00:00:15 Celebrating the 100th edition of Lokad TV and answering supply chain questions.
00:01:17 Discussing the origins of Lokad and its focus on supply chain industry.
00:03:15 Supply chain forecasting with bias using quantile forecasts in 2011.
00:04:53 Shifting to a programmatic approach in 2012 and the challenges faced.
00:07:47 Reflections on the early years, mistakes made, and the evolution of Lokad.
00:08:48 The impact of cloud computing on the entrepreneur’s business.
00:11:31 The evolution of their company and major technical breakthroughs.
00:13:26 Exploring Bitcoin, economics in action, and its relation to supply chains.
00:15:34 Growth of the company and transition to supply chain science practice.
00:17:54 Difference between hiring data scientists and supply chain scientists.
00:19:39 Future plans for Lokad and challenges of rapid growth.
00:22:33 The story behind the name “Lokad”.
00:23:37 Biggest setbacks faced by Lokad.
00:25:45 Coronavirus highlighting the need to transform supply chain models.
00:26:59 Emphasizing resilience and capacity to embrace uncertainty.
00:28:11 How Lokad’s algorithms performed during COVID-19 disruption.
00:29:00 The importance of adaptability and managing changing demands during lockdown.
00:30:01 Adapting supply chain models during crisis situations.
00:30:59 Bitcoin, blockchain, and its impact on supply chain security.
00:33:30 The importance of professionalism and understanding for accurate forecasting.
00:36:23 Challenges in implementing narrow AI solutions for business forecasting.
00:38:01 Discussing bad data and the impact of poorly qualified data on ERP systems.
00:39:08 Debating the longevity of global supply chains and the influence of specialization.
00:41:22 The future of local supply chains and the impact of automation on production locations.
00:42:19 Challenges of implementing the quantitative supply chain approach and organizational change management.
00:45:15 Identifying the areas where the low-cut approach can unlock the most business value.
00:46:01 Discussing supply chain optimization for various industries.
00:47:34 Reasons for rewriting the Locad software from scratch.
00:49:10 Impact of key design decisions on software development.
00:50:23 Coexistence of Locad and S&OP-type solutions in organizations.
00:51:01 Discussing the challenges large companies face with fraud management.
00:51:57 Comment about who writes the blog and the impact of the podcast on content production.
00:53:00 Importance of reflecting on past mistakes and realizing when you’re wrong.
00:54:02 Learning from past mistakes to avoid similar issues in the future.
00:55:41 Conclusion and call to action.


Lokad founder Joannes Vermorel discusses the company’s journey and its focus on supply chain optimization in an interview with Kieran Chandler. Vermorel talks about Lokad’s early struggles with its forecasting-as-a-service model, its adoption of quantile forecasting, and its move to a programmatic approach. He emphasizes the need for supply chain companies to plan for uncertainty and embrace risk management in an increasingly unpredictable world. Vermorel also discusses Lokad’s algorithms’ performance during the COVID-19 disruptions, the potential value of cryptocurrencies in supply chain management, and the future of global supply chains. Finally.

Extended Summary

In this interview, Kieran Chandler and Joannes Vermorel, the founder of Lokad, discuss the company’s journey and its focus on supply chain optimization. Vermorel started Lokad in 2008 while pursuing a PhD in computational biology, but was drawn to the potential for innovation in the supply chain industry. The company initially struggled with its forecasting-as-a-service model, but later made significant strides, such as embracing forecasting with bias using quantile forecasts in 2011 and switching to a programmatic approach in 2012.

Vermorel explains that the company’s early approach to forecasting was based on eliminating bias, but eventually, they realized that biases could be useful in supply chain optimization. Quantile forecasting allowed them to be more profit-oriented, though it was initially considered a “weird” idea.

Lokad initially followed a traditional enterprise app model with screens, buttons, and menus. However, as they signed more clients, they realized that supply chains were too diverse to fit into a rigid app structure. The company pivoted to a programmatic approach, where calculations and features were customized for each client, focusing on productivity and reliability.

Reflecting on the company’s journey, Vermorel acknowledges that there were many lessons learned, and the path of an entrepreneur is filled with regrets. One significant change came with the rise of cloud computing, which forced the company to rewrite most of its products. Despite these challenges, Lokad has continued to evolve, embracing new generations of machine learning and focusing on solving better-defined problems in supply chain optimization.

The founder of Lokad, about the company’s history, technological breakthroughs, and future plans. Vermorel explains that cloud computing and deep learning were key breakthroughs for the company, as well as adopting a financial perspective on supply chain management.

Vermorel also shares his interest in Bitcoin, which he views as microeconomics in action, with parallels to supply chain management. He finds inspiration in the technical insights of the cryptocurrency community, which he believes can benefit Lokad.

Lokad moved from a data science approach to a supply chain science approach after finding that data scientists were too focused on data problems rather than supply chain issues. Vermorel emphasizes that the commitment of Lokad employees should be to improving the supply chain for clients rather than just producing fancy machine learning models.

When asked about the future of the company, Vermorel envisions more organic growth. He acknowledges that rapid growth may not be suitable for supply chain businesses, as it could lead to major issues if things go wrong. Lokad aims to grow at a sustainable pace while ensuring its employees have sufficient experience to handle complex supply chain problems.

Finally, Vermorel shares the origin of the name “Lokad.” Initially inspired by “local advertising,” he later embraced the interpretation “looking ahead” suggested by an IBM consultant.

