00:00:00 (Re)Introduction of Knut
00:01:51 Knut Alicke’s work on supply chain resilience
00:02:59 Companies’ response to first lockdown
00:04:15 Joannes’s perspective on supply chain changes
00:06:35 Defining risk and resilience in supply chain
00:10:06 Knut’s key ingredients for resilient supply chains
00:13:09 Importance of end-to-end visibility
00:14:42 Importance of data interpretation
00:15:55 Case study: Pharmaceuticals
00:17:28 Software-driven supply chain disasters
00:19:28 Lokad’s approach to machine learning tools
00:21:21 Sophisticated software making companies fragile
00:28:32 Complexity of supply chains
00:30:29 Benefits of probabilistic approach
00:33:08 Factoring in inflation risk
00:40:33 Supply chain resilience as insurance
00:44:32 Explanation of the CHAIN model
00:50:00 Example of B2B retailer service
00:52:12 Importance of dollar-based metrics
00:58:41 The effectiveness of automated systems in risk management
01:00:37 Example of aircraft maintenance narrative
01:04:11 Critical skills in supply chain
01:05:31 Importance of clear writing
01:08:16 Knut’s call-to-action
The pandemic has forced companies to reassess their supply chains, focusing on risk reduction and resilience. In this interview, Knut Alicke of McKinsey and Joannes Vermorel of Lokad discussed the need for systematic planning, digital tool utilization, and software automation. Alicke emphasized the importance of visibility and pre-warning systems to detect potential disruptions, while Vermorel highlighted the need for a digital culture to understand data nuances. Both agreed on the importance of scenario planning and a probabilistic approach to manage potential issues. They also stressed the need for strategic thinking, effective communication, and cultivating options in supply chain leadership—things Alicke covered in detail in his recent (co-authored) book, From Source to Sold.
The recent pandemic has forced companies to re-evaluate their supply chains, with a focus on reducing risk and increasing resilience—as explained by Knut Alicke, a partner at McKinsey, and Joannes Vermorel, CEO and founder of Lokad.
Alicke, who has been working in supply chain for almost 30 years, noted that companies have had to become more systematic and agile in their planning processes. However, he pointed out that there is still a gap in terms of experience in supply chain and the best use of digital tools. Vermorel, on the other hand, emphasized the importance of software automation to handle mundane decisions and tasks, freeing up time for people to focus on unusual or extraordinary situations.
Alicke discussed how companies have reacted to disruptions in the past, such as the Fukushima disaster in 2011, and more recent shutdowns and lockdowns. He noted that while many of the ideas for resilience existed years ago, they were not seen as important. Companies would often return to normal operations after a disruption, focusing on lean and cheap supply chains rather than resilient ones.
Alicke emphasized the need for visibility and a pre-warning system to detect potential disruptions in the supply chain. This could be issues with a supplier’s supplier or problems with logistics, production, or quality. He also highlighted the importance of planning, particularly scenario planning, to mitigate potential delays or disruptions. This could involve expediting shipments, substituting products, or flying in alternatives.
Vermorel agreed with the importance of end-to-end visibility but suggested that companies often lack a digital culture to understand the nuances in their data. He argued that the problem is not the lack of data or its quality, but the lack of understanding of the data.
Vermorel also discussed the importance of understanding what an algorithm is trying to do, rather than how it works. He noted that software allows for rapid scaling, including the potential for large-scale mistakes. He also pointed out that even relatively simple calculations can become opaque due to the limitations of the human mind.
Vermorel further explained that even if data scientists replace planners, the same problem of opacity persists. Some machine learning tools are opaque even to those who use them, and understanding the algorithms doesn’t necessarily mean understanding the results.
Vermorel discussed the operationality of scenarios in supply chain management, explaining that maintaining multiple scenarios can be high maintenance. However, a probabilistic approach, which considers all scenarios at once, can be more manageable with the right mathematical and software tools.
He explained that this approach allows for the consideration of various potential issues, such as a warehouse having a 1% chance of being flooded each month, without needing to know the exact cause.
Vermorel likened the probabilistic approach to a quantum perspective, where all possible futures are considered and mathematical instruments deal with infrequent phenomena.
Alicke agreed and emphasized the importance of companies being prepared to take action based on the insights gained from scenario simulations. He noted that companies often lack the readiness to implement solutions even when they have the necessary insights.
Vermorel discussed the importance of cultivating options in supply chain management. He explained that the probabilistic approach allows for the constant consideration of options, such as alternative transportation modes, which can be activated when conditions are right.
Alicke shared an example of how scenario planning helped a client become more resilient by identifying a bottleneck resource that required 12 weeks to move from one plant to another.
Vermorel discussed the importance of strategic thinking in supply chain management, which can be hindered by constant firefighting.
Alicke emphasized the importance of communicating the need for strategic investments to the board, likening it to paying for insurance. He noted that this requires a strategic decision by the board and the ability to effectively communicate the story to them.
Alicke also discussed the inspiration behind his book, “Source to Sold” (co-written with Radu Palamariu), which includes interviews with people who have made it to the board with a supply chain background, and discusses the chain model they developed based on these interviews.
Alicke explained that ‘C’ stands for collaborative, ‘H’ for holistic, ‘A’ for adaptable, ‘I’ for influential, and ‘N’ for narrative. He emphasized the importance of building relationships, understanding the big picture, adaptability, empowering people, and using the right language to explain things.
Vermorel discussed the fear of second order effects in supply chain, such as the expectation of discounts by customers. He argued for the need to have a KPI that includes judgment calls and forces a long-term view.
Vermorel criticized the lack of imagination in taking into account elusive factors that are difficult to measure. He emphasized the importance of developing narratives to convey technical and rational things in a concise way.
