00:00:03 AI in supply chains.
00:01:02 Joannes Vermorel’s view on practical AI.
00:02:40 AI’s potential in more accurate forecasting.
00:04:07 AI in practice: solving car part compatibility.
00:06:33 AI’s role in handling supply chain edge cases.
00:08:01 AI’s potential for detecting system edge cases.
00:09:59 Applying AI to supply chain forecasting.
00:11:42 Need for human supervision in AI implementations.
00:14:27 Resistance to AI adoption due to job fears.
00:16:00 Inventory’s dual roles in retail.
00:17:33 AI in determining precise inventory needs.
00:18:39 AI’s impact on supply chain and marketing roles.
00:19:32 Challenges of AI implementation.
Summary
The discussion is centering on the application of AI in supply chain management. Vermorel, who is the founder of Lokad, is emphasizing the potential of AI to address complex challenges that go beyond mere statistical problems, such as car part compatibility, and manage edge cases that are traditionally handled manually. Even with the prevailing fear of job losses, he is arguing that AI often removes tedious tasks, thus enhancing job quality. However, he is acknowledging that the disruptive nature of AI can incite internal conflicts, as he cites an example where AI-driven inventory optimization is impacting marketing and supply chain responsibilities. Vermorel is suggesting that such organizational changes, rather than the deployment of software, could be what slows the adoption of AI in companies, signaling a significant shift in business norms over time.
Extended Summary
The discussion centers on the topic of Artificial Intelligence (AI) within supply chain management, highlighting potential benefits and implications. Kieran Chandler, the host, kicks off the conversation by asserting that AI is a recent buzzword in the tech industry. This prompts the guest, Joannes Vermorel, to ponder the timing of AI’s integration into supply chain operations.
Vermorel concurs with Chandler’s remark about AI’s buzzword status and proposes that genuine experts in the field seldom employ the term “Artificial Intelligence”. He stresses that despite AI being a buzzword, it doesn’t dismiss the significant developments happening under the AI umbrella. He names three key components: enhanced mathematical methods, an increasing amount of accessible data, and escalating processing power.
Vermorel posits that these advancements can lead to more accurate predictions within supply chain management. Yet, he insists that the timeline for total integration is indistinct and will probably span decades due to unique challenges in the supply chain industry.
As Chandler requests clarification on the potential benefits that sophisticated statistical methods or deep learning techniques can bring to supply chain management beyond improved forecasts, Vermorel argues that AI’s impact is multi-faceted. He clarifies that the AI revolution is like experiencing color after only knowing black and white; it’s not solely about higher resolution but also about unveiling new perspectives and dimensions.
He further accentuates that the most significant benefits of AI for supply chain management might emerge from areas that don’t initially appear as statistical problems. These hidden opportunities for AI application, Vermorel suggests, is where AI’s true value will shine.
To exemplify his point, Vermorel offers an instance of a non-obvious application of AI in the supply chain: car part compatibility. He outlines the difficulty of maintaining a database of car-part compatibilities, a formidable task considering the millions of unique parts and hundreds of thousands of unique vehicles in Europe alone.
Vermorel reveals how his team at Lokad utilized machine learning (a subset of AI) to address this issue. Their algorithm demonstrated 98% accuracy in identifying incorrect compatibility claims in the database, as well as missing compatibilities. This case underscores the potential of AI to solve complex problems within the supply chain that extend beyond typical statistical issues.
Vermorel initiates a conversation about how supply chain complexities go beyond standard catalogs or ready-made solutions. He emphasizes that most supply chain challenges are found in the edge cases—situations that diverge from the norm. These edge cases, he claims, are often addressed by large teams of people manually modifying and correcting anomalies through extensive use of tools like Excel. This laborious process, although necessary, indicates an area where AI could provide substantial benefits.
The discussion then shifts to the possibilities AI presents in detecting and managing these edge cases. Vermorel illustrates that AI could potentially mitigate some issues faced in supply chain management, including delays and accidental stock-outs. However, this AI solution might not resemble familiar voice-operated systems like Siri or Cortana. Rather than a single, multifunctional AI, Vermorel envisions a series of highly specialized micro-use-cases of AI designed to handle specific facets of the supply chain.
Vermorel also comments on the predictive abilities of AI in supply chain management. He notes that, apart from demand forecasting, AI can also offer a probabilistic forecast about supplier issues, such as delays or quality problems. He mentions that AI can forecast customer returns, a particularly crucial factor in the context of fashion e-commerce. These predictive capabilities of AI could play a vital role in optimizing supply chain operations, mitigating numerous uncertainties inherent in the process.
