getting better and technologies and whatnot. But the reality is that if we go back to a simple test, I would say in real time of adaptive enterprise, page 7 lists nonlinearity as the very first principle, which Carol also pointed out. So that’s the very first principle of complex adaptive systems. Sounds impressive, but let’s pick the simplest nonlinearity that we can have in supply chain: MOQs, minimum order quantities. Surely demand driven would have something profound to say about MOQs. Well, not really. Across a thousand pages, MOQs are mentioned six times. That’s nice in every single book, so about two times on average for every book. So that’s quite a lot of material that we have.

And let’s take an example. Page 63, we have an example of an MOQ that is so small it might as well not exist because numerically it has no impact whatsoever on calculation. Fascinating stuff. And then page 115, we have a container order situation. Interesting nonlinearities from several fronts with an MOQ. And so what is the situation? We have an order size of 100 units, a container size of 100 units, and an MOQ—wait for it—of 100 units. So what a coincidence. It’s like the stars align just so they wouldn’t have to deal with any actual nonlinearity. You can repeat this process with price breaks, perishable goods, cross-docking, repairable equipment, you name it. Demand driven has absolutely nothing to say about these common nonlinearities. Nothing. Zilch.

And that’s the essence of demand driven: a flashy theory that sets grandiose goals for itself, leveraging the best that technology has to offer. Yes, but technology gives you computers to do calculations, and there are so many equations, and yet they’re doing nothing. So essentially, we set grandiose goals for ourselves, but then we have nothing to offer to deal with common decision-making problems. And so should we believe that demand driven can tackle real-world supply chain challenges? Let me think. No, absolutely not.

Conor Doherty: Several seconds to spare. Thank you, Joannes. Carol, your five-minute rebuttal when you are ready.

Carol Ptak: Thank you. Again, I’m very disappointed that Joannes is choosing to use a book report rather than to debate the model that we were supposed to be debating. But let me first address the case that he is citing in the book, and I would welcome you to come join us in Frankfurt next week where you could speak to the person that actually did that implementation. David Poveda will be there from Medan Colombia, and he can give you the very specifics.

Demand Driven World next week, we also have, because I know you’re always very concerned about cases, that it’s, trust me, believe me, and it’s always the case studies are done by the software company or by the consultant, which always try to put a shiny face on the apple. We don’t allow that at the Demand Driven Institute. All of our case studies are done by the practitioner. So I’d invite you, Joannes, and all of our listeners, if you’d like to register for the Demand Driven World next week.

We have nine case studies coming in, new case studies from companies like Assa Abloy, where Fredrik Helgesson, the logistics director, will be presenting. Another case from Mexico, from Mega Alimentos, where Antonio Treviño, the supply chain director, is coming. Mettler Toledo is coming with the head of their global planning, or A2A with their managing director, or Gelwin with their VP of supply chain, or Sapo with their head of planning, or Koch Engineered Solutions with their global head for planning and scheduling, or PPG with their Latin American supply chain director.

Those are just the case studies that are going to appear next week in Germany. I would encourage anybody, trust me, go ahead. We put all of our case studies on our website. They are only done by the practitioner. We do not allow the software company nor the consultant to even co-present. These practitioners say, “This is what we did, this is why we did it, this is the problem that we had, this is the results we got,” and very exposed say, “And if we had to do it over again, this is what we would change.” We do not monitor or edit any of their comments.

So as we look at the idea of MOQ, I think you’ve misquoted the number of times MOQ appears in there since it appears every time that the net flow equation also appears in there. But I really still think that you’re missing the point of what the Demand Driven Adaptive Enterprise is. It is really three separate relevant time ranges with the tools required that are relevant for that time range.

Now, what is relevance? And that’s a definition that is in the book. Relevance is how I establish and how I connect the requirements with what’s going on in that time range. So how do I more closely connect my assets to what’s going on in the market? By just implementing DDMRP, which is the engine inside the Demand Driven Operating Model, typically on average, companies achieve a third to a half reduction in inventory, and typically their on-time and in-full goes up to well over 90%.

I would refer you to the Coca-Cola Africa case to hear Coca-Cola Africa talk about what was going on there. Now, before they did DDMRP, their forecast accuracy was about 50%. They implemented, they got better results, their inventory went down, their on-time in-full went up, and at the end, their forecast accuracy was about 50%. Does that mean we don’t forecast? No, of course not. We need forecasts to be able to run the tactical and the strategic range. What I was hoping we would get into in this debate is more conversation about how the DDAE model works rather than a page-by-page book review.

So as we consider the idea of forecasting, you know, on the probabilistic forecasting, yeah, it definitely has a role, but it has a role only in the tactical and strategic ranges, which allows us to help modify and adapt the operating model of which DDMRP is the planning engine. So we look at that, we have to consider the DDAE model can only take what it is that we can affect. So outside of our consideration has to be our market-driven innovation, and on the other side, we have to consider actual market demand.

And as I said earlier, if we are fortunate enough that our total cumulative lead time sits within our customer expectations, that’s an easy company to manage. However, that’s not the world that we’re living in. Our customer tolerance times are significantly shorter than our cumulative lead time. So we must have some management model to be able to sense the changes in the marketplace, adapt our planning and production, be able to translate an adaptive business plan into what it means from operational capability, and also exploit our unique operational capability to take advantage strategically. I think I gave you the three seconds back.

Conor Doherty: With change. Thank you. Well, thank you very much, Carol. With that, Joannes, I return to you for your final two-minute remark.

Joannes Vermorel: So nearly a thousand pages of Demand Driven materials plus a few minutes of commentary can be summarized on a sticky note. Here is nothing but the most damning thing: the Demand Driven paradigm is utterly impervious to reason. I could spend all day quoting lines, highlighting whether each one is trivial, nonsensical, or downright delusional, and we’ll still be stuck in the same place, like a hamster on a wheel but without the entertainment value. Why is that? Because every time I point out a flaw, it’s like trying to play chess with a pigeon. It knocks over the pieces, it craps on the board, and then it struts around like it won.

Carol did not respond to any of the serious criticisms I brought, including the basic ones such as the flagrant misuse of the Taguchi capability index. She did not explain the pseudo-equations. She could have attempted to refute my arguments one by one, yet she did not. And she did not because she cannot. So instead, we are treated to a series of digressions, mostly authority arguments. Let’s not kid ourselves. Case studies are just a fancy way of saying, “Trust me, I am a professional.” Ladies and gentlemen, I appeal to the most elevated form of human reasoning: the duck test. If it looks like a duck, swims like a duck, and quacks like a duck, then it’s probably a duck. If a theory looks like garbage, smells like garbage, and sounds like garbage, then it’s probably garbage.

In conclusion, can the Demand Driven Adaptive Enterprise model address real-world supply chain challenges? No. But I will concede this: if you can somehow trick your competitors into thinking it can, then you will definitively gain an edge as they will crash and burn.

Conor Doherty: Thank you, Joannes. And Carol, I turn to you for your two-minute concluding remark, please.

Carol Ptak: Thank you. Well, I’m very disappointed in Joannes, to be very honest. I really expected an open, honest discussion rather than him reading from his pre-prepared notes without considering the points that were made.