Vermorel argued for the need to have insights that deeply resonate with what businesses are trying to do, as opposed to relying on easy metrics that are irrelevant to the problem at hand.
Alicke agreed, adding that numbers support the narrative and help identify root causes when something goes wrong. He emphasized that effective leadership requires people with the necessary skills to activate the vision expressed through the narrative.
Alicke suggested that everyone in supply chain should understand end-to-end processes and train colleagues from supply chain and other areas. He mentioned that he and Vermorel teach at universities to increase the community’s capability and promote supply chain as an interesting and important topic.
Vermorel added that clear writing is a crucial skill for collaboration, creating narratives, and organizing reports. He criticized the low quality of writing in many departments and encouraged students to improve their writing skills throughout their lives.
In conclusion, the interview highlighted the importance of understanding and managing risk and resilience in supply chains, the role of data and algorithms, and the need for strategic thinking and effective communication. It also emphasized the importance of cultivating options, understanding end-to-end processes, and improving writing skills.
Conor Doherty: Given the recent pandemic, most companies have re-evaluated their supply chains with an emphasis on reducing risk and increasing resilience. Today’s guest, Knut Alicke, has written extensively on these issues as well as supply chain leadership in his new book, “From Source to Sold”. Knut, welcome to Lokad.
Knut Alicke: Thanks a lot for having me.
Conor Doherty: Well, I said welcome to Lokad, but it’s probably more accurate to say welcome back to Lokad. You were with us, I think, 3 years ago, almost to the day, in fact.
Knut Alicke: That’s correct. This is my second episode with you. So, it was three years ago, you’re right. We talked about the future of the supply chain, work skills, and everything. It has been an interesting 3 years for all of us with a lot of disruptions and a lot of things going on in the supply chain.
Conor Doherty: Absolutely, and we’ll get back into that. But for anyone who might have missed that episode, could you maybe reintroduce yourself to the audience, please?
Knut Alicke: Sure. So, my name is Knut Alicke. I’m working for McKinsey. I’m based out of our Stuttgart office in Germany, and supply chain is my passion. That’s what I’ve been doing for almost 30 years. So, next year it will be 30 years. We’re getting older and older. What I do here is basically all topics in terms of planning, so forecasting, S&OP, supply planning, production planning, inventory, but then also physical flow, warehouse optimization, transport network optimization, setting the right governance organizational structure.
Over the last three years, I’ve been clearly working on supply chain risk and resilience to help our clients to be better and to have a more resilient supply chain. And next to McKinsey, I’m still teaching. So I’m developing, so to say, the new class of supply chain professionals because that’s what we always hear, that we don’t have enough supply chain professionals. We don’t have enough people who really understand end to end, who understand trade-offs, and who are into the topic.
Conor Doherty: Well, actually, if we can just then go back to what we were discussing. Because in, you mentioned again, it was three years ago, we talked about the future of supply chain and the skills required. That was mid pandemic. In the years that have followed, now we’re basically post pandemic, do you think that the situation has changed? You know, risk and resilience have become more of an issue. So, is it the same skill set required or has that changed?
Knut Alicke: A lot happened. If we just look back three years, a lot of companies started, kind of after the first lockdown, to set up firefighting war rooms, control rooms, whatever they called it, and to solve problems. That was not always done in a systematic way. That was not always done in kind of really thinking about end to end. And then they realized that, hey, we need to do more. Right? We need to prepare, we need to make sure that we have the right visibility in place, we need to have the right levers in place that we can pull, and we need to make sure that our planning processes are agile enough and fast enough.
So, a lot of companies did reduce their planning from a monthly planning to every two weeks, and in the S&OP, the operational planning from a week to every two days. And that all requires talent. That requires talent that understands supply chain, that understands digital and brings everything together. And what we see here is there is still a huge gap. The gap got smaller. People did, I would say, educate their own staff. There was a lot of hiring going on from external sources, but still, there is a gap in terms of experience in supply chain, how to best use digital tools to plan and to improve the performance of the supply chain.
Conor Doherty: Well, thank you. Joannes, you were on that panel as well. Have you changed your perspective in the intervening years?
Joannes Vermorel: I mean, evolved, yes. I don’t know to what degree it counts as a change, but the gist of it is, for my perspective, the more disruption you face, the more automation you need. Because you see, if your routine is already keeping everybody busy, you know, firefighting and dealing with the mundane, if you’re already at 100% busy just coping with the mundane, when the extraordinary hits you, then you have like zero leeway into coping with this extra stuff.
And I don’t mean in supply chain capacity or hard assets, but just the mental bandwidth to deal with a problem. If everybody in the organization is already at full speed just to keep the company operating on a normal day, when you have an abnormal day, then everything kind of explodes or gets delayed. So, and I don’t have, I would say, a silver bullet to free up this bandwidth. However, one of the next best alternatives to a silver bullet is extensive software automation.
So that at least all the mundane decisions and mundane stuff gets out of the way, robotized, and so people have the time to focus on what is fairly unusual. And by unusual, I don’t mean the usual fluctuations of the demand being a little bit higher, a little bit lower, or all the times similarly varying. I mean structural change where you have suppliers that disappear, suppliers that become way more expensive with no going back to the earlier state of affairs, tariffs, or things that really modify the structure of the market in which you operate.
Conor Doherty: Well, it occurs to me in a discussion on risk and resilience, it probably would be better to actually define the terms. So, Knut, if I can come back to you again, post pandemic, people talk about the importance of risk and resilience, but I mean, risk and resilience existed pre pandemic. So, in your expert opinion, how exactly have these concepts changed? Like materially, how have they changed as a result of the pandemic?