Later, Chandler and Vermorel debate the level of AI expertise required to implement such a system. The question is whether companies need AI experts to leverage AI’s benefits
in supply chain management. Vermorel considers that the AI elements of the operation can be outsourced, thus removing the need for companies to maintain an in-house team of AI specialists. He suggests that organizations can delegate their AI needs to a company like Lokad, which specializes in supply chain optimization.
Vermorel pinpoints one of the major hurdles in the adoption of AI in companies as potential internal conflicts due to the disruption of the status quo. The fear of job losses with AI implementation also comes up in the discussion. However, Vermorel considers this fear, while valid, often misguided. He states that AI tends to replace tedious, menial tasks like spreadsheet management, which he dubs “the worst kind of job ever”. Instead of generating resentment, this kind of automation can liberate employees to concentrate on more meaningful aspects of their roles.
Still, Vermorel acknowledges that AI adoption can lead to discord within a company, but this conflict would manifest at a corporate level rather than among rank-and-file employees. He illustrates this with a retail example. Inventory in a retail store serves two functions: meeting customer demands and making the store appealing to customers. Here, AI could determine the optimal inventory needed to fulfill both purposes.
The issue arises when determining which department within the company should bear the costs for each function of the inventory. Supply chain would naturally cover the inventory cost to meet customer demands. However, the cost of inventory aimed at making the store attractive (which Vermorel equates to marketing expenses like TV ads) would logically fall under marketing. This allocation, driven by AI’s precision, could lead to significant disputes, especially if, for instance, marketing directors suddenly find themselves burdened with large unexpected expenses.
Vermorel suggests that it’s precisely these kinds of issues that will decelerate AI adoption in companies, rather than the fear of job losses. Although deploying the software itself could be done relatively quickly, the organizational changes it prompts could take much longer. Vermorel believes this reassessment of operational norms and responsibilities, prompted by AI implementation, will constitute the main challenge for businesses in the coming years.
Full Transcript
Kieran Chandler: Today, we’re going to be talking about artificial intelligence, a subject that’s been a buzzword in the world of technology over the last couple of years. Today, we’re going to try and rise above some of that buzz and instead focus on its application and what it can do for the world of supply chains. Joining me once again today is Joannes Vermorel, and he’s going to help me with today’s discussion. So Joannes, thanks for joining us again.
Joannes Vermorel: Hello, Kieran.
Kieran Chandler: So, if we listen to some of the experts in the field, they claim that artificial intelligence is going to replace half of the jobs in the world by 2050. However, if you look at supply chains in general, they’re not quite there yet. They’re still very much managed by Excel spreadsheets and in a very human-driven way. So, when do you think artificial intelligence will come to the supply chain industry? What sort of time frame are we talking about here?
Joannes Vermorel: That’s a very interesting question. I agree with your statement that artificial intelligence is a buzzword. And I even believe that one way to recognize if someone knows what they’re talking about when speaking of artificial intelligence is whether they use the term “artificial intelligence” or not. The most competent people often don’t. But just because it’s a buzzword doesn’t mean there’s nothing underneath it. So what do we have underneath? We have about three things: better mathematical methods, a lot more data, and processing power. You need a method that can convert all these data with better mathematical methods into better results, which in supply chain translates into something like more accurate forecasts. In terms of timing, I believe that there are a lot of specific angles to the supply chain world that need to be addressed, and the timing is quite fuzzy. It will take a lot of time, literally decades.
Kieran Chandler: So if we put aside the concerns with the terminology and say that artificial intelligence basically refers to some really advanced deep learning techniques, what can supply chain practitioners expect from these advanced statistical methods? They could probably expect better forecasts, but is there anything more that they can expect?
Joannes Vermorel: Yes, there is quite a lot more. One of the problems with forecasting before this wave of artificial intelligence is that we didn’t have many real-life examples of what a better forecast might look like. By definition, a more accurate forecast is a better forecast, but is it the only way? Artificial intelligence provides other examples that suggest it’s not. It’s a bit like we were looking at something in black and white, and now we get the color. It’s not just better resolution, it’s another dimension. For supply chains, I believe the biggest benefits will come from problems that don’t look like statistical problems at all, and that’s where artificial intelligence will really shine.
Kieran Chandler: So, what you’re saying is that there are problems that, on the surface, don’t look like statistical problems at all, but can actually be used as a forecast of some sort in order to make it possible to use AI technology. Could you expand on that a little bit more?