As to the Taguchi function, as I said in my five-minute rebuttal, the adaptive business plan then creates an operating model. An operating model has a target, has an upper and lower specification limit, and when we compare that to how the process runs, because the Demand Driven Adaptive Enterprise model now allows us operationally to run with process control as opposed to transaction control as we have in the old days of MRP, well then the Taguchi function obviously fits because we want to see how well is our actual performance against that defined range.

As I said, I didn’t expect a page-by-page book report or book review. What I really expected was a discussion of the methodology itself. And it’s not “trust me.” I would suggest that you talk to the actual practitioners and look at their actual results. To me, that’s what really speaks louder than anything. It’s not “trust me.” It is, “This is what our business problem was, this is what we’ve implemented, this is the results that we’ve achieved, and if we had to do it over again, this is what we would do differently.”

And when we talk about does the Demand Driven Adaptive Enterprise model address the needs of this VUCA world that we’re living in and provide real-world results, the answer is absolutely and unequivocally yes. The tens of thousands of people that have been through the DDI education, the results of the companies, the increases in ROI, the company’s ability to be able to survive the pandemic as their demand patterns were going all upside down and squirrely and still were able to improve revenue and ROI, I think the results speak for themselves.

Conor Doherty: Well, thank you both very much. And Carol, thank you for those remarks. At this point, I’m going to transition to some of the audience questions. Actually, there are already quite a few in the live chat. Just to be clear, we ask that the questions be designated who the questions are for, but I’m obviously going to pitch them to both of you. And again, it’s not timed, but try to keep the remarks terse so everyone gets a chance.

But before I get to the audience questions, there is just one that I wrote down because I listened to both of you speak now for the last 33 minutes. And you know, you went back and forth on the books and whether or not it’s about the books, that’s fine. But unless I missed it, at no point did either of you actually define what you see the actual challenges of real-world supply chain decision-making being. So Carol, I’ll open with you. As tersely as you can, what do you actually see the challenges of real-world supply chain decision-making being?

Carol Ptak: Well, the biggest challenge is what I said, which is how do I respond to a world that is variable, volatile, uncertain, complex, and ambiguous? And how do I do it in such a way that I can grow my return on investment?

Conor Doherty: Johannes?

Carol Ptak: It’s about as concise as I can make it. It is. And if Johannes wants to put it on a Post-It note, he can write that. It sums up the Demand Driven Adaptive Enterprise model in one Post-It note: it’s all about flow.

Conor Doherty: Well, thank you, Carol. Joannes?

Joannes Vermorel: My take is that the supply chain is the mastery of optionality. You have limited resources of everything, and you need to allocate them, which represents in practice millions of decisions daily for a sizable supply chain. So solving the problem essentially means making those decisions. They are very basic. They are: what do you buy, what do you produce, what do you allocate, what sort of price point do you have, do you increase or shrink your assortment, etc., etc. And so in my view, all of that indeed for profit. But in my view, the supply chain is a theory and a practice that lets you do this decision-making at scale, which involves nowadays a lot of things to be computed so that it could be automated with computers. That’s pretty much it.

Conor Doherty: Is well, Carol, now that you hear Joannes’ take, do you wish to amend your own or you agree or disagree?

Carol Ptak: No, not at all, but I think, and you know, having been around since the earliest days of computers and a conversation that I had with a computer company and a software company, he said, “We don’t mandate that our customers do things the way we tell them.” And I said, “You most certainly do because what you build into your software is what you consider is industry best practices.” Now, what if those practices are wrong?

So the methodology goes with the computing and the technology goes with the methodology. You know, for example, at the Demand Driven World next week, we have Simo coming who can do a full digital twin of a company to be able to start to make some of those strategic decisions to which Joannes is referring. But does that with the potential of a DDMRP engine sitting in there so that it’s understanding where do I position strategic stock buffers, how do I plan those, how do I then get to real-time response to my marketplace? So technology in and of itself is necessary but not sufficient. Good book title.

Conor Doherty: You wish to add anything or I can push on?

Joannes Vermorel: No, push on.

Conor Doherty: Push on. So, this question is directed to Joannes. I’m reading this verbatim as it was put to me. Could Joannes share his thoughts on the theory behind DDMRP, specifically DDMRP and how it builds upon existing supply chain practices?

Joannes Vermorel: In short, DDMRP is a set of trivialities. They dimension buffers with three colors. There is nothing really specified at the decoupling point. You have no algorithm to know how to put them, so basically they just provide extremely ambiguous guidance. There are also gross mistakes. For example, they say when MOQ is present, you need to have the green zone to be as big as the MOQ, which is absolutely insane because there are plenty of situations where reordering up to your MOQ is insane. So that should absolutely not be part of whatever the DDMRP refers to as green.

But bottom line, it is very, very thin. You know, that’s the thing that for something that is quantitative, my take is that it could be summed up in about three pages and that’s it. And so it is very, very weak. It’s even an insult to operations research, which came before, to say that it would be the descendant. It is not. Operations research was already years of sophistication ahead of DDMRP.

Carol Ptak: Well, and I would challenge sophistication versus results. Just because it’s sophisticated doesn’t mean it’s better. DDMRP actually sits on the idea of lean manufacturing, MRP, DRP, theory of constraints with some innovation that actually now harmonizes all those things that previously we thought were the antithesis to each other. So it’s really all about flow.

And as to the how do I position those buffers, I think he probably missed those pages in the book. There are six criteria about where those buffers are positioned, and that includes customer tolerance time, market potential, lead time, waters, external variability. So there are six of them, and that is what is then optimized and considered in a digital twin to consider once I’ve positioned those buffers.

Typically, what we see is supply chains tend to stabilize because we’ve eliminated the system nervousness, and then both the positioning and the quantity needs to change. So this is the adaption cycle. So it’s not just pure pull; it’s position, protect, pull, and adapt. But we’re very clear about where those buffers are positioned and green, yellow, red because that’s that practicality versus sophistication. Everybody understands green, yellow, red.

And so I understand the rules. What happens when I see green, yellow, red? That’s why planners love it and companies are very quick to implement it, and implementations typically go much faster than what had been originally planned.

Conor Doherty: Joannes, no comment?

Joannes Vermorel: No comment.

Conor Doherty: I’ll push on. This one is directly to you, Carol. I’m reading as it’s written. Why did the demand-driven approach struggle during the COVID crisis and what should companies do to adapt in such situations?

Carol Ptak: Well, there was an interesting conversation during the COVID crisis. We didn’t struggle. I think every IT project out there, every process improvement project out there during COVID got shut down, and that was unfortunate. We spent a lot of time on the phone with senior executives that would say, “Well, we’re going to get back to implementation when we get back to normal.” And our message to them was, “Welcome to the new normal.”

The question isn’t if disruptions are coming, it’s when and where, so you best be prepared. And so what we’ve seen is after COVID, the actual demand for our seats of education has risen to a record level, and the number of implementations globally has risen to a record level because as executives realized that what they have to deal with is this variable, volatile, crazy world that we’re in. Not only did we have COVID, we had the Russian invasion of Ukraine, we had the next pandemic coming through, we had the craziness at the American ports, we’ve got dock worker strikes. It’s not about if the next disruption is coming, it’s when and where.

And unfortunately, during the COVID crisis, a lot of senior executive teams said, “Well, when we get back to normal,” and our message was, “Welcome to your new normal.”

