00:00:08 Debating the claims and benefits of digital twins in supply chain industry.
00:01:50 Difference between buzzwords with substance and digital twin.
00:03:00 Lack of innovation at the core of digital twins.
00:05:07 Comparing digital twin to glorified supply chain simulators.
00:08:20 Supposed benefits of digital twins and Locad’s approach.
00:09:34 Introduction to digital twins and the concept of forecasting.
00:11:29 Concerns about accuracy and the lack of metrics in digital twins.
00:15:00 Improving digital twins and addressing core concerns.
00:16:05 Semantic adequacy and the simplified representation of supply chains.
00:18:01 How digital twins integrate with classic ERP systems and the need for clarification.
00:18:57 Discussing the limitations of digital twins based on data accuracy and approximations.
00:21:00 Criticism of marketing communication around digital twins from competitors.
00:22:48 Explaining what an actual digital twin is and its connection to Monte Carlo simulators.
00:24:53 Analyzing the claims of intuitive dashboards and the importance of accuracy in digital twin simulations.
00:27:14 Understanding the non-intuitive nature of simulators and the importance of accuracy for supply chain applications.
00:28:33 Discussing the idea of a supply chain simulator and its accuracy.
00:29:32 Agent-based simulation and its high degree of customizability.
00:30:31 The practicality and value of manually tweaking simulation parameters.
00:33:12 Comparing different decisions based on a wall of metrics and the challenges it presents.
00:35:34 Addressing the question of identifying the best decision using a digital twin.
00:37:35 Low resolution Monte Carlo process in supply chain simulation.
00:38:07 Digital twin as a rally point and adding real time and synchronization qualifiers.
00:39:07 Importance of a system-wide approach for supply chain and its value.
00:39:47 Critique of digital twins: missing elements and focusing on cheap capabilities.
00:40:58 Conclusion and closing remarks.


In an interview, Joannes Vermorel, founder of Lokad, discusses digital twins in the supply chain industry. Vermorel highlights the limitations of digital twins, considering them a buzzword with limited innovation. He compares them to demand sensing and suggests they only offer incremental improvements. Vermorel emphasizes the importance of accuracy in digital twin simulations and the challenge of integrating them with existing systems. While he acknowledges their potential value as part of a holistic, computerized approach to supply chain management, Vermorel argues that digital twins alone are insufficient for significant improvement and should not be seen as a complete solution.

Extended Summary

In this interview, Nicole Zint speaks with Joannes Vermorel, founder of Lokad, about digital twins in the supply chain industry. Digital twins are commonly described as virtual representations of supply chains that can simulate different scenarios for decision-making. However, Vermorel expresses skepticism about the claims made by digital twin proponents and highlights their limitations.

Vermorel explains that “digital twin” has become a buzzword in the supply chain field, serving as a rallying point for professionals facing similar challenges or seeking similar solutions. He argues that the problem with digital twin as a buzzword is its lack of depth and innovation, in comparison to other buzzwords such as deep learning.

Vermorel asserts that he has not seen any significant innovations at the core of digital twins. He compares them to demand sensing, which he previously criticized as vaporware, but concedes that digital twins may have slightly more value. Any innovation in digital twins, however, is likely to be incremental rather than transformative.

The conversation touches on the vagueness surrounding digital twins, as most vendors fail to clearly define their nature and capabilities. Vermorel explains that digital twins are essentially virtual representations of supply chains, often promoted using marketing gimmicks and futuristic imagery. He compares this approach to Facebook’s rebranding as Meta.

Traditional forecasts in supply chain management are typically limited, as they rely on point time series forecasts. While digital twins may have more versatile forecasting capabilities, Vermorel questions their accuracy, emphasizing the importance of measuring it to ensure the virtual representation makes sense.

To improve digital twins, Vermorel recommends addressing accuracy and ensuring semantic adequacy between the virtual representation and the real-world supply chain. He acknowledges that even advanced simulations are highly simplified compared to actual supply chains. He also highlights the challenge of integrating digital twins with existing enterprise systems, such as ERPs, warehouse management systems, and CRMs, which were not designed to collect scientifically accurate data for digital twins.

Vermorel expresses skepticism about the communication surrounding digital twins and their practicality. He notes that historical data collected through ERP systems often presents a distorted view of supply chain reality. Moreover, he questions the usefulness of operating a digital twin and whether the generated insights and KPIs are worth the cost of employing people to interpret them.

