00:00:07 The role of Microsoft Excel in the supply chain industry.
00:01:33 Reasons for Excel’s popularity and its key strengths.
00:03:01 Heuristics and their implementation in spreadsheets.
00:04:43 Excel as a technological dead-end and its limitations.
00:06:46 Scalability problems and misconceptions about Excel’s limitations.
00:08:01 Limitations of Excel and spreadsheet programming model.
00:09:38 Scalability issues due to complex and replicated logic.
00:11:39 Moving away from spreadsheets and the need for better programming capabilities.
00:13:27 Going beyond heuristics and embracing modern approaches.
00:15:00 Learning from Google and Amazon, and the role of machine learning in supply chain optimization.
00:16:00 The importance of probabilistic forecasting in supply chain optimization.
00:17:10 Addressing the skepticism of using advanced technology over Microsoft Excel in supply chain management.
00:18:19 The impact of companies like Amazon and Alibaba on the industry.
00:19:45 The consequences of remaining in a technological dead-end.
00:20:01 The appropriate uses of Excel and its limitations in predictive supply chain optimization.

Summary

In this interview, Kieran Chandler and Joannes Vermorel, Lokad’s founder, discuss the role of Microsoft Excel in supply chain management. Vermorel calls Excel the “Swiss Army knife” of supply chains, acknowledging its strengths in programmability and customization. However, he believes spreadsheets have reached a technological dead-end in managing complex supply chains, with the programming model’s limitations resulting in replication and maintenance issues. To optimize supply chains, Vermorel suggests companies adopt advanced techniques like machine learning, replacing heuristics with data-driven insights. While Excel has merits, for complex optimization, organizations must move beyond spreadsheets and follow the footsteps of tech giants like Amazon and Alibaba.

Extended Summary

In this interview, Kieran Chandler, the host, discusses the role of Microsoft Excel in the supply chain industry with Joannes Vermorel, the founder of Lokad, a software company specializing in supply chain optimization. The conversation revolves around the reasons behind Excel’s popularity, its strengths, and the use of heuristics in the industry.

Vermorel believes that Microsoft Excel is the Swiss knife of supply chains, with an estimated 90% of supply chains worldwide being run through Excel. He attributes its popularity to the lack of superior alternatives until recently, as many supposedly better options were not actually superior in several ways.

Excel’s key strengths lie in its programmability and expressiveness, allowing for a high level of customization. Its distribution throughout organizations means that supply chain practitioners across various locations and product lines can craft their own heuristics or numerical recipes for managing their supply chains. Vermorel defines heuristics as numerical recipes that are not provably correct but are approximately correct. These heuristics have been tried, tested, and adjusted over time and vary from one division to another and from one location to another.

An example of a heuristic in supply chain management is keeping in stock exactly twice the number of units that were sold in the same period last year, considering a three-month window. Although the reasoning behind such heuristics is not always clear, they have been found to work and are widely used throughout the industry.

The host, Chandler, observes that basic approximations have been good enough for the supply chain industry for decades. Vermorel agrees but highlights the opportunity for improvement and optimization as the industry evolves.

They discussed the limitations of spreadsheet-based approaches in supply chain optimization, particularly focusing on Excel and its equivalents. Vermorel explains that companies have already reached the maximum potential of heuristics in spreadsheet-like environments, and this technology has hit a dead end. The discussion delves into the reasons for this and the inherent problems with using spreadsheets for complex supply chain management.

Vermorel notes that companies began exploring the potential of spreadsheet technology in the 1990s and reached a relatively stable point in the early 2000s. Despite some advancements, he believes that spreadsheets, including Excel and similar programs like Google Sheets and OpenOffice, have reached a technological dead end. This is because after companies have optimized their heuristics, the only remaining changes are inconsequential.

Chandler asks Vermorel to clarify the limitations of the spreadsheet-based approach. Vermorel explains that some people mistakenly think the problem with Excel is its inability to handle large amounts of data. However, he believes that the real issue lies in the programming model. He argues that if Microsoft wanted to increase Excel’s scalability to handle billions of lines of data, they could, but they choose not to because they recognize it as a dead end from a practical perspective.

The programming model in spreadsheets, according to Vermorel, is not scalable because it involves massive replication of logic. When users want to apply a piece of logic to more data, they copy and paste it across the spreadsheet, which results in an inefficient programming process. This replication becomes more problematic when organizations attempt to consolidate multiple heuristics across a large organization, leading to increased complexity and difficulty managing the information.