Conor Doherty: All right, thank you, Carol. Joannes, forgive me, why do you believe the demand-driven approach may or may not have struggled during the COVID crisis?

Joannes Vermorel: This question was not addressed to me, so I may just comment on the answer of Carol. Because again, I have no facts on because I’m not really privy to what is going on exactly in the companies that run those things. But what I would say is that to a question that is as factual as this one, what we get, and that’s something that is very typical of the demand-driven paradigms, is an endless list of factors: you know, regression, Ukraine war, volatility, uncertainty, etc. Buzzword, buzzword, buzzword, problem, problem, situation.

You see, it’s like a profusion of things. But when I start again, and the books are exactly the same, you have the list in every single page. They would go on 20 tangents, and every single time I think, “Okay, now they have opened like 20 chapters to address every single one of those tangents,” and you get nothing in terms of concrete, mathematically sound, and when I say mathematically sound, I don’t mean high math, I mean even primary school math, something that is not ambiguous, that gives you a rule that you can compute, and then nothing. You just move on, and again, that’s just a profusion of endless factoids. And I think that’s really a pattern, and I would like the audience to pay attention to those profusions of factoids.

Conor Doherty: Well, actually, if I can push on because the next question will go to Joannes and then for Carol, I’ll throw it to you. But does Lokad offer, I didn’t write these, does Lokad offer a different approach for handling disruptions like the ones seen during COVID, and if so, how does it address such challenges?

Joannes Vermorel: So the long answer is in the series of supply chain lectures, but that’s a very long answer. The short answer is we use probabilities and probabilistic forecasts. The idea is to have an economic model where events that have a low probability and a large economic impact can be factored in. So you need probabilistic forecasts, and then on top of that, you need a second instrument. So that’s the predictive instrument, and then the optimization instrument is stochastic optimization, which is the general term for any kind of general solver that can give you an optimized answer under uncertainty.

Bottom line, you assess the probabilities of all the possible futures, step one. Step two, you look at all the possible decisions, I mean obviously shrunk to what a computer can manage, and you optimize what gives you the highest risk-adjusted return on investment. That’s the short answer on how Lokad does it, I would say, in very, very technical terms.

Conor Doherty: Carol, earlier you spoke that the DDAE model, like the all-encompassing hierarchy of concepts, is compatible with probabilistic forecasting.

Carol Ptak: Absolutely, absolutely. I mean, probabilistic forecasting is something that would help us design how the operating model is defined. But, you know, to challenge back to Joannes on his answer, that was a very complicated scientific answer that basically boils down to, “You know, the answer came out of the computer, trust it.” And I don’t know a single planner on the face of the planet that is going to say, “Oh, it came out of the computer, trust it.” The DDAE model is more understandable.

All right, in layman’s language, I don’t have a PhD or two or three. And so, you know, what I would put up is say, “Okay, first we got to agree on the problem. What’s the problem we’re trying to solve?” And that’s why we go on and on about variability, variety, you know, the real problems of the real world and how DDAE solves that. And, you know, the other question I would have is, “Okay, Lokad, where is your page with your case studies of how you’ve solved problems for your customers in the real world with the real bottom-line results that are presented by your practitioners?” And I would put that page up against what the Demand Driven Adaptive Enterprise model has done any day. And like I said, come join us next week in Germany, meet these people face to face, talk to them.

Conor Doherty: Any comment? No more factor and more digression and an authority argument on Cherry and the cake. So, no further comment.

Well, if I can follow up there, Carol, again, and I don’t want to put words in your mouth, so correct me if I’m wrong, but the way you framed your response to Joannes’ comment was almost like, “Well, I don’t have a PhD either, so hey, I’m not a doctor. I came out of computer science and numbers.”

It seemed like you were positioning yourself and your approach as not necessarily being anti-academic but understandable. My question to you as a follow-up is, if it’s understandable but less effective than a more sophisticated solution, would you be okay with that?

Carol Ptak: No, I wouldn’t because I think it is more understandable and more effective. When planners and managers can understand how something works, then they’re going to use it. Like I said, there’s not an executive on the face of the planet that’s going to say, “Oh, the numbers came out of a computer, good.” Because I would also challenge Joannes that you cannot optimize a supply chain because supply chains are complex adaptive systems. You can look at alternatives and you can select one, but the reality is that unless and until you don’t have any variability in execution, there’s always going to be a range of possibilities that the actual results are going to see.

In demand-driven, I would say that not only is it highly understandable, we don’t use anything higher than fifth-grade math. So, I can understand why Joannes would be insulted by the primary academia of the mathematics, but at the same time, we don’t use anything higher than fifth-grade math. It’s very understandable, so companies use it and they see incredible results. There’s a great case study; it was the last one when we did Germany a few years ago. She says, “Yeah, I know, same thing as everybody else has got inventory is down by half, on-time full is up 90%, boring.” And I was like, “Man, when you get tired of seeing those results, I’m in the wrong spot.”

So, I would suggest to you that not only is it easier to understand, but it is more effective. But it’s not at the antithesis of probabilistic forecasting because that mathematics can help us understand as we start to move through the model once the initial implementation is done. How do we adapt? And that’s where I think the probabilistic forecasting, the digital twins really come into play, is to understand all those relationships. But first, the first step has got to be to stabilize the supply chain to be able to mitigate that operational variability.

Conor Doherty: Okay, well, Joannes, to be fair, you made some notes. Do you have a response to that?

Joannes Vermorel: I mean, first, again, pointing out things that are slightly nonsensical. Yes, the DDMRP and the complex adaptive system and all this theory does an optimization. It’s stated at the very beginning: it optimizes return on investment. If you try to push a number upward or downward, you are doing an optimization. That’s the definition of an optimization. So, when you say, “You see, that’s the sort of thing that is completely schizophrenic,” where you say, “Oh no, we don’t really do that, we don’t do optimization,” and then you just mentioned in the next minute that you’re trying to optimize return on investment. That’s like, sorry, this is the very definition of optimization.

And then if we go back to…

Carol Ptak: We are trying to grow ROI, not optimize it.

Joannes Vermorel: But that’s the same thing. Grow, optimization is literally a way to take a target function that can be ROI and move it a bit in the desired direction. That’s literally the Wikipedia definition of optimization. So, this is exactly what you’re doing. So, that’s insane to me, this sort of approach.

And then probabilistic forecast, I am very sorry, but the formulas and everything that is given in those books, they are very weak. The formulas, yes, I can also, again, this is a little bit of an authority argument on my side, but it is completely incompatible with probabilistic forecast. Just to give you a taste of what, if you apply probabilistic forecast, what it looks like, the first thing is that you don’t want to look at your SKUs independently. You will weight the contribution of every single unit independently across the entire company. That’s literally probabilistic forecasting 101 that you will get.

So here in this methodology, you are dealing with the buffer one buffer at a time. So, that’s, sorry, it is just not, those things do not even exist in the same plane. They are not compatible, neither in terms of concepts, nor in terms of methodology, nor in terms of technologies. They are widely, widely different.

Carol Ptak: Did I say that the probabilistic forecast would be one buffer at a time? I think one thing we’ve always said about DDAE is that we look at the holistic and the cause and effect across. And again, I’d invite you to pop on a training, come on over to Frankfurt next week. We’ve got about three presentations of where probabilistic forecasting is looking at the entire network and being very successfully used inside of DDAE model.