Vermorel describes digital twins as glorified Monte Carlo simulators for supply chain purposes, often using agent-based modeling. However, he questions the accuracy of these simulators and the level of trust that should be placed in their outputs. Vermorel acknowledges that simulators can produce visually appealing dashboards, but he emphasizes the challenge of determining the accuracy and trustworthiness of the presented data. He also points out that simulators are complex black boxes by design, and while they can capture non-linear phenomena in supply chains, they may generate unexpected responses when parameters are adjusted.

The conversation focuses on the practicality, value, and limitations of using digital twins in supply chain management.

Digital twins, as Vermorel explains, are a repackaging of decade-old simulation concepts, specifically Monte Carlo processes, made possible by cheap and powerful processing capacities. These simulations can now span an entire supply chain, generating significant interest in the industry. However, Vermorel emphasizes that digital twins should be seen as just one algorithmic ingredient in a comprehensive supply chain solution, rather than a complete solution on their own.

The interview delves into the challenges of comparing different supply chain management approaches, given the multitude of variables involved. Vermorel notes that digital twins enable the measurement of various metrics, such as service levels for each SKU, inventory costs for suppliers, and quality of service for customers. The true benefit of digital twins lies in their ability to simulate an entire supply chain end-to-end, although there are limitations due to incomplete data.

Defining digital twins as a series of vendors repackaging old concepts using Monte Carlo processes, Vermorel acknowledges their appeal in a system-wide approach for supply chain management and the value of bridging silos. However, he maintains that digital twins should be viewed as only one ingredient in a larger solution.

Vermorel’s main criticism of digital twins stems from what they are not, rather than what they are. He believes that many elements are missing from the digital twin concept, and while Monte Carlo simulations are useful, they are not sufficient for truly improving supply chain management. Vermorel suggests that digital twins can be valuable as a rallying cry for a more holistic, computerized approach to supply chain management, but they should not be seen as the sole means of achieving improvement.

Joannes Vermorel shares his insights on digital twins in supply chain optimization, emphasizing the need for a more comprehensive approach to supply chain management. He acknowledges the potential value of digital twins but warns against viewing them as a complete solution. The conversation showcases the importance of understanding the limitations and potential of digital twins, as well as the need for a broader perspective in supply chain optimization.

The founder shared his thoughts on digital twins in supply chain management. Vermorel criticized the digital twin concept for what it lacks, rather than for what it provides. He believes that while Monte Carlo simulations are useful, they are not enough to bring about a significant improvement in supply chain management. He suggests that digital twins can be beneficial in driving a more holistic approach to supply chain management, but they should not be viewed as the sole solution.

Vermorel emphasizes the need for a more comprehensive approach to supply chain management, which entails integrating a variety of tools and techniques. While digital twins have the potential to add value to this approach, they should not be relied upon as the only means of achieving improvements. Instead, Vermorel suggests that supply chain management professionals should consider a range of factors, including data analysis, simulation, optimization, and machine learning, to identify and address the key challenges facing their organizations.

Overall, Vermorel’s insights suggest that digital twins have a role to play in supply chain management, but they are not a silver bullet. He advocates for a more nuanced approach that incorporates a range of tools and techniques to achieve the desired outcomes. Vermorel’s emphasis on the need for a holistic, computerized approach to supply chain management is likely to resonate with professionals in this field who are seeking to improve their operations and enhance their competitive advantage.

Full Transcript

Nicole Zint: Now, Joannes, what does a digital twin really look like for the user from afar?

Joannes Vermorel: My perception of digital twins is that it’s one of those buzzwords where it’s more form than substance. Technological and scientific communities need buzzwords as rally points, so that people who are looking at a problem the same way can come together and have something like a scientific conference or a business strategy. For example, a buzzword with substance behind it would be deep learning. It’s an entire set of scientific and technological undertakings. However, with digital twin, I see the buzzword and some sort of mechanics at play, but when we start digging a little bit on what is underneath, my perception so far is that it’s very shallow. There is no grand computer science revolution, no mathematical revolution, no machine learning revolution, and it’s very hard to even pinpoint any truly innovative aspect of what could be put behind those digital twins.

Nicole Zint: So, you’d say there’s no hardcore innovation at the core of a digital twin?