Knut Alicke: The good question is whether they changed. If you think about, like, what was it, 2011 when we had Fukushima? That was like 12 years ago, and companies reacted in a way they also reacted to the recent shutdowns, disruptions, and lockdowns. So, I would say that a lot of the ideas did exist many, many years ago, but they were not seen as important. Companies would not focus on that. They would say, hey, the disruption is over, let’s go back to normal and let’s just make sure that our supply chain is as lean as possible, as cheap as possible, but not as resilient as possible.
So, if you think about what is necessary to be resilient, we need to have the visibility in place. So, we need to have something like a pre-warning system that, hey, something is cooking in let’s say tier three, tier four. So, not our direct supplier, but the supplier of the supplier of the supplier has some problems. Maybe there’s a logistics issue, maybe there’s a production issue, maybe there is a quality issue.
We know exactly that this will manage all the way down to us, to our production line, and will cause a disruption. If we know that early enough, we can react. Or let’s say, hopefully, we can react. To be able to react, we also need to make sure that we have something like planning in place. So, if we, for example, then see that, oh, this container will probably arrive two weeks late, then this information itself is not helpful. The information that this two weeks delay leads to a stock out in our components, that leads to a production stoppage because we cannot assemble whatever we want to assemble, or we have an availability issue with this, we cannot deliver to the retail store that desperately needs our products, this is very important. And for that, we need scenario planning.
So, we need to analyze what is it that we can implement to mitigate this delay. Is it that we need to expedite the shipment? Is it that we substitute the product? Is it that we need to fly in something else to make up for the delay? And this is where a lot of companies still have a problem. That, hey, we create one plan, but we are not able to create plans just in case a disruption happens or some delay happens. And this is super important. If you now think about what is necessary to do that, we need to have data in place, master data. We need to have the capabilities in place, we started to talk about this, and we need to have the organization in place that also accepts that, here in this scenario, we come to the conclusion that air freight is the solution, and then we go for air freight. All of this needs to happen to make sure that we have a resilient supply chain that is still able to deliver.
Conor Doherty: Well, actually, again, you’ve identified three ingredients, and that was something that you mentioned in a recent survey that you wrote at McKinsey on tech and regionalization. You mentioned the most resilient supply chains have end-to-end visibility, high-quality master data, and do effective demand scenario planning. So, Joannes, to turn it back to you, why do you think that those are absolutely critical ingredients to have for a resilient supply chain? Or would you add anything else to that?
Joannes Vermorel: Yes, I mean, from my perspective, the challenge with data is very specific in the sense that data quality is usually excellent. That’s weird, I know most vendors complain about bad data, but the reality is that when we look at, let’s say, Western companies, maybe not Asian companies, but Western companies have been digitalized for three decades and usually in terms of accuracy, when there is a record saying this thing has been sold on this day, this quantity, it is 99.9% accurate. So yes, there are some clerical errors here and there, but it is very accurate. Now, the problem is not that the data is usually incorrect, it’s that the semantic is very fuzzy.
Just to give an idea, most of our clients, and I’m thinking of the bigger ones like public companies, usually it is very fuzzy to count what they have in stock. The problem is not that they don’t have the data, the problem is that just imagine you don’t have one ERP, you have 20 ERPs and they all count the stock not in one way but in 20 different ways. And then the stock is not binary, it’s there or not there, it can be on hold at the customs, it can be on hold for quality testing, it can be in storage, it can be reserved for some clients somewhere. So, you see there are plenty of complexities.
And then when you think about the demand, same thing, it becomes very quickly very fuzzy. Let’s consider for example a B2B distributor. You’re selling to businesses so usually you have multiple order dates, not one. You have the dates where the client tells you they want this in the future but it’s not a firm order, it’s in the future. And then there will be a date when they pass the order and then there will be a date when they want a portion of the order first delivered and then another date for the second part of the order being delivered and so on.
So, I absolutely agree with the end-to-end visibility, it is a critical ingredient. But where I think that companies are frequently lacking is that there is a lack of digital culture to apprehend the nuances that get into this data. The problem is not so much that the data is bad or that they don’t have the data, the problem is that they have literally thousands and thousands of tables and people are drowning into bad KPIs, simplistic recipes and whatnot that just do not tell them what they need to know.
For example, companies that operate multi-on supply chain, we have seen people think about service levels in the middle of the network, but the service level of the middle of the network tells you nothing about the perceived quality of service from the client side. It’s pure artifacts. So, I would say those problems are the same but there is some twist in the way you look at them and that’s where I think the biggest skill gap is.
When we talk about master data, what does that mean to have mastery of the data? That’s kind of a play on words, but I would be more on the problem is more the mastery of the data rather than the lack of data or the lack of quality data.
Conor Doherty: So, Knut, just to throw it back to you, do you agree that it’s more a matter of actually how you interpret the data richness or the data source and not the intrinsic quality of the data?
Knut Alicke: Honestly, I saw both, but I would agree that to use the data and create insights from the data is very important. Let me just add one element to this because that is also what I see in many of our clients happening.
The planner has a system to use, right? Using data and then there is an algorithm and that algorithm does some calculations, a forecast, a production plan, supply plan, whatsoever. What we often see is that there’s much more algorithmic intelligence in there than the planner is ready to use. And why is that? That is because for most planners the algorithm looks like a black box. What they would like to do is they would like to open the black box, look inside, understand and then use it.
For a big pharmaceutical company, we did an analysis after they implemented one of the big planning systems and there were only eight people logging in and using the system. All the other hundreds of planners were logging in but then they logged out very fast and then they logged in again and then logged out. What does that mean? They downloaded all data to their Excel sheets, they did their usual changes and planning and uploaded the results again.