As an example, Vermorel describes a scenario where a small-scale spreadsheet contains a few hundred products and two or three heuristics. When the scope is expanded, and more heuristics are needed for larger segments, the complexity problem arises. Attempting to manage hundreds of heuristics across an entire organization using spreadsheets becomes an unmanageable nightmare.

The interview highlights the technological dead end reached by spreadsheet-based approaches to supply chain optimization. The limitations lie in the programming model, which involves massive replication of logic and an inability to manage complexity when scaling up to larger scopes and organizations. This makes spreadsheets unsuitable for addressing the intricate needs of supply chain management in today’s business landscape.

The conversation revolves around the challenges of using spreadsheets for complex supply chain management and the need to move beyond them for better efficiency and scalability.

Vermorel highlights that spreadsheets are often used to manage supply chains, but they are not an optimal solution due to their limitations in handling complexity. He points out that the programming model used in spreadsheets often leads to duplicated logic, making it difficult to maintain and debug. This becomes especially problematic when dealing with large spreadsheets containing hundreds of different formulas to accommodate different heuristics used by supply chain practitioners.

When asked how companies can move away from spreadsheets, Vermorel says that simply replicating the spreadsheet logic in another system would only result in marginal improvements. Instead, organizations need to fundamentally rethink their approach and adopt more advanced methods, such as machine learning, to replace heuristics with data-driven insights.

Discussing the lessons that can be learned from tech giants like Google and Amazon, Vermorel explains that these companies have moved beyond rule-based systems by using machine learning to learn from historical data. This enables them to optimize their supply chains more effectively. However, he notes that the key to success with machine learning is adopting a probabilistic forecasting perspective, which has been demonstrated by Amazon’s research and publications.

Addressing the concerns of skeptical supply chain practitioners who are hesitant to move away from Excel, Vermorel acknowledges that Excel has many positive qualities, such as stability and scalability. However, he warns that it represents a technological dead-end for predictive supply chain optimization. He urges practitioners to consider whether their industry can afford to remain at a plateau, especially when competitors like Amazon and Alibaba are aggressively pursuing technological advancements in supply chain management.

Vermorel concludes by emphasizing that Excel is not inherently flawed, and it can still be useful for data entry and other simpler tasks. However, for complex supply chain optimization, companies need to move beyond spreadsheets and adopt more advanced techniques.

Full Transcript

Kieran Chandler: Today, we’re going to discuss this aspiration and understand why replacing it is easier said than done. So Joannes, what do you see as the role of Microsoft Excel in the supply chain industry?

Joannes Vermorel: I mean, it’s literally the Swiss knife of supply chains. It’s the thing that is like BIC lighters and used everywhere for pretty much any sort of purposes. It’s actually very impressive how much gets done through Excel. In my estimation, I would say probably over 90 percent of the supply chains worldwide are run through Excel, not SAP or ERP systems. ERP systems manage assets, but as far as predictive supply chain optimization is concerned, I would say over 90% of it is done through Excel.

Kieran Chandler: If over 90 percent of the industry is using it, why is it so popular and why are people so reliant on it?

Joannes Vermorel: It’s interesting. I mean, the first part of the answer is because, until very recently, there were not that many superior alternatives. Most of the supposedly superior alternatives are actually not superior in several ways that we can describe. So first, people did not drop Excel not because they were stupid or religiously attached to it, but just because there were no credible alternatives.

Kieran Chandler: So, what are the characteristics of Excel that make it so strong, and why do people like it?

Joannes Vermorel: One of the things that makes it very powerful is that you can combine programmability and a level of expressiveness that comes with this sort of system. The second thing is that it’s heavily distributed in your organization. Many supply chain practitioners across countries, locations, and product lines can craft their own heuristics.

Kieran Chandler: By the way, what do you mean by heuristics? How do you define that?

Joannes Vermorel: Heuristics is kind of a numerical recipe that is not provably correct. It’s a best attempt at having something that is approximately correct. Usually, from a pure mathematical perspective, the heuristic is not even correct, but it kind of works. For example, a common supply chain heuristic is having in stock twice the number of units that were sold last year at the same period, considering a three-month window. These heuristics have been tried and tested, and the magic numbers like the duration for the time window or the factor used have been adjusted over time and vary from one division to another and from one location to another. What is great about spreadsheets is that you can have them embedded into your organization through a sea of spreadsheets that implement all those diverse heuristics.