Joannes Vermorel: I haven’t seen one. It might actually be a little bit better than demand sensing, which we reviewed a few months ago and is just pure vaporware. However, even if there are ways in which I can see digital twins making some progress, it’s going to be very incremental in terms of the type of innovation. Vendors who sell digital twins are phrasing it as a supply chain simulator. Would you say that it’s supposed to be a simulator?

Nicole Zint: Most vendors who are in the process of selling digital twins remain extremely vague about what it actually is. They would say it is like your virtual supply chain, a representation of your supply chain. What is the virtual supply chain?

Nicole Zint: That’s virtual, which means essentially that it’s not your real supply chain. This is essentially a representation of your supply chain. So far, it remains incredibly vague. A battle plan on paper of your supply chain could be a virtual representation of the supply chain. Typically, it’s associated with a computer-aided representation of your supply chain. “Virtual” has this kind of cool, positive connotation attached to it.

Joannes Vermorel: It is also a bit surfing on the sort of virtual worlds, virtual reality sort of things, a bit like Facebook rebranding itself as Meta. This is again the same sort of vibes that I can see. The interesting thing is that, indeed, when you start to try to figure out what it is technically – at best, I can assess because, again, most of our competitors provide an incredibly small amount of technical details about what it is – it looks like glorified simulators about supply chains. Again, prove me wrong, but I haven’t seen elements that let me think that these supply chain twins are anything but fancy simulators.

Nicole Zint: So you say it’s kind of similar vibes as Facebook changing its name to Meta. What do you mean it’s the similar vibes to that?

Joannes Vermorel: I was just referring to the sort of marketing style, the style of communication that goes with pushing the product – nothing else. It’s a way to package it. You would see that every decade when people want to project something that is futuristic. They are not pushing the same metaphors or themes. For example, in the 1950s, everything would be about humanoid robots. You would see plenty of futuristic advertisements where people wanted to think about what the future looks like. You would have people that were literally disguised as robots with something that is very dated today, where you have people covered in metal plates pretending to be robots.

And for artificial intelligence, you would see plenty of people that try to communicate with cognitive technologies as if they were copying the brain. There is part of the imagery that comes with AI to have the brain, the cognitive aspect, as if you had a mind in the machine. Digital twins are playing on another thing. They are playing on the idea of virtual realities, the metaverse, the Matrix sort of vision for the future. Again, I’m not saying this is scientific; it’s just like a marketing gimmick. It’s the way you approach that. And by the way, every single scientific undertaking is attached to a certain way to sell itself, even if it’s pure science. You always have to kind of market it to the community at large.

Nicole Zint: Not a bad thing in itself, it’s okay to have some sort of themes and imagery that comes with it. But I believe it’s important in the world of enterprise software to identify that.

Joannes Vermorel: Why is it important? Well, because people are first and foremost trying to sell stuff to you. It’s not like we are doing that for the beauty of the human mind. It is first and foremost a for-profit undertaking to improve supply chains.

Nicole Zint: Hold on, let’s unpack this a little bit. What are the supposed benefits then of a digital twin according to the communication of many other vendors that could be qualified as competitors of Lokad?

Joannes Vermorel: Lokad does not sell digital twins. I believe we are doing tons of things that are very much under the umbrella of the expected benefits of digital twins. However, it’s a choice; we don’t market ourselves as digital twins. So take this with a grain of salt. I’m basically trying to describe in a way that is not too unfair what our competitors are essentially trying to sell under this umbrella.

Nicole Zint: A comment on Lokad and digital twins, because what we do here at Lokad when we look at our probabilistic forecasting, we look at all the sort of expected outcomes of all the different decisions that you can possibly take to compare them against each other. So essentially, when we think of the claims of a digital twin, it is to be able to project all the different decisions in a what-if scenario and then see the impact of that. Isn’t that kind of a similar way, except a digital twin is kind of more gamified in a way?

Joannes Vermorel: I think what is put forward with digital twins is the supposed superior capabilities in being able to have a much higher degree of expressiveness, so that you can see tons of possible futures and variations in the future. And I would say it’s only fine and well in terms of intent. However, where I’m very puzzled is that as soon as you start to do that, essentially, if you have any kind of projection for any kind of virtual supply chain, then there will be the question of accuracy because what you’re doing is a forecast.

