Joel Beal:
David, welcome to Shelf Life. I'm so happy to be here with you today.
David Simchi-Levi:
Great to be here. Joel, Happy to talk supply chain with you in the next 45 minutes.
Joel Beal:
Always good to talk about supply chain. You obviously have a very impressive background. I'm not sure I told you this in advance. I am a failed academic, if you will. I started the path of an academic career in economics, ended up dropping out, moving full time into technology. You've had a very impressive career. I know MIT is one of the leading institutions around supply chain. I'm curious to start out how did you end up in this space?
David Simchi-Levi:
Actually, this is a great question, Joel. I started 30 years ago at Columbia University as an assistant professor, focusing on doing a lot of theoretical work around the analysis of analytics for vehicle routing problems. Most of the work was purely theoretical but at some point in the first couple of years I received a call from the New York City Board of Education. This is think about the timing. This is about 1993. The person on the other side of the call is telling me we are developing a new decision support system for school bus routing. We would like to work with you on the analytics, on the engine that will generate the routes that the school buses will follow. I was not that interested. I was doing theory. I was focusing on my academic career. I expressed this concern to the person on the other side. At some point there was a pause and the person was asking me are you interested in making a big impact? Do you want to be relevant? I'm thinking here is my opportunity to transfer the theories that I developed into practical algorithm. I started collaborating with the Board of Education in New York in the development of a new system. Remember the time existed in 1993. As a technology that we see today was not available at the time. We needed to develop the analytics. Then we needed to connect the analytics, the route generated by the analytics, with digital maps show, present the routes generated to the planners, make sure that the routes follow all the requirements, the safety requirements, introduced by the Board of Education. All of a sudden we developed a system that was used in the entire New York City and dramatic improvement in performance of the system. This technology won the best window-developed technology award that was given by Microsoft in 1994, 1995. All of a sudden I realized, hey, I can develop a system that uses analytics as the main engine but requires way beyond the analytics. But I was looking where are the opportunities to make an impact? How many cities you have that are the size of New York? Very few, and so I did not know anything about supply chain. And all of a sudden I realized there is this new emerging area called supply chain where maybe this analytics can make an impact. And I basically founded the first company called logic tools that focused on supply chain analytics. And this company that was founded in 1996, 1997 had three types of technologies one for supply chain network design, one for multi-ethic loan inventory optimization and the third one for production planning and scheduling. Ten years later, think about the timing. We had 350 clients using our technology on a day-to-day basis. We saw the company and became part of IBM technology infrastructure in 2008, and then I was. I joined IBM, helping them with the new technology, and then in 2011, I founded a business analytics company. Think about the timing. This is a time where nobody talked about application of AI and machine learning for businesses. We founded one of the first companies that focus on business analytics for operation and and and supply chain, and very quickly I followed this up with a cloud technology company that provided opportunities for developers and for companies to implement to customize their solution to their own environment. These two companies the business analytics and the cloud technology companies became a part of Accenture technology infrastructure. I'm company free. I don't push any company, but right now I lead the MIT data science lab, and let me tell you just a little bit about the MIT data science lab. That that I lead the MIT data science lab is a partnership between MIT and about 25 companies, focusing on some of the most challenging problems that these companies have by bringing to together data models and analysis. We have companies, partners from different industries, high-tech, cpg, finance, even government, are part of the partnership, as well as software companies, and we have a global footprint and we cover varieties of areas. One of the areas that we cover and I will end up with this story is the area of supply chain resiliency. Everybody, as you know very well, joel today is excited and focused on supply chain resiliency. We at the lab started focusing on supply chain resiliency