Kieran Chandler: So, what we’re seeing in the industry is that this basic approximation is good enough, and a lot of supply chains have basically been running on that for decades.

Joannes Vermorel: Exactly. When you say “good enough,” it’s interesting because they have been running on it for decades, and from my perspective, it’s interesting because now…

Kieran Chandler: Nowadays, I would see Excel as a technology called dead end. So companies have already had ample time to come up with the heuristics, refine them, and get the most out of them. It’s interesting that they got to the point where they had already reached the most of those heuristics that you can have with Excel. And when I say Excel, I don’t mean just Excel, I mean any kind of software that gives you a spreadsheet-like environment. So, for example, Google Sheets would just be exactly the same as Excel in this respect. It doesn’t matter if it’s exactly Excel; what matters is the spreadsheet data model that is important here. The fact that it’s Excel or maybe the OpenOffice alternative doesn’t really matter. So, interestingly, those companies explored what you can do with a spreadsheet during the 90s, and I think they reached the point for many large companies of having something relatively stabilized in the early 2000s. We are now nearly two decades down the road of having something that has been already kind of stabilized, where there is nothing really new in this respect. And it’s indeed a technology called dead end because once you’ve compiled your stakes, once you’ve done that, the only things that are left are, I would say, things that are slightly inconsequential. So, you say it’s a technological dead-end, so what’s really missing then? What’s the problem with this kind of spreadsheet-based approach?

Joannes Vermorel: Some people misunderstand the limitations of Excel. A common misunderstanding is that you have a scalability problem with Excel, that you cannot process a lot of data. Yes, indeed, you cannot process terabytes of data with Excel spreadsheets, but that’s not actually a real problem. If Microsoft decided not to have spreadsheets that could deal with billions of lines, it’s not that they couldn’t do it. They did increase the limit from 65,000 max lines to 1 million something lines in Excel 97. They could bump up the limit to a billion lines with a different version of Excel geared to large-scale data processing. So the question is, why does Microsoft not just bump up the scalability of Excel? It’s because they also know that it’s a dead end from a practical perspective.

What is not scalable about Excel or spreadsheets in general is the programming model. The programming model is that whenever you have a piece of logic in a spreadsheet, if you want to do more of the same, you will basically copy and paste this piece of logic across your spreadsheet. From a programming perspective, what you’re doing is a massive replication of your logic. You have one formula, and now you have a million copies of your original formula. If you have a large organization, the good attributes of this organization through spreadsheets was that everybody could have their own heuristics. But if you take a spreadsheet with a few hundred products and you have two or three heuristics that work well, if you say now I’m going to consolidate in a larger spreadsheet the 20 something different heuristics that I need for this bigger scope, suddenly your spreadsheet ends up with a complexity problem. Your spreadsheet starts to contain not just two formulas that have been cut and pasted, but 20 formulas that are not used everywhere in the same way in the spreadsheet, and that starts to be fairly complicated. If you try to scale up to hundreds of heuristics through the whole organization, then it becomes a complete nightmare.

Kieran Chandler: Spreadsheets that are kind of clunky and these calculations seem to take a while. It’s that replicated logic which is the reason behind it, right?

Joannes Vermorel: Yes, to a large part. The programming model results in duplicated logic all over the place. The problem is all about the maintenance of this logic. How do you maintain an Excel spreadsheet that contains literally hundreds of different formulas? I’m not talking about hundreds of different formulas with just one distinct formula per column, because that’s the easy way. Imagine an Excel spreadsheet with a million lines, and some of those lines have a formula that is not just the same formula as the one that is above or below. Different supply chain practitioners dealing with different product lines and segments are using different heuristics. If you want to consider that, you end up with a spreadsheet that is super complicated and very difficult to maintain. Spreadsheets do not cope well with increased complexity, and it becomes a nightmare to maintain, debug, and even understand what is going on in these large spreadsheets.

Kieran Chandler: So, how can you move away from these spreadsheets? Organizations have spent years constructing them, and there’s a lot of logic held within them.