So, one very important element is explainability. We need to create trust in terms of all the algorithms we have. We either need to explain it or we need to have other ways to show that the algorithms are working the way they should work and with this only the planners will finally use all this cool stuff that is out there.
Conor Doherty: Actually, a quick follow-up to that and it’s relevant to something I read at Lokad. I’m not going to say who wrote it, but it was in a paper on MRO and they said that more important than understanding how the algorithm works, it’s more important for the practitioner to understand what it is trying to do. And I’m curious with what Knut just said and what I’ve just said, what is your take on that, Joannes?
Joannes Vermorel: So, I completely agree with Knut in the sense that sophisticated methods introduce new classes of risk. And when you look at some of the biggest supply chain disasters of all time, they were software driven. That’s the Nike disaster of 2004, that’s Target Canada, that’s Lidl that wasted half a billion Euro. So, software lets you do things at scale super fast, including super dumb things. And yes, opacity doesn’t take anything that is super fancy to get super opaque.
The beauty of computers is that the human mind is left behind with only something like 10 multiplications. And then, even if you’re super smart, any modest calculation that does more than 10 multiplications and additions, you can’t intuitively follow up what is going on. So, it doesn’t take brutal numerical sophistication to be absolutely opaque. Even something that is still relatively simple in terms of computer processing power is already way beyond what you can follow.
So, this is a big problem and by the way, even if you replace planners by data scientists, you still get the exact same problem. There are classes of machine learning tools that are very opaque even to the people who wield those tools. So, even if you have a deep understanding of the algorithms, it doesn’t mean that you understand what if the results that you’re looking at is really what you intended. That’s another class of problem.
The way Lokad approached that is mostly by being very opinionated on certain classes of machine learning tools, especially differentiable programming that let you operate with semantic variables. So, the idea is that it’s not any kind of machine learning, it’s the sort of models where every single variable has a name and a semantic attached to it. So, that means that you can inspect what is going on piece-wise in your model to gain an understanding of whether the behavior looks correct.
Just to give an example, if for example we have cyclicities, day of the week, week of the year, week of the month, it means that those cyclicities are going to have named parameters that you can check. There will be literally a variable called the Ramadan effect or the Chinese New Year effect. It may sound very anti-machine learning because we don’t autodiscover the patterns, but the idea that all the patterns are named and so variables have a clear semantic makes it much easier to inspect piecewise the model.
So, even if the output is strange, you can still go and inspect the pieces that constitute the model and it doesn’t take a PhD in mathematics to do that. That’s only part of the solution but the rest needs different methodologies. But yes, the technological risk, I mean introducing sophistication in trying to make your company more resilient, the history is a little bit against software vendors in general terms. More sophisticated software technologies tend to make companies more fragile overall compared to cruder, simpler ways of organizing companies.
Conor Doherty: Well, Knut, to actually bring it back to risk and resilience, I remember in the survey from I think it was November this year, you noted that of the three ingredients that were mentioned, visibility, master data and demand planning, scenario demand planning was or had the least adoption. I think it was only about a third of those surveyed said that they had effective demand scenario planning in place at the company. I’m just curious, why do you think that there was a drop off between the first two ingredients and the last one and what effect does that have on resilience at the company?
Knut Alicke: Planning is not easy. It sounds simple, you just say, “Why don’t you evaluate your overall end-to-end plan for the scenario that says we have less capacity or we have higher demand or the supplier is not able to deliver?” But just imagine that a lot of companies still calculate one plan per week. So it’s still the weekend is required because it takes 14 hours and it blocks a lot of the IT resources.
Even in these days, this is often the case. So how would you tell these companies that, “Hey, please calculate five scenarios where you evaluate different solutions,” where they say, “Okay, that takes a week to calculate.” So there’s the sheer compute is often not there. Then there’s very often, it’s not clear how to populate the scenario. So what should we calculate and how to evaluate, right?
All of the planning solution providers have the ability to calculate scenarios. Then you need to evaluate what is better for our current setup and for our customers and for our supply chain. So they need to be clear on, “Hey, it should be optimized for service, for cost, or for our inventory.” That’s often not clear.
Unfortunately, we still see a lot of S&OP processes or IBP process or end-to-end planning processes coming up only with one solution. And then the discussion is very interesting because you can only accept this one solution. There’s no way that you can say, “Hey, why don’t we do something different here?” So there is a lot to catch up and to improve, to be able to calculate the scenarios, to understand and value the trade-offs, and then come to a joint decision what is the best for our customers or company or value.
Conor Doherty: Well, Joannes, I’ll come to you for a moment. I’ll come to you in a moment because I know you’ll have something to say about this. But just to follow up on that, Knut, when it comes to evaluating the viability of any given scenario, do you see that as being unique to each company or do you think that there is an overarching metric or philosophy that every company could use to evaluate the viability of a scenario?
Knut Alicke: So we always talk about the three most important elements of a supply chain, and that is cost, service, and capital. It would probably start even with service. And then you have trade-offs. Service goes up, “Oh yes, we can do that if we increase inventory or if we increase cost.” Cost down, “Yes, okay, but then service might go down.” So, to understand these trade-offs is super important.
Talking to a lot of our clients, we often do a very simple exercise. We just ask them individually, “What is most important for you? Where would you invest, let’s say, 10 EUR to improve if you would have something or a thousand or 100,000? Is it reducing cost or optimizing cost? Is it improving service level or is it reducing inventory?” And you often get a completely mixed picture. So everyone talks about different things.