Conor Doherty: Okay, next question. Again, this one actually, Carol, is directly to you. There are quite a few. Whenever we get tired, we can stop, we don’t have to answer them all. How does DDMRP, again, I’m reading verbatim, how does DDMRP address the issues inherent in MRP logic? Does it require running multiple times a day to be effective?

Carol Ptak: The closer you get to running DDMRP in real-time, the more stable it becomes, because it allows our planners to have the most relevant real-time information. Is it necessary to run in real-time? No. How it addresses the limitations of the MRP logic is the power of MRP is everything is dependent, and the bad news about MRP is everything is dependent. So, a delay anywhere is a delay everywhere.

How DDMRP logic addresses that is by inserting these decoupling points based on one of the six criteria to determine where those positions of independence are going to be so that it absorbs variability from both sides. It decouples and it provides our primary position for planning. In between the decoupling points, it’s dependent like it always has been. So, that’s why we took a lot of criticism when we named it DDMRP, and it’s because MRP is still in there. Because between the decoupling points, it’s still dependent planning as we always have. So, it addresses the limitations of MRP by insertion of those decoupling points, and those are the primary positions for planning.

Conor Doherty: Thank you. Joannes, turn to you for comment.

Joannes Vermorel: Yes, I mean, there are several things here. First, MRP is really the wrong baseline. At its core, it’s using a traditional database, and the problem is that a transactional core is absolutely trash when it comes to analytics, all kinds of analytics. So, this is insanity. This is an insane baseline, and so I think it is incorrect to compare MRP to anything. This is an antiquated baseline that should not even be considered.

Then, when it comes to real-time, I mean, again, this is something where you should really challenge where the question comes from. Because the reality is that a modern computer, as a baseline, gives you a 2 GHz processor. It means that you can do two billion operations per CPU. And a modern computer has, your phone has eight CPUs, so that’s literally tens of billions of operations per second on a smartphone.

So now the question is, what do you have that cannot be done within a latency of microseconds? And the short answer is that when you design across a system on top of a transactional database, you get absolutely horrid performance. And so, vendors who just manage to kind of mitigate the absolutely horrid performance refer to that as real-time. It’s really nonsense, I mean, really, really nonsense. It’s just misuse of modern computing hardware. I could go into the detail, but I would say here we have a really wrong baseline for MRP and for real-time. That would be my comments.

Conor Doherty: Carol, I think some of that you might agree with, an improper baseline being MRP or no?

Carol Ptak: Well, the reality is this: MRP is being used by virtually every company around the world. So yes, I agree it’s antiquated. I agree it needed a move into the future, and that’s why we did DDMRP. That’s why we had to implement the decoupling buffers, which allowed us then to run operations according to process control as opposed to transaction control, which is what MRP does. Everything is transaction control. You’re either okay or not okay in MRP. You don’t know how okay or how not okay you are.

And you know, MRP real-time first came out with PeopleSoft in 2001, and our customers hated it. I mean, I have the advantage over Joannes that I’m really old. So, when I was teaching at the university, I had students tell me how they admire how I did the research on the history of IT, and it’s like, yeah, it wasn’t research, this is anecdotal, I lived it.

And we really thought that as computers got faster, it would solve our problem. But we discovered as computers got faster, our problems got worse, and that was because of the system nervousness. My very first APICS meeting 46 years ago was about system nervousness. We knew about it then; we just didn’t know how to solve it. And we didn’t know how to solve it until DDMRP came along to be able to stabilize the planning function.

But the whole idea of APS, I mean, there’s not an APS implementation out there that has been successful. To give Joannes his definition, success is: has it increased the company’s ROI? And it’s because it’s trying to do this multi-echelon optimization based on an incorrect business function. And I agree with him, you know, the technology has to change when there’s a change in business rules, and the business rules have to change when there’s a change in technology. That’s what Eli and I wrote in 2000 when we wrote “Necessary but Not Sufficient.” We’ve known that for a long time.

Conor Doherty: Thank you.

Joannes Vermorel: Yeah, I would comment again, misuse of terms. When I say transactional for a database system, I mean it in a very specific way. It refers to the way it is used when you design databases. And when you say transactional, it has nothing to do with finance or some sort of process, etc. It means essentially the ACID property: atomicity, consistency, isolation, durability. Those are properties granted by your storage.

And DDMRP is just as transactional as MRP as a paradigm. And all the implementations I’ve seen, your vendors that do DDMRP, they do it on top of SQL databases just like everybody else who was doing MRP. So again, there are so many things where you’re using words, but you’re not using them in the proper way. So that means that if you refer to transaction, you refer to something that has nothing to do with the point, which was the design of database systems. You will go on a tangent on using transaction for something that is more like the methodology of DDMRP.

And again, it’s completely different things. I’m sorry, so I’m just pointing out that we have, again, we had factors, but we are also constantly shifting the semantics of what the words actually mean.

Carol Ptak: Well, and I think that’s where the conversation that we had when we first set up this debate is to get to the definition. Because my perspective on the world comes from a lifetime of running manufacturing and being an operations planner, being a planner on the shop floor, being a supervisor, being an operations VP, working in the IT industry as the industry expert.

You know, coming at it from the practitioner real-world perspective and not from, you know, what we used to call the little white house, which is the IT back in the old days that had the raised floor. And that’s where you wanted to go out in the summertime because they had to be air-conditioned. So I’m not coming from what we call the bit twiddlers. I’m coming from the real-world perspective of how it is that you actually run an operation and run a manufacturing facility as part of an integrated supply chain.

So yeah, I would say we probably have very different definitions, but my definitions would be the ones that are used by the, you know, this was part of our debate: does it address the real-world challenges today? And that’s the world I come from.

Conor Doherty: Okay, sorry, I’m just going to push on a bit because there are quite a few questions to get to. But again, we can circle back to this point later on. So to Joannes, and again, you’ve touched on this already, so you can keep this one light, I guess. What is the minimum technology we need to build optimization?

Joannes Vermorel: I suggest to frame the problem the other way around: what are the technologies that are explicitly in the way of you achieving that? You see, because the reality is that data science, generally speaking, needs very, very little. That’s why, for example, Python is so popular.

So my take is that the curse nowadays is that modern enterprise systems are like a thousand layers. You have the database, you have the operating systems, you have all sorts of caches, you have all sorts of data retrieval layers, etc., layers upon layers. And so essentially what modern, I would say, enterprise software systems do is that they tend to just move data from one layer to the next, and that involves tons of computing resources, memory, CPU, bandwidth, and whatnot.

So bottom line is there is no minimum requirement, but you have to be aware of all the stuff that is in the way. And in this modern state of software technologies, it is huge. So my message is don’t think about the stuff that you need; think about the stuff that you don’t need and get rid of it. And then once you’re back to the core, the algorithmic core, you’re good.

Conor Doherty: Carol, I know you said that the supply chain couldn’t be optimized, but, you know, indulge me. If you thought they could be, what technology would be required?

Carol Ptak: Oh, that’s to me, technology is, you know, that’s, I’ll leave that to Joannes. You know, I live in the real world with the real-world problems of looking at the methodologies. And then I always work very closely because I worked for IBM for a while, and I had the great honor of working with the Watson Research Center. You know, those are the brilliant PhD guys. I’m not one of those. You know, I’m just a very pragmatic operations manager that has been very blessed to have a very successful career over the last 45 years.