And I can’t help myself to think about all those vendors who have immense struggle when it comes to the accuracy of their forecasting technologies, and that suddenly through digital twins, they make the problem disappear, at least in the marketing brochures. One of the things that I haven’t seen at all in those discussions about digital twins is that everybody is hyped about the idea that you can do so much in terms of looking at all the possible futures. So essentially, you’re doing a forecast, and then people would say, “No, no, it’s not just a forecast. It is much more versatile than that.”

I would say, fine, if it’s a very versatile forecast that lets you look at tons of possible futures, what you are technically doing is known in the scientific community, the statistical community, as a probabilistic forecast. So you’re looking at lots of probable futures and even like policy-driven if you want to be able to inject into that higher-order constructs such as policies, your pricing policy, your worst punishment policies, and whatnot.

Nicole Zint: There is this question of accuracy, and it puzzles me that many vendors who are pushing digital twins don’t seem to realize that there is a massive accuracy problem. In your opinion, is a digital twin essentially a forecast, but packaged differently, so the concern about accuracy goes away?

Joannes Vermorel: The problem is that the word “forecast” in supply chain circles is typically applied to an incredibly narrow type of forecast, which are point time series forecasts. There is an entire range of forecasts that exist. Digital twins are not forecasts if you define forecasts from the perspective of point time series forecasts – they are more than that.

Nicole Zint: Agreed. So, digital twins are not just time series forecasts?

Joannes Vermorel: Let me rephrase that very clearly: digital twins are not forecasts if you define forecasts from the perspective of point time series forecasts. However, if we take a broader definition of forecasts as non-ambiguous and quantitative statements about the future, then digital twins, at least the way they are presented by our competitors, very much fall into this category. My first concern is that as soon as you have a forecast of any kind, there is a question of accuracy. If you don’t even ask this question and don’t develop the tooling needed to assess your accuracy, you don’t know if what you’re doing is any good. You might just be toying with large quantities of numbers, which, by the way, is incredibly easy with modern computers. You can use a lot of processing power, numerical recipes, and mathematical formulas, but that doesn’t mean what you get by combining all of that is scientific or even reasonable. What you can get is some sort of delusion about your virtual reality that has no real high-quality relationship with your actual supply chain. If you don’t even measure your accuracy, no matter the type of forecast you have, you don’t have the slightest clue if what you’re doing is even sensible.

Nicole Zint: So how do you improve on an existing digital twin?

Nicole Zint: But I mean, that’s the first concern you see existing between these systems. If I look at what our computers are doing, there doesn’t even seem to be any metrics. So, if you don’t have a measurement, I’m not exactly sure what they are actually optimizing.

Joannes Vermorel: That’s not the only concern. That was just the first. So, you see again, I’m just saying that if we even want to be able to pretend that what we’re doing is not just a pure empty buzzword, we need to address the core concerns. The first core concern seems to be completely unaddressed by the supply chain vendors who are selling digital twins, which is the problem of accuracy. But it’s not the only concern. We have another class of concerns that is an exceedingly major concern, which is the adequacy at a semantic level between whatever you’re doing in terms of virtual representation of digital counterparts with reality.

Because you see, those digital twins are not like the Matrix, the old movie where you can recreate an alternate universe that is almost impossible to differentiate with the real world. Doing that remains a feat of complete science fiction. We are decades, if not centuries, away from being able to do anything that looks like the Matrix. So, whenever we want to simulate or have a digital counterpart of a supply chain, we have essentially something that is an exceedingly simplified view of the supply chain.

Even at Lokad, when we are doing the most advanced methodization that we can for supply chain, we need to have the humility to realize that even what we consider to be state-of-the-art in terms of sophistication and granularity of representation, or the digital counterpart of the supply chain, is still an incredibly simplified vision of the supply chain.

Moreover, the data that we need to feed this virtual representation, this digital representation, doesn’t fall from the sky. The data that we are going to use comes from enterprise systems, business systems, ERP systems, warehouse management systems, CRMs, EDI extracts, and dozens of other sources. The point is that those systems have all been engineered to operate the supply chain, not to collect scientifically accurate data about the supply chain.

Nicole Zint: So, how does a digital twin blend with a classic ERP system?

Joannes Vermorel: First, we still haven’t even started to address what is actually a digital supply chain, a digital twin for supply chain. You see, people, vendors want to avoid this question. They want the client to directly jump to the benefits they get, etc. But I insist that we need to first list all the concerns to even know if what we’re looking at is something that is even really authentic.