Vermorel discusses the biggest setback the company faced in its early years, which happened around 2011-2012. During this time, Lokad won benchmark competitions, providing clients with improved forecast accuracy. However, these clients found that their supply chains worsened as a result, and their planners were frustrated with the software.

Vermorel reflects on a particular meeting in New York where angry clients confronted him, stating that while Lokad’s software provided better accuracy, it made their lives miserable and was not addressing the real problems in their supply chains. Ultimately, Lokad lost some clients over this issue.

The conversation then shifts to the topic of the coronavirus and its impact on traditional supply chain models. Vermorel believes that the pandemic is just one of many sources of uncertainty that can disrupt supply chains, mentioning examples such as political decisions, tariffs, or viral social media incidents. He emphasizes the need for companies to plan for uncertainty and embrace risk management, rather than relying on forecasts that pretend to predict the future with certainty.

Vermorel asserts that companies like Amazon, which focus on resilience and the capacity to embrace uncertainty, are the ones that are succeeding in the face of crisis. He suggests that the best financial institutions are also starting to align with these ideas and that supply chain companies should follow suit to stay ahead in an increasingly unpredictable world.

They talk about the performance of Lokad’s algorithms during the COVID-19 disruptions, the potential value of cryptocurrencies in supply chain management, the importance of professionalism and business understanding in supply chain optimization, the challenges in implementing narrow AI solutions for business forecasting, and the future of global supply chains.

Vermorel explains that during the COVID-19 crisis, Lokad’s algorithms did not perform well on their own. However, the company’s supply chain scientists were able to adapt and optimize the models in a short period of time, demonstrating the importance of human intervention in times of crisis.

In response to the question about narrow AI solutions for business forecasting, Vermorel expresses skepticism about the term “AI” and highlights the importance of understanding different classes of machine learning algorithms. He also discusses the quality of data in supply chains, stating that while data is not necessarily bad, it is often poorly qualified, leading to issues in interpretation and application.

Finally, the conversation turns to the future of global supply chains. Vermorel does not provide a definitive answer but raises the issue of climate change and its potential impact on the sustainability of global supply chains, suggesting that the current model may need to evolve.

They discussed the global nature of supply chains and how specialization limits the local production of certain goods. He also talks about the eventual return of some supply chains to local areas due to automation. Vermorel addresses skepticism around Lokad’s quantitative supply chain approach and the challenges of implementing it in organizations. He highlights that Lokad performs best in complicated supply chains with many options and decisions. Finally, Vermorel explains the motivation behind the complete rewrite of Lokad’s software and shares how Lokad works alongside Sales & Operations Planning (S&OP) type solutions, by mostly ignoring them as they are detached from the real-world effects on the supply chain.

They discussed about how Lokad operates side-by-side with data science teams that produce detached models that are not used. He also mentions that he writes the company’s blog, but at a much slower pace than before due to time constraints. Vermorel emphasizes the importance of revisiting past mistakes to understand what went wrong and how to avoid similar mistakes in the present and future. He believes that looking at a problem from a different angle can lead to breakthroughs rather than simply doing it better. Vermorel encourages viewers to send in questions to the podcast and to subscribe to future episodes.

Full Transcript

Kieran Chandler: Hello and welcome to a rather special edition of Lokad TV. Today, we’re live here in Paris to celebrate our 100th edition, where we’re going to be looking back at the Lokad journey so far and answering your supply chain questions.

Joannes Vermorel: I really didn’t think we would make it to a hundred episodes on something as bizarre as supply chain. The reason why we started all of that was just because I discovered this nice cool software called OBS, and I started to play with it. I found that it was a great piece of software, so I wanted to try it. But really, in reality, I think we are using it for the first day today since it’s only used for live events. No, I wasn’t really planning that much ahead.

Kieran Chandler: So today, the whole idea is we’re going to look back at the Lokad journey so far and sort of the lessons we’ve learned along the way. Perhaps if we start off by casting your mind back to 2008 when you first started the company, why did you decide to start a company within the supply chain industry? What was it that interested you?

Joannes Vermorel: At the time, I was a PhD student in computational biology, but I never completed my PhD. The number of excellent researchers in the field was astonishing, so it was humbling and very enthusiastic. However, I could see that the world would be just fine without me. When I started to look at supply chain, what I saw was mostly 19th-century math. I realized that there was potential to do things better in this area, which is absolutely gigantic. So, with a lot of enthusiasm, I launched my own company.

Kieran Chandler: How did those first couple of years go? Was it easy to launch? Were people interested in what you were talking about, or was there a lot of hesitancy at first?

Joannes Vermorel: No, it was terrible. It took us years to have something that actually worked. Lokad was founded on the idea of forecasting as a service, which is actually a very bad idea, both technically and from a supply chain perspective. The start of the journey was quite slow, precisely because it wasn’t working.

Kieran Chandler: Let’s talk about some of the big steps you took along the way. The first one you mentioned was in 2011, the idea of forecasting with bias using quantile forecasts. Why was this something that was a little bit controversial or different?

Joannes Vermorel: It wasn’t controversial; it was just plain weird. In statistics classes and in all the supply chain circles I was aware of, the idea of forecasting with bias was not well-known.