Joannes Vermorel: Last year, I had the opportunity to work on a very intriguing problem, which is car part compatibility in the auto industry.
Kieran Chandler: You basically have cars that need to be repaired and for those cars, you need parts. Just to give you a sense of the problem on the European market, there are several million different parts. It’s a bit insane when you think that there are only 300 million Europeans. Moreover, there are over one hundred thousand distinct vehicles. There’s an entire industry, albeit small, that competes on one thing: establishing a database of compatibility between cars and parts. All these companies are doing is building a list of which car is compatible with which part.
Joannes Vermorel: Indeed, these databases consist of millions of lines and they are entirely maintained by hand with literally hundreds of people dedicating their lives to maintain this one database. My team at Lokad, which specializes in machine learning, not artificial intelligence per se, managed to develop an algorithm. We tested this algorithm on a real-life setup and it achieved 98% accuracy in detecting the compatibility that is claimed by the database. The algorithm also demonstrated 98% accuracy in detecting missing compatibilities, so there could be a part that can actually be mounted on your car, but you or anyone else doesn’t know it yet because it’s quite hard to keep track of so many cars and parts.
Kieran Chandler: Using artificial intelligence to work out if a car part is compatible with my vehicle seems a little bit overkill. I would have thought a simple catalog might work for that or an off-the-shelf basic solution. But what about supply chains in general? What can artificial intelligence do for them?
Joannes Vermorel: The point I want to illustrate is that most of the challenges in supply chains actually lie in the edge cases. These are situations that usually work but then you have exceptions. These exceptions don’t solve themselves. It takes people, and a lot of them, to solve these edge cases. You end up with entire armies of people essentially tweaking Excel sheets because the way you notice there are so many people dealing with these edge cases in supply chains is the vast number of people editing Excel sheets. They aren’t wasting their time. They are dealing with these edge cases that do not really fit into the main ERP system. They have to revert to Excel to manage these. So technically, whenever you see people having to manually deal with a multitude of edge cases that typically involve Microsoft Excel, this is a situation that most likely artificial intelligence can solve.
Kieran Chandler: So having AI detect these edge cases sounds like a great idea. It would certainly go a long way to solve some of the problems we see here at Lokad, such as resulting in things like delays and accidental stock-outs. But what would this actually look like in practice? Would it be something like Siri or Cortana, like a voice in the operator’s ear that sort of tells them what to do and when to do it?
Joannes Vermorel: The idea of having your phone suddenly say, “look left, you have a problem” is pure science fiction. These current AI systems, like Cortana and Siri, are more a series of heavily specialized micro use cases. For example, the people who implemented Cortana and Siri have a special use case for ordering a pizza. They build a lot of code just to make it sufficiently flexible to work pretty much everywhere in the world, so you can order a pizza.
Kieran Chandler: Achieving a successful pizza delivery anywhere in the world is actually quite a challenge. It’s all about very specific use cases. Those AI assistants are nothing but a collection of use cases that are well integrated. Now, for supply chain, it will be pretty much the same. You’re going to get forecasts for all areas where there is uncertainty. Future demand is not the only thing that is uncertain in your supply chain, there are many other things such as, for example, lead times. Your suppliers are not perfectly reliable. It is not clear how reliable or unreliable they are.
Joannes Vermorel: Absolutely, and this is where AI can help. Artificial intelligence can provide a very accurate probabilistic forecast of the issues that your suppliers might generate. And it’s not just about delays. Maybe your supplier will deliver the goods you’ve ordered on time, but when they arrive at your warehouse and you inspect them, you might find that there’s a quality problem.
So, it’s not just about delay, it’s also about the quality of what you received. If you’re a fashion e-commerce platform, for example, you sell products to your customers and sometimes, because it’s fashion, they just don’t like it. So what they do is they return those products to you. Knowing in advance who and how many people will return items to you is highly useful for optimizing your supply chain.
There are tons of areas where you face uncertainty, maybe not as significant as predicting future demand, but still crucial to address. I believe these future technologies based on AI will significantly contribute to supply chain management.
Kieran Chandler: Moving on to the human supervision side of things. These AI technologies won’t be able to work on their own. How much AI expertise is actually going to be needed to make a project like this work? Large companies, like Google, might have the resources to employ a large number of AI experts, but what about the rest of us? For instance, fashion e-commerce companies tend to be quite ahead of the trends when it comes to technology, but they might not have AI experts in their ranks. How do you see this actually working in the real world?