There will be a third class of questions, which are about the expressiveness we can achieve. We have the problem of making a statement about the future based on data that does not have a true alignment with reality.

Nicole Zint: Of your supply chain, there is no such thing. You know, what you have is the historical data as seen through the ERP. This should not be confused with the realities; this is just a very, very distorted view. Fine, we have to be able to work with that, but make no mistake, there are huge approximations that are going on, and those approximations can be very harmful with regard to the objective of bettering the supply chain that we might have using this digital twin. Then, we have probably a third class of concern, which is, how do you even operate over such a digital representation of your supply chain? It’s not clear that just because you have a digital counterpart, that automatically good things for your supply chain will come out of it, especially when people tell you, you will get KPIs or key insights. I would say fine, you have key insights, but essentially until proven otherwise, those key insights mean that the company will have to pay employees just to have a look at them.

Joannes Vermorel: So, you see, when you say that a piece of software delivers you KPIs and insights, essentially, it is still on the cost side of the equation for the company because, well, no matter how interesting those things might be, those numbers might be, essentially, the company has to pay for people to look at them. And so far, it still does not produce return on investment. And, by the way, that’s a thing that I discussed in one of the previous episodes with the bureaucratic core of supply chain. In supply chain, it is always very much tempting to do all sorts of stuff that are bureaucratic in nature. That’s the problem of having highly specialized people dealing with fairly technical tasks.

Nicole Zint: So you’re very critical of digital twins?

Joannes Vermorel: I’m not critical of digital twins. Again, let’s be precise. I say that whenever you present a concept, we have to be very specific about all the sort of challenges that need to be addressed by the thing that’s critical. And my criticism is very precisely of the communications that are associated with digital twins as presented by my competitors. So, the criticism so far is not about the digital twins themselves. We can get to that in a minute if you want, but it’s about what is surfacing in terms of elements of communication. And so far, I would say the sort of elements that emerge for me are striking, as if they were missing the elephants in the room. And not like one elephant, but at least three major elephants are kind of dismissed or disregarded or non-existent, if you want. And I would say that that makes me wonder whether they are paying any attention to the reality, to the reality of the problem being solved.

Nicole Zint: Fair enough, but now let’s get back to the digital twin itself.

Joannes Vermorel: Yes. So one of the claims that I’ve seen vendors essentially saying is that their digital twin is able to essentially have this intuitive dashboard that can instantly let you see the impact of different what-if scenarios. What is your take on that? What would be your first criticism, if you wish, but also what are the benefits that we can get from a digital twin, in your opinion?

Nicole Zint: So, first, I would say, what is an actual digital twin as implemented by the vendors who are vending digital twins?

Nicole Zint: To ask the question, what is it technically? And here you see there is a value judgment, and I would say those are glorified simulators, Monte Carlo simulators more precisely.

Joannes Vermorel: Despite the fact that the amount of technical information is very thin as pushed by many of our competitors, they still have a few screenshots floating around and a few tidbits of technical nuggets. That’s what I’m using to make this statement. Essentially, when people say that they have a digital twin, what they have is a piece of enterprise software that gives you some kind of modeling capabilities. They have something that is very much oriented toward a Monte Carlo mindset. It’s going to generate, with a certain degree of noise, stuff that is supposed to represent the future states of your supply chain. They are typically going to have some sort of things that are more inspired from the agent-based modeling. They try to represent the supply chain network as a collection of agents that have pre-configured behaviors, potentially learned behaviors to a limited degree. Then they just run the simulator and collect metrics as if you were putting probes at specific locations or on specific patterns in your supply chain. So, at the technical level, a digital twin is a kind of simulator, a Monte Carlo simulator that is geared toward supply chain use cases.

Nicole Zint: When it comes to claims on, for example, having dashboards?

Joannes Vermorel: With any simulators, you can put probes all over the place to measure the outcomes from your simulators, and it is very easy to collect thousands of numbers. If you have thousands of numbers, it is very easy to compile them into a visually pleasing way, like a dashboard. The main problem is how much trust and confidence you should have in those numbers, and that brings me back to the question of accuracy. Having fancy dashboards is most certainly something I trust my competitors to be able to do. But let’s be realistic; it is possible to have very pleasing reports with Excel, and it has been possible for three decades. So, it’s very incremental at best for this sort of category of benefits.

Nicole Zint: What would your idea of benefits then be? Let’s challenge another quality you see, that’s one that you mentioned: intuitive.