Kieran Chandler: The idea was you need to eliminate the bias, you know. Large companies have entire teams of demand planners who are spending their entire days eliminating and adjusting the model so that they are not biased. Why would you actually have people who do the opposite and add bias and do it on purpose, not by mistake? That was just the point, it was not controversial, it was like, dumb. Why do we have an entire team working on removing the bias and you want to add bias?

Joannes Vermorel: Actually, it took me several years to even come to the conclusion that this might be a good idea. For me, it wasn’t a controversial position; it was not a position at all. It was a non-problem until, by elimination of all the other stuff that didn’t work, I came to a conclusion. That was, I think, the quantile forecasting breakthrough. Yes, biases were very, very useful in supply chain because you want to prevent, you want to be biased toward profit, and thus we had to kind of completely re-engineer the technology around this idea.

Kieran Chandler: Okay, and then another step you took was in 2012, where you decided that instead of following the majority of the market, which was taking that kind of enterprise plug-and-play approach, you decided to do something very different and use more of a programmatic approach. Why was that something that you thought was good for supply chains?

Joannes Vermorel: Again, Lokad was started with the very classic way, you know, with screens, buttons, menus, and options – just the sort of things you expect from any kind of enterprise app. But the reality is, every single time we were signing a new client, we realized that there were so many things that didn’t fit. So, we were literally implementing tons of new features to accommodate each client.

Usually, when you start a software company, you think that yes, we don’t have all the features that the market wants, but we’re going to add a few more features and gradually converge to something that is feature-complete. So, it’s okay to start with a minimum viable product, and then you rinse and repeat, add a few features, and hopefully, you converge to something good that has a market fit. But I was literally four years down the road, and I didn’t see any convergence; if I saw anything, it was divergence.

We were starting to manage to win larger clients at the time, and I was seeing that it was even more diverse than what I had during the first few years when I was dealing only with SMBs. So, if anything, I was not on a convergent path; I was on a divergent path. When I looked at my competitors, I saw monsters – monsters in the sense of their software products being monsters, not the people. The software products had thousands of screens, literally thousands of options, and it was a completely divergent development process.

At the time, the challenge was, am I going to follow this path? It doesn’t even make sense. Is there any way to have some kind of convergence? And then, I finally came to the conclusion that supply chains were way too diverse to basically fit into a rigid app with menus and buttons. Instead, we needed a programmatic approach.

Kieran Chandler: Can you tell us about the beginnings of Lokad?

Joannes Vermorel: Yes, certainly. Lokad was founded with the idea of creating a platform for programmatic optimization and predictive optimization of the supply chain. We envisioned a platform where the menus, buttons, and calculations would be completely bespoke, and thus, you need to program them for every client. But if you program stuff for every client, then what is your problem? Your problem becomes producing productivity and reliability. You want to be able to do it super fast, super cheap, and thus envision a platform for programmatic optimization, predictive optimization of the supply chain was born.

Kieran Chandler: Did you ever look back at those early years and are there any big mistakes that you made and any big regrets that you have?

Joannes Vermorel: The path of an entrepreneur is full of regrets in the sense that if I knew at the time back in 2008 what I know presently, you know probably we would have taken three times less time. We would have been returned faster to the market than what we have been. But you know, it’s difficult to replace, even intellectually, to replay the past. For example, when I started in 2008, I started with the tech of the time, and then one year down the road, in 2009, it became very clear that, for example, the world of software has changed completely, and that we had to move toward cloud computing.

Kieran Chandler: Can you explain what cloud computing is?

Joannes Vermorel: Sure. The classical perspective of looking at a computer problem, that’s how I started back in 2008, is you have a machine, a computer, to do a calculation, a data crunching, you know, a task that you want to execute. So how long will it take? Well, it takes as long as the program is running. You have one machine. You launch the program, and when it’s done, it’s done. So what is constant is the machine. The problem varies, and thus, the computational time it takes to complete the resolution of the problem varies.

The cloud computing mindset is the complete opposite. What is constant is your target delivery time for the result of your calculation. So you say, “I want my calculation to be delivered in 30 minutes,” and then you can dynamically adjust the amount of computing resource you allocate to solve the problem. If I need a thousand CPU to basically deliver the result in 30 minutes, so let’s dynamically allocate those 1000 CPUs. The key insight was when we shifted from this idea of one side was the hardware is the constant and what varies is the problem and that’s the delay to deliver the solution to the problem versus the cloud computing perspective where the constant is a delay, and then you adjust the computing resources to deliver within the timeframe. Suddenly, we had to rewrite almost entirely everything that we had done at Lokad.

Kieran Chandler: If you kind of look back, you can see we’ve gradually evolved as technologies evolve with us. And if you look on our website, you can see those generations of machine learning we followed. What would you say was the biggest breakthrough from a technical perspective?

Joannes Vermorel: The thing is, it’s not just evolution. It was literally a complete change. People think, “Oh, it’s just evolution,” but no, it didn’t work like that through the history of Lokad. It was more like we had a product, we threw it away, and started from scratch, usually on a better problem. So, it’s not just a better product because it has the same features, just better. Usually, it’s literally a different problem because it tackles the issue with a better understanding, which usually completely changes the technology or the architecture of the software.

I think, in terms of machine learning, the biggest breakthrough was deep learning. From an infrastructure perspective, the biggest breakthrough was cloud computing. That’s the idea that you want to have hard deadlines to deliver your results, and the rest varies. But from a statistical perspective, the biggest breakthrough was probably deep learning, even if it’s not what we have right now in production. It’s differentiable programming, but the breakthrough itself was coming from deep learning.