way before the pandemic in 2011, 2012, after events like the tsunami in Japan and the flood in Thailand, and, if you remember, the volcano eruption in Iceland in 2010, and we were fortunate enough to collaborate with the Ford Motor Company, develop a new way to measure the resiliency of any supply chain, identify hidden risk and develop mitigation strategies. Our technology, the MIT data science lab technology, was implemented by Ford a cosy inter supply chain. We received the best engineering technology implemented at Ford in 2015 and we published a few technical papers, but the ones that I will refer to here is the Harvard Business Review article that was published in 2014. Between the publication of this article in 2014 focused on executive and supply chain leaders. In the beginning of the pandemic, Joel, I was knocking on many doors trying to convince executives and leaders to focus on supply chain resiliency very little interest. Most companies, as you know very well, focus on supply chain efficiencies, on cutting costs in the supply chain. Everything changed at the beginning of the pandemic, but perhaps not in the way you expect. Think January 2020. The pandemic was in China, it was not in North America, it was not in Europe. I started collecting data on what's happening in China and I realized I can use the technology I developed six years earlier to understand the impact of what's happening in China on supply chain in North America and in Europe, and I very quickly published a paper in February of 2020 in Harvard Business Review saying supply chain in the title and in the first paragraph supply chain in North America and Europe will stop by mid-March. And this is precisely this is exactly what happened. March 17, fortune magazine reported the entire automotive industry in Europe is shutting down. The New York time, march 18 2020, reported the entire automotive industry in the US, in Mexico, in Canada, is shutting down, and after that we got an enormous interest in our technology. Different consulting companies started implementing what we develop. Others have used the concept that we developed, emphasizing that they focus on supply chain resiliency, and I will end up just with a short comment which says last year, in May of 2022, the US president economic report had a section on our technology, recommending to companies and encouraging companies to use supply chain stress test that what I called we call our technology to make sure they are ready for the next disruption. We all you myself, your colleague, we are all supply chain professional. Think about this.
Joel Beal:
The highest decision-maker in the country recognize that what we all are focusing on is important for the national security of the country well, they do say timing is everything and you've obviously been studying and focusing on supply chain for, you know, decades at this, this point, and you're very interesting to walk through that timeline. You, as you say, you know publishing those, seeing this, this wave of disruption coming and then having it happen. Before we jump into that article, because I definitely want to talk about that, I'd encourage anyone listening to to read it, a fantastic article that you put in Harvard Business Review. How fast did you see that change happen when COVID hit? I mean, was it just all of a sudden, you know, your inbox was flooded and phone calls were coming in, or did it take people a while to you know realize the magnitude, even when COVID hit, of what was going to happen?
David Simchi-Levi:
It was relatively quiet until the publication of our paper in middle of February. And then the first call that I got was before mid-March, a call from Ford. The people I collaborated with six years earlier said hey, we read your article. Nobody understood why you are predicting mid-March. Can you come and give a talk? And I think March 2nd, on March 3rd I don't remember the exact day I gave a talk at Ford with hundreds of people participating from the entire organization trying to understand. Hey, these guys that worked with us Seekers earlier, he's saying he used the same concept, the same technology to predict. Two weeks later I get another set of calls from, called from Ford people hey, the prediction was highly accurate.And after that was a flood of calls and interest in what we did
Joel Beal:
So let's talk a little bit about, or maybe a lot about, supply chain digitization, because that's really kind of the overarching theme of your article. There's many different elements here that we'll dive into, and I know that you've been brought into a lot of companies you know to talk about this and what Folks can you just start? I'm sure people here are familiar with the term. A lot of consulting firms are out there pushing supply chain digitization. Can you explain what it is and why it's important? It's always been important, but it's probably particularly important today.