So that means that the supply chain strategy is not aligned. If the supply chain strategy is not aligned, how would you evaluate what the best scenario is? Because one part of the company would go for higher service level, often the producing part would go for lower cost because that’s their local incentives. So that is something where you, in the bonus structure, if you look into the bonus structure, that is often contradicting these trade-off discussions for scenarios. So that is something that needs to be addressed, needs to be solved, and then you can decide on, “Hey, this is really the best solution for our company.”
Conor Doherty: Thank you. And Joannes, your take on how to evaluate the viability of scenarios?
Joannes Vermorel: I would revisit a few other things first. Because you see, first, let’s discuss about the compute requirements. That’s something that I hear frequently, “Oh, it takes hours to compute.” But let’s consider that a smartphone, just a regular smartphone, it does out of the box something like 10 to 20 billion operations per second. And that’s a smartphone. If you get into a workstation, an actual workstation, we are very easily, cheaply, into the 100 billion operations per second. If you’re crazy and you go $5,000 and you put graphic cards and GPUs, you’re into the thousand billions operations per second. Again, cheap stuff.
So now the question is, what exactly are you doing with this processing power? Because that’s the thing. At Lokad, we have the typical discussion. I hear people say, “Oh, five scenarios takes 40 hours of compute.” And then at Lokad, we say, “Oh, but we just run about a thousand scenarios a second.” So first, I would say, we have several problems.
First, modern enterprise software has a problem of having layers upon layers that pile inefficiency. And people may not realize, but most enterprise software is built upon 40, sometimes 50 years’ worth of inefficient layers that never went away. And so you lose your processing power by a factor of 1 million, sometimes more, in inefficiencies of having literally this sort of lasagna software design where it’s a piece of software that talks to another piece of software that talks to another piece of software, etc.
For example, if you try to do this sort of things with a transaction system SQL database, it’s going to be insanely inefficient. I mean, when I say insane, by a factor of something like anything between a thousand times slower than it should be and possibly up to a million times slower than it should be. So supply chains as objects for numerical simulations, they are not super complex. Even an incredibly complex supply chain is like 100 million SKUs, maybe 200 million SKUs. A modern video game is now simulating real time about a billion triangles, 60 frames per second. So that just gives you the scale.
So we are talking of something that in terms of modern computes, even a gigantic supply chain, Walmart scale, is small. It’s smaller than your average video game nowadays. So we have that to keep in mind. And so if you have a calculation that takes more than minutes, you need to really pause and consider, “Am I doing something that is truly complicated that really needs all that processing power? Or am I just starting from something that is incredibly inefficient?” So my proposition is that most of the time, we are talking of things that are incredibly inefficient.
And that’s if you approach it the right way, it’s a non-issue. Then the second thing is the operationality of scenarios. My approach, I mean at Lokad, what I discovered that was a bit more than a decade ago, is that the problem with scenarios is that they are high maintenance. If you have a dozen scenarios that you want to maintain, it’s a lot of effort. And the trick, and it was literally kind of a trick, is that if you go for a probabilistic approach where you look at all scenarios at once, and so that means potentially millions of scenarios, then if you have the right instruments, mathematical instruments, and software instruments, it becomes a lot easier.
And that’s surprising because you would think, “Oh, if I look at all the possible futures at once, it has to be much more complicated.” But the reality is that with the right approach, it is not. And the answer is because suddenly all the stuff that you want to consider becomes a lot more manageable. You don’t have to make hard choices about what about the warehouse. Okay, let’s say the warehouse at every given month has a 1% chance of being flooded or suffering something that would severely impact its operation. We don’t need to know exactly what, we just say, “Okay, 1% chance a month that we will lose half of the capacity of the warehouse for whatever reason, a strike, a flood, an electrical problem, a small fire.”
And we can say, “0.1% of chance that we lose the warehouse for six months.” And you know, it’s a guess, it’s okay. And then the interesting thing is that you don’t do that in isolation of the other stuff. The beauty of the probabilistic approach is that you can say, “We add this risk to the warehouse and then we’ll add the risk of having a port in China that is blocked, again 1% chance every month.” That’s an estimation, we can revisit that. But the interesting thing is that you can suddenly parallelize the track to think about those risks.
It’s not you craft a scenario where you decide exactly what are the risks taken into account and which one are not. It’s that you can add a risk for the warehouse, you can add a risk for a port in China, you can add a risk of surge of price for a supplier. And that’s the beauty of it, is that all of that blends in. And in terms of maintenance, once you decide to include a risk, what is left to do? The answer is nothing, because your probabilistic forecast is embedding that and the decisions that come out of the system are risk adjusted out of the box.
Joannes Vermorel: I would say this sort of puristic perspective versus classical scenario planning is that first, you can decompose entirely the way you analyze the different risks. So if you have different people that analyze different risks, they can play with the same system at the same time. And then once you get to an agreement about a level of risk, you immediately get risk-adjusted decisions out of it as soon as you turn it on. That’s it, nothing to do, and that’s the beauty of it.
So, in terms of practicality, if you think that inflation has a 1% risk to be above 20% during the next 12 months, okay, factor that in. And if people agree, then we have that and we have immediately all the decisions that are risk-adjusted for that.
The interesting thing is that when you express things like that, yes, you may end up with a couple of dozens of high-level risks, but they are not very complicated to express and they are not very complicated to maintain. That’s the beauty of it. It’s much easier to maintain a high-level risk such as a 1% chance of 20% plus inflation over the next 12 months for, let’s say, Germany, as opposed to maintaining and crafting a scenario where you would respond to this risk in specific ways.
The probabilistic approach is more like the quantum perspective where we say, well, we look at all those possible futures and we let the mathematical instruments deal with those infrequent phenomena. But in aggregate, they are inevitable. If you pile a series of 1% risks per month, you’re guaranteed over the course of the next 5 years to hit several of those problems. The question just becomes when one of them will happen. You don’t know, but that’s fine.