Joannes Vermorel: That’s interesting because that’s absolutely not my experience with anything that is like a simulator. Simulators are very much black boxes, complex by design, and they are absolutely not the sort of numerical recipes that lend themselves to easy explanation. By the way, Lokad is using quite extensively Monte Carlo processes, simulators, and generators, and it is something that has a very

Nicole Zint: Lokad has a strong affinity to probabilistic forecasts. However, even if Lokad is using those methods, I recognize that they are not especially intuitive, especially when it comes to the results. It is pretty much by design. What you want to capture with a simulator is typically all the sort of non-linearities that you cannot capture through other methods. But as soon as you deal with phenomena in your supply chain that are highly nonlinear, it becomes very difficult and very black box-ish to get a grasp of what is going on. It means that suddenly you just adjust ever so slightly a parameter, and you have an enormous response at the other side of the network, and it was kind of unexpected.

Joannes Vermorel: If the simulator is accurate, then it is good. It means that it gives you a tool to basically gain control and get a better grasp of the unintended consequences of seemingly small actions in your supply chain. However, it all depends on the accuracy of the simulator. It’s not going to be something that is intuitive in any way. It’s going to be very black box-ish at best. And again, it is very much a complex numerical model that is going on when you’re simulating the world’s supply chain. It doesn’t really fall into the category of things that I would qualify as intuitive.

Nicole Zint: So, a simulator is essentially a forecast behind the simulator curtain, where the accuracy is in question. The idea is that we can see the supply chain all in one screen and toggle with different parameters to see the output. On paper, that sounds great, like a magic ball that can see into the future. But, of course, when you don’t question the accuracy, that’s essentially what it is.

Joannes Vermorel: Yes, and there are also plenty of other questions. When you have a simulator, you have agents, which are essentially the building blocks of your simulation. When you say you run a simulation across the supply chain, it means that you’re going to simulate every single SKU to associate, for example, a replenishment behavior or consumption behavior. So, we have plenty of tiny agents that have their own behaviors, and when we let the simulator play out, we just let all those agents operate and give us a potential future state of the supply chain. We can do that many times.

Now, indeed, by design, this agent-based simulation lends itself to a high degree of comparability. You can touch every single agent and modify them. This is something you can do, and indeed, you can have your wall of metrics that you will obtain by just tweaking the parameters.

Nicole Zint: Running the simulator now, that creates a question about whether it is a realistic exercise. We have potentially thousands of SKUs, if we are talking about a large-scale supply chain. Does it actually make sense to pay people to manually tweak the parameter that controls the agent, you know, the modelization of every single SKU, one by one? Yes, you can do that, but is there a point? Is there value in that?

Joannes Vermorel: That’s also another big part of the concern. Absolutely, you can do that, but this is part of the realm of the given capabilities that are very much what you get out of the design. But then there is a question of should you do that. By doing that, you will get numbers, but how do you decide that a tweak is even better than the other if we can just tweak something, say what happens if I order this much of this product? So essentially, you tweak something, and then you see the output of that as in just one scenario.

Nicole Zint: That sounds like a time series to me. So, first, what you’re not going to get typically if you run a Monte Carlo simulator is one scenario; it’s like the aggregation of the average outcome over many scenarios. Well, hold on, but they claim that it’s an intuitive way to see the outcome of different what-if scenarios. So, if I ordered this much, I should get an instant picture of what the future of that would be, not different pictures, just one.

Joannes Vermorel: Yes, I mean, because essentially, what you do with simulators is you average out results. So, it is a time series forecast; at least, it is a point output. The difference is that the sort of time series assumes that you have a vector of information that is granular over time, but it has a point. It essentially gives you an average estimate. What you will get is a point estimate of the consequence of your adjustment. In this regard, it’s fine to do that. You cannot maintain all future possibilities until the very end; at some point, you have to say I have a decision, and I want to assess the economic outcome of this decision. So, this part is fine. What may not be fine is if users have to go through potentially millions of parameters by hand. Then you have something that is heavily impractical, distracting, and time-consuming, and the benefits that you may get in identifying better decisions may not suffice to cover the cost of paying all the people that need to tweak the simulator.