And then, from a supply chain perspective, the biggest breakthrough was the idea that you need to take a completely financial perspective on the supply chain end-to-end. You put dollars of error and reward and opportunities everywhere. This financial mindset was probably the biggest breakthrough – to look at everything through the lenses of a financial analysis instead of looking at it from the lenses of service level, with percentage stages of error that you’re trying to improve.

Kieran Chandler: Another maybe slightly odd route that many would have looked at and seen as a bit bizarre was back in 2016 when we dipped our toes in the world of Bitcoin R&D. Why did we take that route, and what did you learn from those experiences?

Joannes Vermorel: Bitcoin has always been a hobby for me, so professionally, Lokad doesn’t really depend on any kind of crypto, blockchain, or Bitcoin. Nevertheless, it is fascinating because it’s economics in action. People have started to engineer software systems around ideas grounded in our understanding of economics, and that’s very interesting because usually, those ideas only belong to the realm of politics. They are never engineered.

You can have a politician that says we need to raise the taxation, and another politician that says we need to lower the taxation. The experience only takes place on the scale of countries, and usually, they are not engineered – they are just the result of an imperfect democratic process at best. The interesting thing about Bitcoin is that it’s a different approach to economics and technology.

Kieran Chandler: So, Joannes, tell us about your interest in Bitcoin and how it relates to supply chain optimization.

Joannes Vermorel: About Bitcoin, it was microeconomics in action from an engineering perspective. You can assess whether it works or not. That’s very interesting because supply chains are pretty much the same. It’s microeconomics in action. You can experiment and assess whether things are working or not. So, from this perspective, I found it very interesting. Bitcoin shares many properties of what supply chain has. It’s distributed, many actors, layers of software, tons of complexity, conflicting incentives, and multi-layered security problems. Obviously, it’s all analogies, not a direct translation, but there is a lot of inspiration to be found in looking at the technical insights found in those communities. Not the speculation, that’s just nice, but the technical insights. They are fairly interesting.

Kieran Chandler: Can you tell us about Lokad and the company’s focus?

Joannes Vermorel: Sure, we are kind of today just around 50 people located in the center of Paris. We provide what we call a supply chain science practice.

Kieran Chandler: Why did you move away from the classical data science kind of side of things?

Joannes Vermorel: I think it’s very kind of you to say that I decided to move away. It was more like we tried the classical data scientist route and failed badly. We had to move away from this. When we hired young engineers, right from the hiring interviews, we define the landscape of what are their loyalty, what are they loyal to, instead of what is their commitment. Are you committed to the vision, a type of problem, a type of skills? What is your commitment? When you go down the path of data science, people are committed to data problems. This is the wrong sort of commitment. You end up with people who focus on the cool problems and the cool tools, and they focus on the nice, coolish problems data-wise. Unfortunately, most of what it takes to solve a supply chain problem is pretty much in the uncool area, at least as far as data processing is concerned. You need to prepare where you need to qualify literally hundreds of fields, document them, and discuss with tons of people to clarify what the supply chain processes are exactly so that you have a chance to do something like an optimization that makes sense in practice. Thus, your commitment should not be to the data, it should be to the supply chain. That’s what we learned a bit the hard way. That’s why what we have right now is supply chain scientists because when we hire those young enthusiasts, we tell them your commitment is not in delivering a fancy machine learning model. This is not what Lokad is about.

Kieran Chandler: Your commitment will be to make the supply chain of our clients better and that’s a very different thing, and frankly we don’t really care that much. You know, if you do it one way or another, obviously we have recipes that we know to work, we have certain classes of tools that have been battle-tested. But fundamentally, you will do whatever it takes with a client to make their supply chain better. And that should be your commitment. That should be your daily challenge, your daily inspiration, and everything.

Joannes Vermorel: It turned out that when we were hiring for, I would say, data scientists, we were getting people that had probably too much interest in the fancy data problems and not enough interest in, I would say, the people, in the business problems, in making sure that there was nothing that would actually get in the way of the solution to be used in production because usually the problems of those, I would say, machine learning-driven initiatives is that they fail not because the algorithm has a problem, but just because there are bigger defects in the overall setup.

Kieran Chandler: Okay. And before we kind of dive into maybe some of our viewers’ questions, as a final kind of question, we were a company that’s very much grown organically over the last decade and a bit more. And so slowly but surely, what are your kind of ideas for the next five years, the next 10 years? What do you see for Lokad for the future?

Joannes Vermorel: Um, more organic growth. I mean, literally, one year, we had 60% growth, and frankly, we were that far from complete collapse. And what people don’t really realize is that when they see startups that say, “Oh, we have this 200% yearly growth,” it’s absolutely fantastic. I would say, “Yes, it’s good if you have something where you can move fast and break things.” You know, if you have a dating app, and you have complete server melts down, frankly, no big deal. Your customer base will come back, you know, tomorrow. It’s not a problem. When you have a complete meltdown of something that is driving a supply chain, and that you suddenly, your clients start to pass massive production orders or purchase orders that happens to be complete nonsense, we are talking of multi-million-dollar mistakes. So it’s very, very bad. The idea of, you know, “move fast and break things” is not completely compatible with supply chains. And what few people realize is that, I would say, looking at the job market as it exists nowadays in Paris, or same thing would be in New York, or other big cities in around the world, is that if you have like 50% growth yearly, and that you have a regular employee turnover at something like three, four years, you end up with a median age of your employee in your company that is six months. You see that? That means that if you have 50% growth, at the end of the year, you know, your company, half of the people have only been there for six months. And literally, if you expect that people with only, you know, six months worth of experience are able to drive, you know, supply chain, we are talking of potentially hundreds of millions of euros, you know, annually. That’s a lot to ask, even if you’re hiring smart, dedicated, brilliant engineers. It’s a lot.