Conor Doherty: Okay, then I shall push on. Carol, again, I’m reading these for the time. Has DDMRP or we’ll say even DDAE been successfully implemented in any large retail organizations with several hundred stores? If so, could you please provide examples?

Carol Ptak: Sure, yeah. One, the first one that jumps to mind is Mick. Most of the retail operations that it’s been implemented in are down in South America. So Mick has, they’ve got several retail stores. I’m trying to think of some of the other ones. The biggest retailer in Colombia has implemented DDMRP. There’s a unique challenge with retail because retail is what’s called, they have what’s called a long tail. Typically, about 10% of their products generate 90% of their revenue, and 90% of their products generate 10% of the revenue.

So it’s a unique application, but most of the retail implementations are actually down in South America and Mexico. And then we do have a retail one coming up from South Africa too. Takealot was supposed to be at the conference, and that’s the largest store down in South Africa.

Conor Doherty: Okay, thank you. I’ll push on. There’s not really much for you to add to that question, Carol. So you’ve mentioned the concept of flow a few times. Could you actually define the concept of flow and explain what it means within the context of DDMRP, please?

Carol Ptak: Well, that’s the foundational pillar. Flow is the rate at which a supply chain converts materials to products required by a customer. And that’s very specific. Flow is the rate at which a supply chain converts materials, inputs, into outputs that’s required by a customer. It is absolutely the foundational pillar underneath DDMRP. It also happens to be the foundational pillar underneath Lean and Theory of Constraints and a lot of the other, you know, more common, more recent, I should say, operations improvement areas. So that is the whole foundational pillar. As I said, if Joannes wanted to write a true Post-It note about demand-driven, it’s all about flow.

Conor Doherty: Thank you, Carol. Joannes, you made some notes. Do you wish to respond? Well, this one is to you. How does Lokad incorporate adaptability at a system level while balancing the sensitivity of the solution to variations in the supply chain?

Joannes Vermorel: So, two angles here. Concerning the sensitivity to the variation, the question is: are they desirable or not? There are classes of numerical recipes that are extremely, I would say, trigger-happy in terms of results, and that’s very damaging because in supply chain, you get ratchet effects. So once you’ve triggered a production batch, you can’t undo that, so you have to live with your decision.

So you don’t really want numerical recipes that are trigger-happy and erratic on their own. By the way, one of the aspects of probabilistic forecasting is that it tends to really make the numerical recipes a lot more stable. A lot of the uncertainty that you have with traditional systems is that when you have a classic forecast, a small deviation of the forecast tends to cascade into massive divergence downstream. So that problem is solved by going to probabilistic forecast and stochastic optimization.

Now we have another angle in the question, which is adaptability. The reality is that when you have a numerical recipe and something catastrophic or completely unprecedented happens, there is no substitute for human intelligence. The way Lokad works is by having supply chain scientists who can, in a very short amount of time, rewrite and amend the numerical recipes to fit the new situation. Again, we don’t have a crystal ball; we can’t anticipate something that is radically unprecedented like the Evergreen blocking a canal.

But when it happens, there are so many changes that it requires a human mind. But the human mind is not there to duct tape every single SKU one at a time; it is to rewrite the numerical recipe. Then we are back in business. All the decisions are automated, and it is done automatically and at scale.

Conor Doherty: Carol, do you wish to add anything to this?

Carol Ptak: I can’t discuss Lokad.

Conor Doherty: Well then, this question was originally for you, Carol, but actually, I think it would be more interesting to pose it first to Joannes and then we can contrast your response. So, Joannes, why are you reluctant to compare case studies of probabilistic forecasting with DDMRP or those of Carol? Let’s just put it that way.

Joannes Vermorel: Because first, I do not believe at all in case studies in enterprise software or enterprise practices. The domain has been fraught with problems since essentially the 1950s. The problem, again, you have massive conflict of interest. Just think of it this way: the vendor will not publish the case study unless it is like putting them on a pedestal.

And then the clients, the managers who risk their reputation when they go for an initiative, they have a massive incentive to have the entire world believe that this initiative went superbly. My casual observation is that 90% of the initiatives in supply chain fail across all companies, all countries, all verticals. 90%, that’s the same baseline.

And how many case studies can I name in my entire career that were showing dismal results? None, not a single one. The only negative case study I could find was through, I would say, brilliant journalists. For example, I encourage this audience to read “The Last Days of Target Canada.” This is a fantastic summarization of all the things that went wrong, but it’s super rare.

Leo lost half a billion euros just a few years ago on an SAP inventory optimization initiative. No case study. So you see my point. The conflict of interest is so massive that it is not about comparing my case study to your case study. This thing needs to be gone. This is a methodology that needs to be rejected outright, period.

Conor Doherty: Right. Well, Carol, the question was originally to you. So why do you think Joannes is reluctant to compare case studies with yours?

Carol Ptak: Well, that’s a very good question, and only he can answer it. I know that he is very reluctant about case studies. I mean, to say the more real-world observer, the question obviously would be, “Do you have any?” And I encourage people to talk to these guys, not just what’s published, but actually come and talk to these guys and get the detail.

Because we encourage them to actually say, “If we had to do it over again, what would we do differently? Where did we fail? What didn’t work? What did we think was going to work?” We encourage that kind of transparency in our case studies. As I said earlier, we don’t allow the software companies and we don’t allow the consulting companies to do the case studies. It’s the people.

That’s the reason why we put on Demand Driven World, to allow these practitioners to talk to each other so that they can have those kinds of conversations of what worked, what really didn’t work, what did they learn, how can we learn from each other. Not only the successes, which is important, but how do we learn from the failures? What did not go well?

And I think that’s just absolutely as critical. If we can help share the failures so that somebody else doesn’t have to step and stump their toe on the same curb, then I think that’s a good thing. That’s why we do Demand Driven World. Most of our implementations are in Europe, so that’s why we’re coming to Europe next week.

But we think case studies are absolutely critical because that’s the first thing we get asked for. Understand, Demand Driven Institute, we are not a consulting company. We are not a software company. We have never been a software company, and we have never been a consulting company. We are just thought leaders in the area of supply chain. So we’re very independent from all the software companies.

But as people considered demand-driven, it shifted about right after the pandemic. I would say it shifted from “Have you tried demand-driven?” to “Why have you not tried demand-driven?” And that was because of the results the company saw during the pandemic that already had implementations going.

Conor Doherty: Okay, well, I will push on, but I will come back to you, Carol. First, again, it’s actually to both of you, but I’ll just start with Carol because you were already talking. In a highly VUCA world with sparse and erratic demand, how would you make decisions without significantly increasing stock levels? And sub-question, how do you manage uncertainty upon uncertainty in such challenging situations?

Carol Ptak: Well, and that’s where you really need to understand the business. That question doesn’t give enough information. What is uncertainty on top of uncertainty? How much of that uncertainty is self-caused? How much of that uncertainty is because of your pricing strategy? There are a lot of layers of the onion to peel back there to get to the root cause.

I was just at a conference in Wisconsin where a software company came up to me and said, “How would you propose you do allocation in short supply?” I asked, “Does your client have excess inventory?” “Oh yeah, they got too much of the wrong stuff, too little of the right stuff.” I said, “Well, solve that problem.” Sometimes what we see is that this variability on top of variability is self-induced.