Conor Doherty: Knut, does that align with your engineering understanding of the situation?
Knut Alicke: That is definitely in line. It would be great to leverage this compute and to be able to have some kind of distributions of responses to discuss this.
For example, let’s say you do these scenario simulations, right? And then you kind of know, hey, with this probability, this and this happens. What is then important is that companies need to be prepared to take levers. Now you know that there might be a disruption, what is next? You need to understand, hey, here I need to have these five things in place and just in case something happens with my early warning system, I would then start to execute.
Often, companies are not really prepared. Even if there is the insight, they’re not prepared to implement the solution.
Joannes Vermorel: I completely agree. And by the way, that’s why in my series of lectures, I introduced supply chain as the mastery of optionality. You need to cultivate options.
Scenarios are one way to make those options more pressing, such as alternative transportation modes. But the problem is that it feels very theoretical until you hit the problem.
My problem with scenarios a decade ago was due to the fact that a given scenario would not come into play most of the time. This 1% chance most of the time doesn’t come, and so there is no readiness for it because nothing in the system is really geared toward the immediate execution of this scenario.
But if you cultivate something where, for example, every single time you make a purchase order, there is the option to have it shipped by freight at a much higher price, it’s always an option that is there. It’s just that usually it is not profitable.
That’s the difference between having the optimization that has the option that is already plugged in, just latent, not exploited because the conditions are not right, versus a scenario where the day this option should come into play, nothing is ready. The people are not used to that, the IT systems don’t immediately respond to the proper decisions, and so people have to think and do a lot of unusual things.
Knut Alicke: Let me give you an example from recent years where we helped a client to be more resilient. We looked into scenarios, looked into an early warning system, and everything, and then found out that if something happens in one plant, we can produce in another plant. But there’s one bottleneck resource, the testing equipment. It required 12 weeks to move it from one plant to the other.
So in your scenarios, you need to decide 12 weeks before, “Hey, do we expect something and should we move?” It was completely new for them. They were always looking into it like whatever 3 weeks before and then, “Oh, it’s too late.” You need to understand the solution space, so to say, the lead time to implement, and then only you can really have a good discussion.
Joannes Vermorel: I think you’re spot on. But for example, the case of testing equipment is very interesting because people are frequently drawn into the mundane emergencies. If you’re already struggling with late suppliers, price surges, renegotiating your contract with clients, and all sorts of other problems, they are a complete distraction.
That means that taking the point of saying, “Okay, we need to double invest and have redundancy in the testing equipment. It’s not going to be super efficient, but in the long run, over the course of let’s say the next five years, there will be a time where it will save our quality of service.” And it’s maybe not that expensive.
That’s the sort of thing where people need time and calm to ponder. If they have to jump from one firefighting to the next, this sort of super strategic thinking just does not happen.
Knut Alicke: Let me just build on this. What I found also super important is how do you tell this story that you just told, that “Hey, we need to have the testing equipment, we need to have a second one.” That requires investment, so all the kind of quarter-end related KPIs will not look good.
That is a decision of the board. And what we often try to explain is we use the analogy of an insurance. You have a car insurance, you pay for your car insurance. If you would translate this into your daily operations, you would say, “Ah, why do I need to pay this car insurance? There is such a low probability that something will happen. Maybe you can just skip it, right? I don’t need it.”
No, you want to have it in the rare case of an accident because then it’s getting really bad and then the insurance comes in. And this is how we think about supply chain resilience. It is something that you develop just in case. It might require some investment, it might require some preparation, but then you’re prepared in case it happens.
The challenge is that most companies think about the next quarter or the next year, but the next disruption might come in one year plus one month. So that is a strategic decision that needs to happen and that needs to be decided by the board. And that’s why this story, to tell this story to the board, is super, super important.
Conor Doherty: When you say telling stories, that almost sounds like leadership, almost like something that might feature in a leadership methodology, something that might feature in a book perhaps?
Knut Alicke: Exactly so, and it’s very nice to see a copy of the book even there. That’s amazing, “Source to Sold”. And indeed, what my co-author and I, Radu Palamario and I, did was talk about why we do not see more people with a supply chain background in boards, right? So as a CEO, also as a COO, why is that?
We joked around that it’s probably because supply chain people speak a different language. They’re so number-driven, they’re so detailed, they don’t see the big picture. And we said that, on the other hand, supply chain people have an understanding end to end. So they should understand the business.
Knut Alicke: We joked around because it’s probably true that supply chain people speak a different language. They are so number-driven, so detailed, they often don’t see the big picture.
On the other hand, supply chain people have an end-to-end understanding, so they should understand the business. We looked into whether we have examples of this. We examined the Fortune 200 and found that only 11% of the companies have a CEO with a background in supply chain. Tim Cook is a well-known example, but there are clearly some more.
We decided to interview a couple of people who made it to the board with a supply chain background. This led to 26 interviews, which we consolidated in the book. We then came up with a condensed version of what we learned, which is the chain model.
The interviews were very interesting. We learned a lot from these people who had very different careers. We had people from all over the world, men and women. It was not so easy to find women, so it’s clear that this is still a white male-dominated field and that needs to change.
We had entrepreneurs, small companies, big companies. The book has received very good feedback.
Conor Doherty: Out of curiosity, in the context of a discussion on risk and resilience, are there any of the interviews that strike you as containing insights relevant to the discussion we’re having now? You can pick anyone, male or female.
Knut Alicke: Literally everyone, because it was the time of the lockdown when we did the interviews. Everyone talked about the importance of being agile, being prepared, being resilient. That’s also what we put into the chain model. The ‘A’ is for adaptable. It’s very important that we understand the risk and are able to communicate the risk to the board.