Joannes Vermorel: And so, I believe they are like, unfortunately, when we’re not, you know, a business like Facebook growing at a thousand percent a year, it’s just not a reasonable option. And that’s why we are going fast, but there are limits in what can be done. Otherwise, we can’t even train the new people that we are constantly hiring.

Kieran Chandler: Okay, let’s dive into a few of the questions because I can see there’s already a fair few and a few familiar faces, friends of the show. First off is one that I think a lot of the staff here at Lokad were really interested in, and we never got told. It comes from Dervish who basically asks: Is there any special reason for the name Lokad? Is it an abbreviated form of the algorithms we’re using? Is it a secret?

Joannes Vermorel: The reality is, when I was doing my PhD in computational biology, I was thinking about starting a business where I would use digital display for advertising. So, I thought of local advertising, and I came up with “LoCad.” It was a very good domain name with five letters. I kept it, and then I think 10 years later, an IBM consultant told me, “Oh, Lokad, obviously it’s looking ahead. What a great name!” And I thought, “Yeah, looking ahead, that’s a cool story and that’s the one I will tell my clients now.” So, the real story was it was for local advertising, but I think this interpretation of looking ahead is way cooler.

Kieran Chandler: We’ve got another question here from Deh. It’s a bit miserable, but it’s all focused on mistakes. What was the biggest defeat or setback that we’ve had in the history of Lokad so far? And let’s particularly frame it maybe from a client’s perspective.

Joannes Vermorel: The biggest setback was, I think, with some large US clients that we had. During the first few years, I didn’t have big setbacks because I didn’t have big clients. So, it took time to actually achieve the big setbacks. The big setback was a turning point, I think it was around 2011-2012, where we were literally winning benchmarks, similar to the Kaggle competition with Walmart. We had more accuracy, classic weekly forecasts, monthly forecasts, and we were putting those things in prediction for our clients’ supply chains. However, their supply chains would get worse.

Then we would redo the benchmark, and compare, and Lokad would have better accuracy. But at some point, the client would have a phone call and tell me, “Joannes, you know what? You’ve completely disrupted our supply chain.” I remember a meeting I had in New York, where they had asked me to come, and I went into a room with 20 planners. Half of them were completely furious, telling me, “Your software is making our lives completely miserable.”

For me, it was such a nightmare. There were 20 people, they were really adamant, and I was thinking, “Yes, in terms of accuracy, we’re better.” But people would say, “Frankly, we don’t care. You’re not dealing with the mess; we are dealing with the mess, and it’s just not working. It’s a nightmare.”

Kieran Chandler: Overtime, the clients were committed to them, and they tried their best, we did our best. And I think we lost them something like three years down the road, but it was such a miserable experience.

Joannes Vermorel: Miserable, and whatever. Okay, we’ll try and cheer up a little bit and talk about another not so cheerful topic, unfortunately, the coronavirus. We can’t seem to avoid it at the minute.

Kieran Chandler: We have a message from SV asking, do you think coronavirus has highlighted a need to transform traditional supply chain models?

Joannes Vermorel: I believe that coronavirus is just one more source of variability. There are so many things that can shake the world. You can have the next president decide to establish tariffs, a country like England decide to exit the union, or your company can be completely disrupted by tons of things. For example, nowadays, you might have employees that post a racist video on YouTube that completely damages the brand overnight and then you lose 20% of your market share just because of this stupid video that becomes viral. There are tons of things that make the world more uncertain. So, I think if it outlines one thing, it’s something that we have been advocating for a lot of years: you should plan for uncertainty. I have no clue whatsoever what the future will be in the post-COVID world, but I’m pretty sure it will be even more erratic than before, and so you need to embrace uncertainty, embrace risk, and manage it instead of doing a forecast and pretending that you know the future with your crystal ball. Lokad doesn’t have a crystal ball, neither do you. So, you need to embrace uncertainty and manage risk. I believe that nowadays, this sort of idea is starting to make its way through finance. Not every hedge fund is aligned with this sort of idea, but the best are getting on board with that. And I suspect in terms of supply chain, the companies that are getting ahead with this crisis are the people like Amazon that precisely emphasize resilience, capacity to embrace uncertainty, and react super swiftly with a lot of digital systems to support that.

Kieran Chandler: I’m afraid we’re going to stick with the coronavirus theme just for one more question from Marcus Leopold, a friend of Lokad’s. He’s asking, how well did the Lokad algorithms actually do during the COVID disruption? Did the customers have to go back to the manual, or did the Lokad algorithms cope with it automatically?