Joannes Vermorel: Firstly, the very best fashion e-commerce companies that I know do have AI experts in their ranks, although that’s an outlier. To answer your question, I believe that supply chain companies, or companies who have a supply chain to manage, don’t necessarily need AI experts. They need something else, and I’ll get back to that later.
The AI component can be completely outsourced to a company like Lokad. If you have any concerns about AI, you can simply become a client of Lokad and outsource your AI component to us. That’s a strategy that can scale quite well.
However, let’s address the elements that don’t scale, and this is where the timing aspect comes in. I believe that the problem with AI is that it forces companies to become more rational. It also compels them to remove ambiguities and challenge the status quo. That’s exactly what I discuss in my book on quantitative supply chain. If you hope to optimize anything, you need to first establish a measurement. This is challenging because it directly questions the status quo. That, I believe, is the real challenge with AI technology in the supply chain.
Kieran Chandler: Companies face the challenge of changing the status quo to improve and leverage AI. This can potentially lead to internal disputes. It’s interesting to explore how this plays out. It seems like there are companies out there who could see AI as a realistic prospect. Yet, there’s also a sympathy for those employees who might have their jobs replaced by these intelligent systems. How long will it be before we have an AI presenter here on TV? All joking aside, is there a concern that this could block or slow down the adoption of technology? Here at Lokad, we have plenty of clients. If we introduce a new artificially intelligent model, how do you see it working with your customers?
Joannes Vermorel: That’s a good point. The fear is valid, but people’s expectations of what will happen are usually incorrect. If you read the press, you’d think that all these jobs will be replaced and people are going to object. However, in supply chain, that’s not the case. Why? Because the jobs that are being replaced are, quite frankly, not great. Imagine dedicating eight hours a day to editing Excel sheets. It’s not an engaging job. People are usually quite happy when that task can be automated. They can then do a more rewarding version of their job that makes more sense and doesn’t involve so much tedious work with Excel. But this doesn’t mean there will be no disputes. These disputes, however, will occur at a completely different level - at the corporate level.
For example, consider a retail network with various stores. The question is - what about the inventory in each store? On the surface, you would think that all products in the stores exist solely to be sold to customers. But it’s not entirely true. The inventory within a store serves two purposes. The first one is that a customer walks into the store, finds what they want, and makes a purchase. The second purpose is to make the store appealing enough that the customer is enticed to buy something. So, the inventory plays a dual role.
Kieran Chandler: The issue we’re discussing today is the reason why stores are so full of merchandise. It seems as if they’re afraid to appear half-empty, reminiscent of a store from the USSR. It’s not a desirable image for customers.
Joannes Vermorel: Absolutely, people in retail are aware of this. This is essentially what merchandising is about. Now, let’s consider what artificial intelligence can bring to the table. AI is so exact that it can provide answers you didn’t even know you needed. First, it can tell you precisely how many units of stock you need to serve your clients. Secondly, it can calculate how much stock you need to make your store visually appealing to customers.
Kieran Chandler: So, if we think about this at the corporate level, who is going to pay for these two types of stocks?
Joannes Vermorel: Well, the supply chain naturally pays for the stock that is needed for the customers. But as for the additional stock that’s there simply to make the store look appealing, that’s essentially a marketing expense. It’s akin to paying for a TV ad - it doesn’t directly sell products but generates interest. So, when you integrate AI into your retail chain, it begins to blur the line between supply chain and marketing. Some people may find this challenging, especially the marketing director, who might suddenly find a huge portion of the budget allocated to stock.
Kieran Chandler: It seems like this shift could cause some internal conflicts within the company.
Joannes Vermorel: Yes, there could be resistance, especially from those who were comfortable with their previous budgeting arrangements. They might say, “No, supply chain folks, please keep that. I was just fine having only TV ads as part of my budget.” But now, the perception is changing. Stock is being viewed as a part of marketing. This is a profound shift, and while implementing AI software can be done swiftly, understanding and adjusting to these changes might take decades.
Kieran Chandler: I see, that’s an insightful perspective. Well, I’m afraid that’s all we’ve got time for today. Thanks for such an interesting discussion, Joannes. It’s been wide-ranging - we’ve started with artificial intelligence and ended up talking about stores in the USSR. That’s the way these things go. Well, thanks for your time, Joannes.
Joannes Vermorel: Thank you, Kieran.
Kieran Chandler: I hope that our discussion helped clear up some of the common misconceptions about artificial intelligence. Thanks for tuning in and for the great feedback we’ve received on our videos so far. We’ll be back very soon, but until then, goodbye.