If I want to be a rapid response, high variability, low volume supplier, you’re not going to do that by importing from China. That’s a different strategy. Your strategy has to be in alignment with your operational capability, and your operational capability allows you to have different strategic advantages. Those things have to line up. That’s why DDAE looks at strategy, tactical, and operations and separates those three relevant ranges.

Conor Doherty: Thank you. Joannes, same question.

Joannes Vermorel: So that’s a very interesting question. Let’s start with sparse, intermittent sort of behaviors. Sparse and erratic, yes, that’s where the probabilistic approach really shines. When you’re dealing with something that is sparse, you have to have a mathematical instrument that lets you deal with subunit patterns.

If you just ask, “How many units am I going to sell over the course of one week?” you could say, “50% chance that I sell just one.” In the classic world, you would say 0.5, but it doesn’t make sense because you can’t fraction the unit; it’s packaged. The classic perspective struggles with subunit predictions, resulting in a lot of nonsense because you end up with fractional numbers that are just not real. They exist in math, but they don’t exist in the supply chain where it’s just zero or one.

With probabilities, you get a nice, elegant solution that actually works where you can have a probability for zero, a probability for one, a probability for two, let’s say, units, and maybe a probability for 50 units as well, which is going to be the erratic spike. So, sparse intermittent, that’s where it really shines.

Now, when you pile up uncertainty upon uncertainty, this is a very interesting question. How do you do in a deterministic world when you add a delay on top of another delay? The answer: you do a sum, an addition that feels like a super normal thing. So, you can add, subtract, multiply. Well, it turns out that when you have uncertainty, if you have something like an algebra of random variables, you can do all those combinations of uncertainties, and you will get through an algebra of random variables. You will be able to actually effectively compute the sort of resulting uncertainties that you have on top of all of that. So, I’m not describing exactly the solution; I’m just describing the instruments that let you get to that.

First, you need to have, I would say, statistical instruments that deal with sparsity and erraticity. So, that’s not going to be your classic sort of forecast. That’s not going to be the buffers that are glorified moving averages that are presented in DDMRP. And second, when you deal with compounded uncertainties, you need to have the instruments that will let you do that. People have been doing that for half a century in finance. This is not magic. Lokad did not invent that. It is just a slightly unusual instrument, but it is very straightforward. Just as adding, subtracting numbers, and multiplying them feels natural to you, you will just learn to do that with uncertainty involved.

Conor Doherty: Okay, well, thank you. I will push on. Okay, so this is quite a lengthy question. I’m just going to try and summarize this in real time. Uh, okay, well, I mean, this will be to you, Joannes, because you’ve already kind of answered this. There are a few moving parts, but I will read for basis.

To Joannes: what is your main critique of the DDMRP model, and what specific aspects of it are you questioning? I think you’ve already answered this, but I haven’t heard a solid argument against DDMRP beyond it being too simple. If a simple model can deliver results, why do we need more complex and sophisticated system dynamics models?

Joannes Vermorel: My main critique is that there is exceedingly little, you know, and that’s why I was pointing out pages. Because when you take the pieces, you realize it’s mostly a lot of nothing. And the idea that out of a lot of nothing put together, you’re going to have, voila, a grand assembly, I think it’s completely nonsense. So, my main critique is that it is very, very weak, both line by line and as a whole.

And then you go back to why does it work so well? The question, if you already assume that all the case studies are true—sorry, I can’t do anything for you. If you assume just like the case study is true, that you can get reliably 60% increased turnover by applying DDMRP to retail while halving the stock in the same process and giving the impression that the store is even more full, if you think that’s the sort of result that you can get, you know, because that’s what is presented—sorry, again, I have a bridge in Brooklyn to sell you. That’s it.

Conor Doherty: Well, Carol, again, I kind of want to follow up, and it’s built on that. So, again, this is a question just listening to Joannes and also listening to the kind of conversation in totality. At the start, you commented, “I was surprised Joannes wanted to talk about the books.” And again, I’m not going to speak for Joannes, but certainly for me, if you said, “Hey, do you want to learn about a thing? Here are several books that will explain, like manuals, how does the plane take off?” You read about aviation or aeronautics, you learn about Bernoulli’s principle. It’s written in a book. So, I don’t learn that planes fly; I read that book to learn how planes fly.

So, when you talk about case studies, and I’ll, for the sake of argument, just say that it works, okay, but I think for Joannes and maybe for people listening as well, the problem is if I want to learn how it works, you’re saying it’s not in the books.

Carol Ptak: Oh no, it’s clearly in the books. Joannes is saying it’s not in the books. It’s in the books. We wrote those three books for three very separate, different markets. The “Demand Driven Adaptive Enterprise” book was written for an executive to understand how the whole thing is put together. The “Adaptive S&OP” book was written for the S&OP team of how do we now link a strategic S&OP process that outputs an adaptive business plan that can be translated to a demand-driven operating model. And the “DDMRP” book is very specifically how the DDMRP engine works.

Now, I love the criticism that it’s too simple. I think that’s the best compliment that I can get. Why? Because it’s very easy to make things complex. It is very difficult to make things simple. And we’ve worked very, very hard to make the concept simple to understand and easy to implement.

So, the whole conversation today is about: does the DDAE model solve the problem in the real-world supply chain today? Well, that’s real-world. We have to have something that is understandable, easy to implement, and drives significant results. You know, as we look at the critical thinking tools, you’re always looking for that breakthrough idea that solves lots of problems and does so in a very profound way. And that’s what demand-driven does.

I mean, I love Eli Goldratt. I mean, he always said things so well. You know, he said, “If you have to use math to explain yourself, then you don’t know what you’re talking about.” I love Goldratt. I mean, he came up with great, you know, so if Joannes’ worst criticism is he doesn’t like what we called an equation, okay, the rest of the world calls those equations. And there’s certain formatting requirements by a publishing company, and I don’t know how many books Joannes has published, but there’s certain publishing requirements in the formatting when you put a book out that you do have to label things chart and figure, all right? And it’s a requirement.

So, you deal with the publishing companies, and we’d be happy to take all that out, but it’s a requirement. So, I don’t know how many books you’ve had the experience of publishing, but you’ll find that that is a requirement when you publish with some of the top-tier publishers, as though all that stuff has to be labeled. So, calling what we do simple is the best compliment that I can think of because we work very, very hard to make it simple to understand, simple to implement, but produce profound results.

Conor Doherty: Okay, thank you, Carol. I throw back to Joannes.

Joannes Vermorel: Yes, I think it’s a misrepresentation of my criticism. I didn’t say that those books are simple. On the contrary, I extensively represented that they are very convoluted to present things that are, in the end, very simple. That’s when I say you spend literally half a dozen pages in English to say, “Add this, subtract that, multiply by that.” It is just insanely difficult to follow what would have been represented with, again, primary school formulas, like super basic.

And on the contrary, you see, that’s the point with this book. I’m not criticizing that they are too simple. That’s not my point. My point is that they are exceedingly weak. That’s a very different criticism. Weakness is not simplicity. You can have things that are exceedingly simple and beautiful. Maxwell’s equations, you know, exceedingly simple, beautiful. Yes, the formalism is quite elaborate, but this is not the sort of problem of simplicity I’m talking about.