Joannes Vermorel: The algorithms are not magic; they had the algorithms themselves fail terribly. But, and that’s a big but, is that Lokad, what we’re selling, is not just a software platform. We always say, nowadays, it’s what we call managed plans. It’s basically having the platform and a team of supply…

Kieran Chandler: chain scientists and literally the team of supply chain scientists were working overtime to first shut down in March pretty much all our European supply chains then one month later shut down all our US supply chains and then two-three months down the road to restart them, you know?

Joannes Vermorel: The challenges we faced involved organizing the shutdown, organizing a restart, and tweaking the model so that this unusual lockdown period would not be interpreted as a drop of demand. You cannot have three months where you’re going to count that as demand, as that’s going to completely distort all your seasonal profiles.

I think the crux of that is our technology has been quite efficient, not because the algorithms were powerful, but because it allowed supply chain scientists a very high level of productivity. When the crisis hit, we didn’t have weeks to prepare for the transition. We received phone calls saying, “Lokad, you know what? Next week we are shutting down our plants and warehouses. We need to prioritize tasks that need to be done before that time. You have 24 hours to adapt the model so that those things will be executed gracefully.” Every hour counted, and the supply chain scientists, even if the entire company worked overtime, had to execute that within a few days. The crux was not the quality of the algorithm but the productivity that Envision grants for explicit modeling of the supply chains.

Kieran Chandler: We’re going to go back to the discussions with cryptocurrencies and Bitcoin. This is a message from John Michelle who was asking, do you see any real value added from those Bitcoin blockchain applications in the supply chain in the near term, so over the next three or four years, or do you just see it as crypto hype?

Joannes Vermorel: I see a lot of value but just not the kind of value you would expect. First, cryptocurrencies redefine what computer security means. The interesting thing is that if you put a Bitcoin on a computer, you can know that the computer is safe because the Bitcoin doesn’t get stolen. That’s incredibly interesting, as it means that suddenly you have a very simple test to see if your systems are safe or not. You can drop some cryptocurrency on it, and if it evaporates, well, guess what? There is somebody lurking around in your systems. People started to realize this when Bitcoin businesses began having machines on the cloud holding sizable portions of cryptocurrencies online, and they all went bankrupt through theft.

Kieran Chandler: So, Joannes, can you talk a bit about the challenges of supply chain optimization in terms of IT security?

Joannes Vermorel: Yes, and literally people realized that nothing was safe, you know. All the cloud computing providers had exploits. All the IoT devices have exploits. All the smartphones have exploits. I mean, literally, people realize the extent of the problem, the magnitude of the prime. So, I believe that as a supply chain, we have the problem, I would say, twice as worse because supply chains are, by design, they are geographically distributed. You cannot have a fortress approach to IT security just because you have stuff pretty much spread all over the world. So, it’s a massive challenge, and what is happening in crypto is very interesting because it gives you insights on all the things that you have to do to secure your system for real. That’s the main added value. So again, I would not suggest that as an investment vehicle. I suggest having a look at the technicalities of those things and especially from an IT security perspective that typically combines social engineering problems with software exploit problems.

Kieran Chandler: Okay. We got another message here from one of our friends, Khalil Mehana, who sort of mentions forecasting. No matter how good it is, it needs information for kind of the users behind it. Two key users are going to be the project leader from the company side and the supply chain scientists from kind of the Lokad side. So, how important is that professionalism and business understanding from those two people, and how much can that impact the result and the accuracy of the final forecast?

Joannes Vermorel: That’s the trick. For one, we have about 100, a bit more than 100, companies in production. We are not tweaking the forecast. This is not how it’s done. You know, people think of, “Oh, you need to have like human insight to shift the forecast.” Now, it’s typically to frame how you are even going to extract statistical information from the data. So, you’re not trying to guess the market by having your human insight and you know stuff. I mean, that happens. That happens. We have some fringe situations, for example, one that comes to my mind, the A380, you know, the aircraft from Airbus. We have some clients that are serving parts of this aircraft. When basically Airbus announced that they are discontinuing this type of aircraft, yes, you can adjust your with market knowledge the forecast, but it’s very rare. It’s very rare, so sort of situation usually. The work of, I would say, the supply chain practitioners and the supply chain scientists at Lokad, there are a lot more. It’s typically about first framing the problem so that the algorithms are learning and optimizing the right thing, which is a moving target. If the situation has always covered, it’s not about injecting knowledge in the sense that you tweak the forecast. It’s literally you’re reframing the very problem that you’re trying to forecast and what you’re trying to optimize. Then it’s usually, I mean, most of the mundane work is about revisiting the economic drivers. You know, we are optimizing those dollars of error, but those dollars of error, it’s not something that you extract from the data. There is no data mining or machine learning algorithm to know whether um…uh.

Joannes Vermorel: Many people underestimate a few forces. First, they underestimate the strength of specialization for countries. For example, there are only three countries in the world that have RAM factories - random access memories. So, literally, if you want RAM, and all the computers use them, there are only three countries: South Korea, China, and the United States. Anywhere else, well, tough luck.

And then, if you want lithium for your batteries for your smartphone, it turns out that the world reserves of lithium are just in three countries: Chile, Argentina, and Australia. I think some people will double check. So, again, if you want to have a local production of lithium, well, tough luck. The reality is the same thing for master watchmaking. I believe that Switzerland has more than half, but I think on the master watchmaking segment, they’re like 70%. So, again, I don’t think it’s always feasible for low-value items like t-shirts.