Joannes Vermorel: First, you need programming capabilities, but you don’t want to replicate what you had before. If you just try to replicate the spreadsheet logic that you had before, you will end up with something that is not going to be better than what you had before. It’s only going to be marginally better in terms of slightly better backups and access rights management. Fundamentally, if you just try to replicate your spreadsheet into another system, you’re going to be stuck in a technological dead-end. You may gain a few percent more efficiency, but that’s going to be very thin. Once you’re done with that, you will get nothing better. You may also lose some agility because the new system might be slightly more rigid. So, you need to think of something that goes beyond what you can get through heuristics. You need to reinvent yourself and go with something that gives you a chance to do better.

Kieran Chandler: You mentioned the Googles and the Amazons who have gone beyond this and are implementing more modern approaches. What can we learn from them and what they’ve implemented?

Joannes Vermorel: The quirk about modern machine learning is how to pass the stage of rule-based systems. The first stage of copying human intelligence in the 60s was rule-based engines, or decision engines. The heuristics used in supply chain are exactly that – they are rules to decide whether you should purchase more, produce more, or allocate more in one area. If you want to go beyond that, you need to reinvent yourself and adopt modern machine learning techniques to improve your supply chain processes.

Kieran Chandler: location or another and we have reached this stage of having rule-based systems and we have tuned the rules. If we look at what Google and Amazon are doing, they say, “Oh, we are doing machine learning,” so it becomes a buzzword and it’s very advanced machine learning that may qualify as AI. Fundamentally, it’s something very simple. Instead of having a static set of rules that are manually maintained, we want to learn those rules from the historical data.

Joannes Vermorel: What you need is basically programming capabilities, but you also need machine learning capabilities, so that most of those heuristics can be learned directly from the data itself. It’s not something exceedingly complicated, but if your programming paradigm is incorrect, then it just does not work, and the machine learning just doesn’t work. The dominant paradigm for quantitative supply chain until a few companies like Amazon or Lokad started to think differently was to have classic demand forecasts where there is only one future. We do the forecast, and then everything is based on this one future. Unfortunately, if you tackle the problem from this starting point, it just does not work, and you never manage to replicate the performance of those supposedly dumb heuristics. If you want to outperform those heuristics, you need to go for a probabilistic forecasting perspective, and then you can have a chance to outperform the heuristics. That’s exactly what Amazon seems to be doing, based on their published research.

Kieran Chandler: If we start drawing things together now, what would you say to a skeptical supply chain practitioner who’s probably watching this and they’ve got their systems which are kind of working, maybe a bit clunky, but they are working using Microsoft Excel? Is there really an incentive to move away from that?

Joannes Vermorel: If you’re using Excel and you probably have been for one or two decades, the first thing would be to acknowledge that you are in a technological dead-end. It might be good, but it’s not going to get any better. Do not expect that the next version of Microsoft is going to solve anything. Excel is already an excellent product. It’s very stable, it doesn’t crash, it’s fairly scalable, and it has many good properties. It’s not going to get much better. The spreadsheet alternatives are not going to make any difference. They might improve marginally, but fundamentally, it’s not going to make any difference. The question is: can you live with the fact that you’re just in a dead-end? Some industries can live being in a plateau, but as far as supply chain is concerned, what I see is that some companies like Amazon, Alibaba, and Zalando are very aggressive technically speaking, and they are going very fast. They are really achieving results, supply chain-wise, and they are not doing that with spreadsheets. So, I think the lesson is you’re on a plateau. Some people are doing way better, and it’s not marketing hype. The growth at Amazon and Alibaba is real. Can you really afford to be and remain in a technological dead-end? Maybe, maybe not.

Kieran Chandler: So, to conclude, would you say that Excel is running on borrowed time and you can actually see a day when there’s no Excel at all in the industry?

Joannes Vermorel: Don’t get me wrong. Excel is used frequently, not for predictive optimization, but for data entry, for tabular data entries. Microsoft won the spreadsheet war in the late 80s not because they had the best calculations for Excel, but because data entry through Excel was easier. There are plenty of situations where using a spreadsheet

Kieran Chandler: Because data entry through Excel was easier, I think, you know, there are plenty of situations where using a spreadsheet is just fine. What I’m saying is that if you want to do predictive supply chain optimization for somewhat complex supply chain networks, Excel is, for this specific purpose, at the end. For plenty of other things, Excel is just fine. Okay, any last thoughts, Joannes?

Joannes Vermorel: Just keep a close watch on this space, I guess.

Kieran Chandler: Same thing for this week. Thanks for tuning in, and we’ll see you again next time. Bye for now!