Nicole Zint: That’s one class of problem, and then you have another class of problem, which is that what you will get out of a simulator is a wall of metrics, literally thousands of numbers. So, how do you compare decision A with a wall of metrics, so you have thousands of metrics, and decision B with another thousand metrics, knowing that some are better, some are lesser, some are dramatically worse, and some are dramatically better? It kind of sounds…

Nicole Zint: Like an RFP process, in a way, you have so many different variables. How do you compare when all the variables are different rather than just one?

Joannes Vermorel: Yes, and that creates a real question of comparison. That’s why you have a wall of metrics because when you have a simulator, you can measure every single thing. For example, you can measure the service level for every SKU. So, your output of your simulator, when you average over thousands and thousands of executions, is literally a service level for every SKU, a quality of service for every customer, and inventory costs for every single supplier. The true benefit of the system-level approach of digital twins is simulating not a single SKU, but the entire supply chain end-to-end, as far as you can reach with the data that you have for your suppliers and as far as you can reach downstream on the side end-to-end for the portion of the supply chain that you control.

Nicole Zint: But it seems like the question still remains, which decision is the best?

Joannes Vermorel: Yes, and I would say that’s also another elephant in the room. What do you do once you have those capabilities? As far as I am looking at the digital twin, it looks like to me that it’s a question that doesn’t have answers. There is a psychological trick used by enterprise supply chain vendors and enterprise vendors in general: as soon as people see a piece of software where they can interact and do things, they gain familiarity and, at some point, they like the software. Even if interacting with the software has a game element, the problem is that the game element means that people can like the product, but it doesn’t prove that it does anything good for your company. It takes the focus away from the outcome. For example, if I was saying that as part of company policy, people should play cards for two hours a day, I’m pretty sure there would be a lot of people that would enjoy this activity and say they love it, but that doesn’t mean that it creates added value for the company.

Nicole Zint: So, to conclude here, if we can just circle back to what a digital twin actually is, could you provide your definition?

Joannes Vermorel: My perception is that a digital twin is essentially a series of vendors repackaging decade-old concepts for simulators. What has changed is that the processing power is now sufficiently cheap that you can have a low-resolution Monte Carlo process that spans over an entire supply chain without too much difficulty.

Nicole Zint: Essentially, you take a big machine with a lot of CPUs. Monte Carlo is very easy to parallelize. It is an embarrassingly parallel, technical term problem. And thus, a lot of vendors find themselves with the capacity to create a product that is cheap to implement that does a simulation of a supply chain-wide system. You know that’s what they can do. And then, because they can do it, they can sell it. And because there is no radical innovation going on, they have found digital twin as a rally point to basically make this product more appealing because if I tell you we are going to do a Monte Carlo approach based on stuff that was discovered in the ’50s, 70 years ago, suddenly people would say, “Yeah, really? Is it?”

Joannes Vermorel: So essentially, they do that and then, to basically, I would say, make the thing even more attractive, they add more qualifiers such as real-time and synchronization. I would say good and well. However, with the simulator by design, you’re never going to get anything real-time except if your simulator is incredibly shallow in terms of sophistication. So that’s really something that is questionable. And nevertheless, I see the reason why there is traction for this product is that there is one key element of truth, which is a system-wide, I would say, approach for the supply chain that really deserves a very, very serious look. So, it is a very worthy undertaking to say we don’t want to do a yes to basically bridge the silos and have a system-wide approach that has, I would say, this is a very appealing concept to me. I can see a lot of value to that.

Now, having simulators, Monte Carlo simulators, is one ingredient. But what I’m saying, and that’s would be my point, is that I don’t criticize digital twins for what they are. A simulator is just well and sound. It is fine, you know? It is an established way to do, I would say, probabilistic forecast in a very generic sense. What I’m saying is that I see an enormous amount of elements that are just missing from the picture, and it looks as if people had been working on the capabilities that are cheap to implement, and they are trying to sell them, while, well, there is an entire class of capabilities that are entirely missing, but they are unfortunately much more difficult to implement and much more expensive. But they are the ones that truly make this capability, Monte Carlo simulation, really useful for the supply chain. So my point is that digital supply chain is good if it’s a rallying cry to have a more holistic approach to the supply chain from a, I would say, a computerized perspective. But if you think that Monte Carlo approaches are anything but one algorithmic ingredient, I think that is a very misguided idea that you, with just that, you can actually improve your supply chain.

Nicole Zint: Okay, Joannes. Thank you very much for sharing your insights on digital twins. Thank you for tuning in, and we’ll see you next week.