Textile has been an industry that has been traditionally super lagging behind because it’s so hard to automate. As a result, textile production moved to China, but now it’s not in China anymore. It has moved to cheaper countries like Vietnam, the Philippines, or Bangladesh. But the thing is, once we are in Bangladesh and the salaries hopefully will rise there, where will those things move? Maybe in Africa? I don’t know. But we are running out of cheap countries.

And the automation is progressing, so I believe that for those basic supply chains, they will probably be brought locally. Unfortunately, don’t expect that it will produce jobs because they will be brought back to the places where those things are consumed once a super high degree of automation is achieved. So, will we have marginally more local supply chains? Yes, I kind of believe so. Again, because when you have things that are incredibly automated, it doesn’t really matter anymore where you position your factories. It does matter, again, lithium is only in a few places in the world, etcetera, etcetera. But suddenly, when you don’t care about the cost of local manpower, you can put your production pretty much where you want.

Kieran Chandler: We’ve got two questions here that I’m going to join together into one big one. One from Kenya and one from Manmeet. Kenya was asking because the quantitative supply chain approach is so different, do you encounter a lot of skepticism by implementing those kinds of practices? And Manmeet builds on that as well, asking what kind of organizational change management challenges do you encounter when you’re implementing something like Lokad?

Joannes Vermorel: That’s probably one of my biggest frustrations, I don’t encounter that much skepticism. And I’ll tell you why. It’s because when I tell people, for example, that time series forecasting, naked time series forecasting is completely broken, it ignores uncertainty. People then know that, and then I…

Kieran Chandler: So Joannes, can you explain what happens when you forecast a product but then introduce 10 more products that compete with it?

Joannes Vermorel: If you forecast a product, but you don’t know that you’re going to introduce 10 more products that compete with it, you’re going to have cannibalization all over the place. And if you have a time series model, it’s just going to completely ignore that cannibalization, and so it’s completely broken. And again, people are not idiots. They know that. So the frustration is they get it. I think we typically, when I’m talking with, let’s say, supply chain directors, head of planning, head of supply chain forecasting and this sort of thing, they’re not skeptical. They say, “Yes, yes, yes, I understand.” My frustration comes from, but you know what? I’m just not going to do it. I know it’s broken, but you know, am I willing to really, you know, in theory, a lot of people say if you ask them the question, “Yes, I’m going to do what’s best for my company.” But unfortunately, in large companies, people mostly do whatever it takes so that they can keep their job. And even if they are fairly high in the management ranks, you would pretend that most people are heroes that promote innovation and whatnot, but no, people, most people have very interesting hobbies, interesting life, and their job is well, just a job. And they are not going to go into a crusade to reform their organization so that it can perform better. Yes, for the shareholders, it’s better. It would grow the company, it would make it more profitable. But let’s face it, you know, most people that are in large organizations, they have a salary. There, they don’t want to harm the organization, you know. They want to be reasonably good at their job, but they’re not going to get into a crusade to kind of bring their company to the next stage. And if the company goes badly, they will go, they will change job and go to the next company, you know.

Kieran Chandler: Okay, we’ll just kind of squeeze in another couple of questions before we finish up. I’m gonna squeeze in this one from Richard Lebenski, mainly because he’s staying up late in his time zone. So yeah, yeah, heroes, heroes, yeah. And so he’s asking, well, in which area can the Lokad approach unlock the most business value where others maybe cannot?

Joannes Vermorel: It’s usually inversely proportional to the amount of mess and complication. And when people tell, “Oh, it’s like a nightmare,” it’s very good because usually when the supply chain is like a nightmare, with way too many options, way too many decisions, way too many things with second-order effect, with multi-echelon, with shelf life, with retrofits, also all sorts of super bizarre effects, that’s typically where Lokad shines the most because usually it’s a sort of supply chain where optimization hasn’t even started yet. So because obviously if you have a supply chain that is already extremely lean because it’s so simple, and probably, I would say the, let’s say, for example, water distribution, you know, there is nothing dumber than water distribution. So the supply chain of water, nobody talks about it because it’s so damn blend that there is nothing left to optimize. Everything that was left to optimize was almost optimized a century ago. So Lokad cannot do anything almost, I would say, for water companies. But

Kieran Chandler: On the most extreme other side, I would say, let’s say aerospace, which is a complete mess in complete disarray, especially with COVID. That’s typically sort of areas where we perform the best. But fresh food can be super complicated, luxury also tends to be super complicated, and there is the additional problem that you have very limited data sets. So the classical statistical methods for the parts typically don’t work in those situations. That’s a bit of a paradox. You would say Lokad performs best when there is a lot of data, but also when there is very limited data, and so all the usual statistics just don’t work. That’s also a very nice sweet spot for us.

Joannes Vermorel: Okay. It seems there’s been a bit of a discussion with Nicholas Vanderpooh, one of our previous favorite guests on the show. He’s been discussing with Edith, and the main question was, well, what was the biggest problem that motivated taking on a huge task like a complete rewind?

Kieran Chandler: You mean a rewrite of the software, Lokad?