Conor Doherty: Could you explain the chain model letter by letter?
Knut Alicke: ‘C’ is for collaborative. We need to be collaborative, which we heard in a couple of the interviews. One of the contributors said that they wanted to implement a new S&OP process and he came up with the idea to integrate the suppliers. There were three suppliers that were really important. Everyone in the company was initially against disclosing our production plan to the supplier. But he pushed it through and everyone was very happy. Building relationships internally and externally with customers and suppliers is super important.
‘H’ is for holistic. We need to understand the whole system, the big picture, what happens end to end. This is something that is in the nature of a supply chain person. It’s not necessarily in the nature of some of the other functions where you’re often more focused on what you do.
‘A’ is for adaptable, which we already talked about. The ‘I’ in chain stands for influential. Here, I would say, empower the people around you to be at their best and to contribute.
The ‘N’ is for narrative, which is the most important part for me. This is really about how you explain things. For example, a supply chain person might explain an improvement in service level by saying that our OTIF increased from 89.7% to 91.2%. This does not necessarily tell a lot. If you use a language that would be understood by the board, you might say that we improved our service level and with this we were able to sell more or the customer is more happy and coming back. This is about using the right language, the right narrative.
We always say that supply chain got a seat at the table in the last three years and now everyone understood that. Now we need to make sure that we keep that seat at the table. We need to prove that we are worth keeping the seat.
Conor Doherty: Thank you for your thoughts.
Joannes Vermorel: The interesting thing is that the criticism kind of goes both ways. Yes, the supply chain director should be able to speak the language of the board. But also, the problem that I see is that the underlying software infrastructure that supports the actions of the supply director usually provides indicators that are incredibly narrow-sighted.
For example, service level means nothing if you’re in a business where you have substitution. If the client can still come to the store and technically 50% of the stuff is absent but there is tons of substitution and they just go for a substitute, as it can happen for example in fashion, it’s largely nonsensical.
We have a problem where the supply chain director doesn’t have a narrative or something that makes sense because all the numbers that are being engineered by his underlying infrastructure, people and software, are not quite completely sensible.
Very frequently, nobody had ever quantified in Euros or dollars the quality of service in a way that really matches even roughly the business. They would say, “Oh, we have service level.” But service level is super easy to compute, but does it reflect the perception?
For example, what is the difference between walking into your store today and not finding what I was expecting, versus placing an order six months ago, giving you six months of leeway to get the thing, and then discovering that six months later you’re still not prepared? In one case, it’s too bad, I was unlucky. In the other case, it’s completely unacceptable and amateurish.
The problem with these very naive indicators is that they tend to entirely miss not just the elephant, but the herd of elephants. It’s very bad. I believe that your narrative can also be an injunction into needing to engineer numbers that resonate more deeply with a business.
It’s not just about having numbers. Those technical numbers do not resonate because they’re plain bad. If you tell a number that is either, “We invest this 1 million EUR in extra quality of service,” or “It will cost us 10 million EUR of turnover per year cumulative for the next five years,” then everybody would kind of get it.
The problem that I see is that many of the traditional supply chain practices are a little bit at fault with their supporting vendors. The sort of numbers that you get out of those practices plus their tools are percentages that are very nonsensical.
Anything that is expressed as a percentage is, in my view, usually very deeply suspicious. If it’s expressed in dollars, it’s better. If it’s expressed in dollars over dollars, it’s even better. So, per dollar that I invest or do not invest, what do I earn or lose? It’s usually this sort of level to get a good metric.
Building any kind of narrative that even makes business sense is challenging because you’re operating on fluff, I would say.
Knut Alicke: I like the push that telling the right story also needs to have the right KPIs in the first place.
So, what you’re basically saying is that my example should already be translated and not by the head of supply chain. That would be an ideal situation where even the CEO can understand that by improving certain aspects, I will increase my revenue. I fully agree. We are probably one step before that, but it’s a great vision that you lay out.
Joannes Vermorel: My perspective on your narrative idea is that very frequently what I see is that people, especially in supply chain, generally fear this sort of second order effects. Things that are not in the books.
For example, whenever you have discounts in your brand at the end of the season, you have two problems. First, you give up your margin immediately, but then you create a bad habit in your customer base that expects the discount. So next year, they will wait before they purchase until you give away the same sort of discount.
These sort of things can’t be readily quantified because it’s the sort of things that develop over multiple years, potentially decades. Luxury brands, for example, never do any promotions just so that they don’t let these sort of things develop in the first place.
But back to that, it means that you need to be able to have a KPI where part of your number is entirely made up. It doesn’t mean that it’s irrational or fake, it just means that it’s more like a judgment call that can be very reasonable but needs to be made.
The sort of narrative forces you to have this long view and to factor numerically these sort of things so that you can’t end up with a decision deemed optimal which is in fact incredibly shortsighted.
Another problem that I see is that people are not imaginative enough. They don’t take into account things that in the company, the broader company, people know but due to the fact that it’s kind of slightly elusive, slightly difficult to measure exactly, they would prefer to ignore it entirely instead of having it like super rough but at least present.
Conor Doherty: Well, it occurs to me, just to quickly respond to Joannes with a follow-up. In Knut’s example, when he was talking about providing narratives that make the concept of demand planning a little bit easier, he used the example of insurance and Lokad does have narratives like, for example, the basket perspective that explains the idea of the interrelation and the additional cost of not having something when you need it. That then enables people to understand the second order effect. So, I want to unpack maybe the basket perspective as the narrative that we tend to use to make that easier.