My point is that those books could have been drastically simplified, actually drastically, again, by sticking to the established norms of when you have things that you want to add, subtract, and whatnot, you just use a simple formula, and you don’t go into literally half a dozen pages of extremely complicated and convoluted explanation to explain what is simple. And my point, the criticism, is that by doing that, you inflate the number of pages, you inflate the mass of words to, in the end, deliver very, very little on, again, 900 pages.

Conor Doherty: All right, I will push on. At this point, we have been going for 80 minutes, so I’m going to start cutting questions that have already been answered. So, again, I’m not going to ask Joannes about DDMRP case studies again. We’ve trod that ground quite well. Yes, so this, I’ll go first to Carol.

Can you jointly define the scope, situations, or conditions where something more sophisticated than DDMRP is required? For example, in disassembly processes, DDMRP appears to fall short. How would you address such scenarios?

Carol Ptak: Actually in disassembly, it worked very well. One of the earliest case studies was a company called Erickson Air-Crane. Sorry, Joannes, to go back to a case study, but the Erickson Air-Crane actually has the flight certificate for the Sikorsky helicopter. And so they have a complete disassembly process. So it actually works very, very well, and it works there very well because of the high level of variability.

When you get an aircraft that comes in, it lands as maintained. Now you got to figure out how it was as built, as engineered, and then now you got to try to retrofit the whole thing back. And then you have an issue with your FAA flight certificate that says one part was retrofitted and is good till October 31st, 2024, but another part was retrofitted and good till June 1st, 2025. The frame is only certified now to October 31st, 2024 because all the parts have got to match. So when you get into that kind of high variability, it actually works pretty well.

What I tell people that I always get asked the question, “What industry does it not fit?” The industry where demand driven does not fit is if you are in a highly reliable industry where your customer tolerance time is shorter than your cumulative lead time and you experience no variability of operations, then it won’t work.

Implicit in there, no, I haven’t found that place in the world, but hey, you know, theoretically, you could push it to that point. The more variability, volatility, uncertainty, complexity, and ambiguity there is, the better it works because it was designed. Demand Driven Adaptive Enterprise was designed for today’s VUCA world, and it works in today’s VUCA world.

Conor Doherty: I’ll let you respond.

Joannes Vermorel: Yes, I would just take this example as, again, to the audience. Okay, let’s talk aviation. So we have parts that have flight hours and flight cycles in them. I’m just making it very simple for the audience. So that means then when you have the way you look at your inventory, you cannot look at, “I have one unit, two units, three units, five units.” It doesn’t really make sense because every unit that you have has a certain amount of flight hours in it and flight cycles, by the way.

So you can end up with thousands of flight hours but with just one part or maybe just 100 flight hours but you have two parts for some reason. So the bottom line is that what you cannot do anymore is to have a one-dimensional representation of your SKU. So you cannot say, “I have one, two, three, four, five extra units.” You need a many-dimensional representation of the SKU.

And again, if I go back to DDMRP and everything that is in the books, those points are never ever touched, not even alone the sort of anything that could address sour points. They are not touched. I guarantee to this audience you will not find anything that lets you deal with many-dimensional SKU problems. And yet it is literally the sort of nonlinearity and complexity that the books set for itself as goals in the very beginning.

Carol Ptak: I agree with Joannes, absolutely. Yes, we do not cover multi-dimensional parts. Does that mean we don’t know how to do it or don’t know how to implement it? Absolutely not. My background is in aerospace. I did a lot of work with the NAA aviation depots at Cherry Point, Jacksonville, and down in California, as well as with the helicopter companies. I mean, that was my career. If you do a research, I spent 20 years in aerospace.

So I understand multi-dimensional parts because you’ve got different SKU numbers with different condition codes with different flight hours on them. And Joannes, you are absolutely right. We do not cover multi-dimensional parts in any of those books. Now, I mean, if you want to read an ERP book, my ERP book is the first time that remanufacturing ever appears in a book. But it’s such a specialized environment that if we put everything in about every single environment that’s out there, those books would be 3,000 pages.

Those are the fundamentals, the building blocks to any demand-driven adaptive enterprise. There are different dimensions that you add, like we’ve already discussed retail, you know, we discuss aerospace, we discuss remanufacturing, project management. How about a company that never uses the same material twice? Very successful implementations of demand-driven. So those books represent the building blocks.

You know, it’s like you mentioned earlier about, you know, if I read about flight, yeah, I’m going to read the books and I’m going to understand Bernoulli’s principle and all that, but it’s not going to make you a pilot of a 747.

Conor Doherty: It would make me an engineer just to complete that simile. But Joannes, your…

Joannes Vermorel: No, I think again we are faced for the audience with an authority argument, which I mentioned initially, which is, “Trust me.” So anyway, I suggest we move on just not to circle back to the same arguments.

Conor Doherty: Well, okay. Well, this one is to Joannes. So APICS and ASCM also emphasize the importance of the push-pull frontier. In your solution, at what point in the supply chain network do you transition from a push to a pull approach?

Joannes Vermorel: First, the distinction between push and pull are again from the wrong baseline. So we are going back to the sort of 1970s mindset where they assume that the different parts of the organization cannot talk to each other. So indeed, you have to have one party that decides when to push or the party decides when to pull. But again, this is kind of nonsense in this age of the internet. Why is that? Very simply, you can put an intelligence on top, artificial or not, doesn’t matter, as long as you have a network.

The only thing is to trigger decisions. If you decide to move 10 units from point A to point B, I mean, it is just a perspective to say that if it’s point A that calls the units, then you’re pulling. If it’s point B that decides, then you’re pushing. Again, this is not a valid distinction in this age of the internet. So my take would be, please do not maintain the sort of concepts that have been made obsolete 25 years ago pretty much by the idea that you have an internet network and so information can flow freely across your supply chain.

At Lokad we don’t really deal with that because it’s a problem that is obsolete, and it only exists in companies that persist in using, I would say, obsolete methodologies and obsolete perspectives.

Conor Doherty: All right, there are two more questions and then we’ll transition because it’s been a while now. But Joannes, go first. How effective are traditional safety stock calculations for a company managing both high volume and high variability in its operation?

Joannes Vermorel: Safety stock is broken by design on so many fronts. I will keep it short, but the bottom line is that why is it completely broken? Whenever you invest $1 in your supply chain, this $1 competes for all, let’s say, investments in inventory. It competes with all the SKUs. All the SKUs compete for this $1. Your safety stock model assumes that you can process a SKU in complete isolation, disregarding everything that is happening on the other SKUs. That’s literally the safety stock model.

So on this premise alone, safety stock is completely broken. And then you have a second problem, which is an implementation detail, but in practice, it’s really a killer, which is the normal distribution assumption that is made on top of that. So safety stock invariably means both in textbook and software implementation, normal distributions to be used for demand and for lead times. And this is insane.

So the big problem is, again, all SKUs compete for the same investment. So any SKU-independent sort of logic is broken by design. And then you have a second problem, which is the math that is being used, which is really inadequate.

Conor Doherty: Thank you. Carol, your thoughts?

Carol Ptak: I’m excited I found another point of agreement with Joannes. Safety stock is fundamentally broken, absolutely. All right, it’s one of the two things that we eliminate in the demand-driven methodology. And the reason is because safety stocks, as they’re computed through any MEIO optimization software, assume that to have better customer service, you must have more inventory and that you can calculate the amount of safety stock required, as Joannes said, in isolation, SKU by SKU, by looking at the variation and looking at the z-score for your desired customer service level.