Joannes Vermorel: Yes. I mean, first, if you see that something is not working, you know that you’re on the wrong path. Usually, the rewrite is just the last resort. It’s when you don’t have hope anymore. So, it’s literally at some point, you do incremental development and then more incremental development, and then at some point, you just run out of hope that it will ever work. And so, at some point, you just rewrite it from scratch. It’s very tough. It’s not something that we took lightly. Every time you know, there is a saying in software that you should never rewrite from scratch. I would say, yes, it’s usually a very bad thing to rewrite from scratch. But when you realize that you have architectural flaws, design flaws that are completely at the core of your architecture, you’re toast. You’re literally toast. And that’s something that I tell most of our clients. Most of the good or bad properties of a piece of software are by design. It’s literally the very key design decision that was made probably during the first three months of the development stage of this software that is driving everything else from that point on. So when you see that there was some key design assumption that just falls apart, you’re kind of screwed.

Kieran Chandler: I got a question here that might be quite entertaining from Slim Kalell. He asks, how well does Lokad work alongside SNOP-type solutions?

Joannes Vermorel: Oh, we just ignore them. The funny thing is that bureaucracies tend to have an incredible property; they systematically outlive their usefulness. So you end up with situations where Lokad is live in production. We are literally the supply chain scientists dealing with a modest team of supply chain practitioners, driving every single decision. So all the buying, all the production, all the inventory movements, even the prices. And then you have the SNOP bureaucracy, where people are still meeting, people are still doing their process, they still ask the salespeople for their forecast, spreadsheets are still filled with numbers, and people still practice their usual sandbagging shenanigans and whatnot. So this whole bureaucracy keeps living.

Kieran Chandler: It’s completely detached from, you know, the reality, so it doesn’t strictly hurt because it has zero effect on the real world, which is physical. I would say it has an effect on the supply chain, but people are under the impression that only governments can maintain useless administration, just like, for example, the UK that managed to have a Department of the Colonies way until the UK ceased to have any colony to manage. They were still like a gigantic ministry. France has done pretty much the same with, for example, we still have the Bank of France, which is to manage the Franc. We don’t have the Franc anymore; this is Euro, but we still have the Bank of France.

Joannes Vermorel: Bottom line, it’s not because you’re a company that you’re immune to the problem. It’s the same problem for any large kind of company, and so we end up with those very paradoxical situations where usually Lokad operates side by side with SNLP that is still doing their thing, just completely detached. And where it becomes even weirder is that when we operate side by side with the data science team because there is a data science team that is still producing models that are not used. They still produce prototypes, usually one or two per quarter, which are completely detached, and we just go on with our lives being in production. It’s weird, but you know what, it’s the way it is.

Kieran Chandler: Okay, and we’ll start wrapping things up now. I’ll sort of finish with maybe a comment rather than a question from Yatin Dinesh, who I know is a big fan of the show as well. He sort of just says, “Loving the podcast so far, years of experience and learnings through mistakes being shared.” But he wants to know, who is it that writes the blog?

Joannes Vermorel: The blog, it’s usually me. I mean, I’m very busy, yes. And by the way, if you look at the rhythm for blog production since we started this show, I have been producing at a much slower rate the blog posts. I will get back to it; I’m planning to get back to the blog, but there are only so many hours a day. But no, I never delegated to some kind of third-party content agency to just have like mass-produced happy talk, you know, that just gives you platitudes and nonsense combined.

Kieran Chandler: It’s always me who normally gets in the way of things. No, no, to kind of wrap things up today and to conclude, why did you think it was important to look back at the Lokad journey, and what is it you hope our viewers learn from today’s episode?

Joannes Vermorel: The thing is, it’s very difficult intellectually to realize that you’re wrong. It’s very hard. First, it’s not super agreeable; you don’t like to realize that you’ve made tons of mistakes. So, usually, your basic instinct is to have a defensive mechanism, so you find excuses. You say, “Yeah, we didn’t succeed, but the clients, you know, there was such a culture shock, it was such a tough situation, they had an ERP deployment in progress that complicated things.” There are always tons of excuses. There is this motto: you can have results or excuses. And there is a third way, which is to…

Kieran Chandler: Understand why it went wrong and where is very important to me. I’m probably doing tons of mistakes right now, I just don’t know which ones. To revisit your past mistakes is a way to think of what can possibly go wrong right now. I mean, obviously, we are doing things that are way better than what we were doing 10 years ago, but it doesn’t mean we are perfect. I’m pretty sure that there are tons of things when, 10 years from now, we’ll realize, frankly, I was just out of my mind, or something. It was just obvious that there was a better way to do it, and it was the elephant in the room. It was fat and obvious, and yet, we were just around that. So, in my mind, revisiting those things is something that I try to do repetitively because it gives you an angle to realize what is going wrong in what you’re doing right now.

Joannes Vermorel: And usually, again, the problem doesn’t lie in the fact that you could do something better. That’s the wrong way to look at the problem because usually, the problem is that you’re not even looking at the problem from the right way. It’s not about looking at the problem better, doing it better because that’s completely, I would say, a linear progression from what you have, like an incremental progression. Usually, the biggest breakthroughs happen when you see that you should have looked at this problem from a different angle. And then you realize there is another problem, another angle that is really worth fighting for. It’s not that we could do better; it’s that we were not even addressing the problem at all in the past.

Kieran Chandler: Okay, we’ll have to wrap it up there, but after a hundred episodes, we’ve probably earned a beer, I reckon. So that’s everything for this week, and if we didn’t answer your question, make sure you drop us an email at contact@lokad.com, and we’ll try to get back to you. Make sure you click on the subscribe button, and we’ll see you again in the next episode. Thanks for watching.