Joannes Vermorel: The thing is that as soon as we start having those factors that are not tangible numbers, that’s what I call second circle of economic drivers. Things that are very important but intangible, they won’t appear in the book. For example, many companies have penalties with their suppliers that they can exercise in theory. In practice, whenever they do it’s open war with the supplier and the trust is lost.
So, when you start optimizing this insurance, the interesting thing is that you internalize the risk and you internalize the risk on things that are never going to be measured. It requires a different sort of thinking.
At Lokad, when we have those sort of systems that run automatically, it becomes a bit like a nice anti-spam system. It’s humming gently but you never see it. It just does its stuff and that’s at some point you might even wonder do I really need this stuff because it’s just humming and there are classes of problems that just don’t happen. But as soon as you turn it off, the problems are back.
I believe that this idea of developing narratives is very important because it’s a way to convey things that are very rational but also technical and you need to convey this message in a way that is very concise. People don’t have the time to be experts in all those sort of risks and balance all of that and compute all the trade-offs.
Test of understanding, is this person really looking at the problem from a perspective that makes sense? Just to give an example, if we talk about, let’s say for example, aircraft maintenance, quality of service, one simple way to approach that is to think in terms of AOG, aircraft on ground. So per dollar invested, how many AOG per year do you avoid? Knowing that when an aircraft is grounded, passengers have to be rerouted and it’s a lot of delays, a lot of cost, knock-on effects on the flight schedule and whatnot.
So if you think in terms of service level, you completely miss the plot because an aircraft only needs one part to be missing for it to not take off. The relevant event is the aircraft on ground problem, not the stock out, etc. Every business needs to have this sort of insight that deeply resonates with what they’re trying to do, as opposed to easy metrics that happen to be available cheaply because it’s prepackaged in the software, even if it’s completely irrelevant to the problem at hand.
I know I have this sort of software bias in my perspective. What narrative do you have, Knut? I love numbers, but the trick is that you would think that numbers are the opposition of narrative, but I would not say so. I would say it goes hand in hand. If you have a way to understand even for yourself the stuff that is going on, that will completely shape the way you engineer your numbers.
So don’t think that the narrative is independent from the numbers. The narrative is literally the story you tell yourself to direct your work as a data scientist. If you get this narrative wrong, it means that most likely you’re doing complete garbage with your numbers. The correctness is not in the mathematical aspect, it’s usually the adequacy between the business and what I’m doing with those numbers.
Yes, there is the factual error where you just multiply a number while you should be dividing, but that’s a super technical error and these sorts of technical errors usually are so immediately harmful to your calculation that they are easy to spot. The problems that are much more difficult is when you’re off in a subtle way.
Knut Alicke: So the numbers clearly support your narrative and it also supports everything that you then do to understand if something goes wrong, where you then kind of go into the details. There’s this thing where you kind of ask and then you kind of go from the missing service, the aircraft on ground, why is that? No availability, why is that? Because we did not have stock, why is that? Because we did not have a good contract with our supplier and so on and so on. And then you find the root cause and then you can solve that.
Conor Doherty: Any form of leadership or whatever narrative you want to propose for leadership, even the chain model, whatever, is still effective. Leadership is still predicated upon having people with the necessary skills to activate the vision expressed through whatever narrative you please. So, Knut, to come full circle again, three years removed, what do you see now as the critical skill people need in supply chain?
Knut Alicke: So I could now repeat the chain model, let’s not do that. You need to have all of these skills. And again, all of us, if I would think about what would be a wish for all of us, is that everyone in supply chain that understands all of these kind of end-to-end things and so on, should make sure that you kind of train colleagues from supply chain and from other areas.
That you make sure that you increase the level of capability, that you make sure that you increase the pipeline, so to say. Joannes and I are teaching at universities for exactly this reason, to teach practical supply chain and with this increase the community and spread the word. Make it super clear that supply chain is a super interesting topic and it’s also opening up the path to the board.
Often people ask, “If I’m in supply chain, maybe that’s a dead end?” No, that’s not it. It’s the topic that was one of the most important over the last three years and will be going forward.
Joannes Vermorel: I very much align. I think in terms of skills, there is probably one, if I had to mention only one, it’s not programming, it’s clear writing. Because all of the idea of collaboration, your large company, it’s distributed so it’s going to be in writing most of the time. Yes, you can have meetups, but most of the time it’s going to be in writing.
You want to have a narrative, again it’s going to be in writing. You want to organize your reports and whatnot, again in writing. And one of the qualities that I think is the most underappreciated in modern corporations, especially in supply chain, less in other departments like marketing, is clear writing.
Very frequently I see that the quality of writing in those departments, generally speaking, is very low. So you have very confused summaries about problems, very unclear problem statements. Even when people are asked to give me a half a page description of their job position and why it exists in the first place, usually the outcome is absolutely terrible.
And that’s a big problem. I think that there are some industries or functions where people have been cultivating for a long time clear writing. Finance is one, where usually it’s very concise to the point. Marketing is another one, out of necessity. If you want to have a good branding, you need to be able to convey things clearly and concisely.
There are some industries like software that are very much written and so I would say on average, compared to other industries, the quality of writing is quite good. But overall, I think for students, writing is still a weak skill that can be improved during the course of their life. It’s not like out of university you’re done, it’s something that you can learn afterward as well.
Conor Doherty: As is custom on Lokad TV, Knut, we’ll give you the last word. Is there anything you want to mention?
Knut Alicke: You should buy the book on Amazon. If you still need a Christmas present, the book is available. It’s available on Amazon and others. Make sure to buy a copy, make sure to spread the word. Spread the word that supply chain is cool and build the network.
Conor Doherty: All right, well on that note, Joannes, thank you for your time. Knut, thank you very much for yours. And thank you all for watching. We’ll see you next time.