That’s ridiculous. That is absolutely ridiculous, and we call that a deep truth. A deep truth can only be exposed by a deeper truth, which is, again, it goes back to that Post-It note I wish I could put in Joannes’ office: “It’s all about flow.” When we have better flow, we get better customer service with less inventory at the same time. It is not a tradeoff, you know.

The MEIO systems that try to optimize those two positions of the amount of inventory to customer service are absolutely fundamentally broken, and the demand-driven approach does not use safety stocks. So I agree with Joannes, absolutely spot on.

Conor Doherty: All right, and again, we’ll do one last question. There were other questions, but again, I want to move on to the next section. Anything that was not answered, we’ll address on LinkedIn. But this is actually a question that came, Carol, from someone who’s a fan of yours, in fact. I’m not going to say the name, but someone who was a fan of yours. So this actually does come in good faith and good spirit.

So, Carol, to you: If Joannes’ critique is completely incorrect, if he’s just completely off the mark, then why, in your opinion, do you think the demand-driven approach isn’t more widespread or more popular?

Carol Ptak: Well, that’s interesting. You know, I don’t… his critique… all right, let me back up. My disappointment was I thought our debate today would be about the methodology, not about page numbers and labeling things, graphs, and figures, which we are mandated to do by our publishers. So I was disappointed at the depth of our discussion today.

I think the questions that we’ve had going back and forth at the end were the better part of it, rather than Joannes reading his pre-prepared notes as he came in. So I was looking for a more dynamic discussion. Why is demand-driven not more prevalent? It’s actually very well known in certain countries, and it depends on the team that’s in the country. In France, very, very well known, which is why we’ve had a target on our backs with Joannes for many, many years.

He’s been going after the demand-driven methodology for many years because of its visibility in France. Our number one country is France. Number two is Colombia. Number three is Mexico. We’ve just expanded to Japan. The United States is growing like crazy. So we’re seeing some very large consumer products companies, like Fortune Brands, that’s been implementing. We have some lesser-known brand names like Toyota and Caterpillar that have been implementing.

So I would challenge the fact that it’s not more well known. It’s been very, very large companies that have typically embraced the idea. We’ve had some small family-owned businesses as well because they understand the impact and the importance of cash flow. The more exciting thing is we expanded to China during the pandemic, and we are just now expanding to Japan. The team in Japan says, “You know, we realize that demand-driven is what we’ve been missing because the Kaizen approach is limited, and we need a breakthrough idea.” They believe that demand-driven is that as well.

So the fact that our demand-driven dictionary is in 12 languages, the exam is in nine languages, I would challenge that it’s not more well known. Those of us in the community tend to look at how many companies are not using it as opposed to the size and breadth of the companies that are. To Joannes’ point, a lot of companies, once they’ve implemented it, you don’t hear their case study because they view it as a competitive advantage, and that’s unfortunate.

Conor Doherty: All right, Joannes, I’ll slightly tweak the question because obviously the reasons why you think it doesn’t work won’t necessarily align with people who don’t have, again, your level of academic training. So why do you think, for other practitioners, it is not more widespread, more adopted?

Joannes Vermorel: I mean, factually, I would say very, very factually, because my take is that it is exceedingly ambiguous. There are some methods, if I were to compare to other supply chain theories—not mine, again, let’s put my own stuff out of the equation—let’s say if I were to go to rival theories, let’s say flowcasting, for example. I don’t believe in flowcasting either, but they are extremely specific in their theory, extremely, extremely specific.

So if I want to implement a flowcasting software, I can take the flowcasting book—it’s called flowcasting—and literally they give me everything I need. It is with almost zero ambiguity on what I need to do to implement it. I’m not saying that flowcasting is good; actually, I think it is quite terrible. But to the credit of the authors, their theory that they present is not ambiguous and not vague. Here, DDMRP, I would say the primary criticism would be it is exceedingly vague, extremely weak, and it is very difficult to frame something.

If I were to take my hat off as a software editor and say I want to implement that, that is so incredibly vague I don’t even know where to start. Sorry, and I know that it’s a subjective thing, so I can only say to the audience, pick one of those books, read 10 pages at random, and ask yourself the question: “Can I take what has been said and unambiguously do something for my company?” Unambiguously. Ask yourself your own question, and the answer that you get to this question that you ask yourself should be the real judge on whether what I’m saying is correct or just nonsense.

Carol Ptak: Well, I would challenge that any book that you pick up and read 10 pages, you’re not going to get the complete picture. The way all of our books are written is first we describe the problem, then we describe the direction of the solution, then we describe how the solution solves the problem, and then we handle what we call the impediments, the “ya-buts”, and then we describe a safe path forward. So taking 10 pages at random, I don’t think in any book would get you where you’re going.

But I’d have to sum up today’s discussion as: It’s all about flow, and roughly right is better than precisely wrong.

Conor Doherty: Well, at this point, I don’t have any further questions, but I’ll just open the floor. Joannes, is there anything you wish to pose directly to Carol without my supervision?

Joannes Vermorel: No, I would like to thank Carol for doing this exercise. I really appreciate it. It has been a true debate. I mean, the point was not to reconcile my views. I’m not going to convince you, and you’re probably not going to convince me, but I really appreciate that you took the time and the effort to have this discussion. For me, it means a lot, and my goal onward would be to have more of those debates. Obviously, there are other theories, so that’s a goal that I set for myself for this channel.

I am very glad that, again, Carol did dedicate a solid, what, 90 minutes of her time. That is, I really appreciate that, and I would like to thank you, Carol, for that.

Carol Ptak: Well, very welcome, and I appreciate the invitation. I had hoped that we would be able to do the debate face-to-face, but then the pandemic hit, so that postponed it. So I’m glad that this opportunity came back up because, if you recall, I had committed to you that I told you anytime, anywhere, I’d be happy to do that debate because I think it’s very important to bring the complete information out to the market and debate those points.

I think, as in a debate, somebody can decide exactly which road they want to go down, and it’s just fine. As I said earlier, if I had to summarize it, demand-driven is all about flow. Roughly right is better than precisely wrong.

Conor Doherty: Well, Carol, I do know that I heard somewhere that France is the number one country for DDMRP implementation. So the next time you happen to be in France, if you’re in Paris, again, I know we would both be very happy to host you, if for nothing else but dinner.

Carol Ptak: That’s my favorite. My guys down in Toulouse, they know when I hit there, it’s got to be foie gras and it’s got to be duck breast. I get my canard, I get my foie gras, and I’m a happy little camper.

Conor Doherty: Well, at that point, I will draw things to a close. Honestly, it’s been quite enjoyable to listen to you go back and forth, I will say. This was several years in the making, I think it’s safe to say. So if it wasn’t edifying, I hope it was at least entertaining for everybody. So again, thank you both very much.

Carol Ptak: Conor, I think you did an absolutely fantastic job, and I appreciate it. As I said, Joannes and I have been talking about this for several years, so I’m glad we were able to finally make it happen.

Conor Doherty: On that note, I’ll draw things to a close. Joannes, thank you very much for your time. Carol, you’ve been a great support. Thank you very much for yours, and thank you all for watching. We’ll see you next time.