David Simchi-Levi:
Again, great observation and a very good question to start the discussion around supply chain digitization, george, and maybe as a background, we need to talk about perception versus reality when we focus on supply chain digitization. If you think about the perception before the pandemic, the perception that executive had about supply chain digitization was that it requires significant financial investment. It will take many years people were talking about five to six years to digitize the entire supply chain because it requires instrumentation of every process and every product and every facilities. What we learned during the pandemic is that the future is here, that with moderate financial investment, by taking advantage of existing data that companies have and complimenting it with additional data that companies can get from third party partners, we can achieve almost all the benefits of full-flight supply chain digitization in a relatively short period of time. Remember, the perception was this is five, six years. Now we are talking about 12 to 18 months. The perception is that we need enormous financial investment. Now we know we don't. Why, because the key about supply chain digitization is to focus on a few key capabilities. What are the key capabilities? The first is what I highlight in the Harvard Business Review article, which is all about unified view of demand. Taking traditional consensus forecast that most companies are using, what is the consensus forecast? In consensus forecast, different functional areas will generate their own forecast. Finance has a forecast, sales has a forecast, supply chain has a forecast. A trade, the people who are responsible for marketing, pricing and discounting has their own forecast. And then they come together in a consensus meeting to agree on a compromise. In supply chain digitization, we replace this process which is mostly manual process and also the compromise does not necessarily match with market demand with the process where we agree on the data and then let the analytic generate a forecast that is used by its different functional areas, a different level of granularity. So that's one important capability. The second important capability in supply chain digitization is to replace one size fits all supply chain strategy with a segmented supply chain strategy. What is a segmented supply chain strategy? We recognize that different product has different characteristics, different channels have different objectives and different limitations and, as a result, what we want to do is to segment, to cluster product channels, market into segments that in some sense are similar and for each segment we generate its own supply chain strategy. And finally, on top of this, we build synergies across the different segments in order to leverage economies of scale and to reduce complexity. So that's the second capability. First is unified view of demand replacing consensus forecast. Second, replacing one size fits all with a segmented supply chain strategy. Once we have these two capabilities, we can take advantage of them in smart SNOP. Snop is a process that companies have used you know very well, everybody knows very well have been using since the late 1980s to continuously balance supply and demand. Now we take advantage of the unified view of demand and SNOP is really one integrated process with segmentation and with unified view of demand. And the last element to emphasize and I will stop after that is as good as our plan is, there are always deviation from the plan supply disruption, changes in demand. How can I quickly identify those deviations and disruption and respond effectively to those challenges? This is what I refer to in the article as smart execution. What cuts it causes for different capabilities is digitization, analytics and automation, and to do that, you want to take advantage of unified or a single platform that allow you to aggregate, allow you to normalize and allow you to visualize the data.
Joel Beal:
If I've understood you here, it's like those first three points that unified view of demand, the kind of having segmentation one size doesn't fit all and then that smart planning. Those all flow into let's create the best possible plan that we can. And then the last point is no matter how well we plan, there's always going to be deviations. That plan is wrong the moment it's printed and so we ought to be able to adjust, and I certainly agree with that. We see that a lot, and I want to drill into some of these a little bit. Let's talk about this unified view of demand, and you spoke about the way companies tend to plan today, this kind of consensus-driven process, each team bringing something forward. Do you see what you're proposing? Is this about speeding up that process? Is it about removing human bias from the process? Maybe it's both, maybe it's neither. Maybe there are other elements. I mean, where do you see the biggest opportunities and the biggest gains here?
David Simchi-Levi:
So if we dive into unified view of demand, there are elements from each of the comments that you made and maybe it will be an effective way to illustrate unified view of demand and its impact by considering one recent implementation at a large consumer-packaged good manufacturing company. This is a large company that we know well, that you know very well. It's a global company serving many, many retailers, thousands of SKUs, many distribution centers and manufacturing facilities. The characteristic of the supply chain before the transformation was very typical to CPG. It was all one size fits all consensus forecast, manual processes, and the company was very good at squeezing costs from different processes, focusing on improving the efficiency of different processes in the supply chain. But when you look into it, you realize inventory is high, service level is not necessarily where they want it to be, waste is high as well, and so the question was how can we change that? And the main focus of unified view of demand was to change the consensus forecast that is internally focused into a consumption-driven supply chain strategy. Let's repeat this is starting with consumption and generating a focus of what the retailer will order from the manufacturer. Now how do you achieve that objective? First, you focus on four different data sources. The first data element is all about the internal data that the company has Orders that they receive from retailers, product characteristics and so forth and so on. The second is consumption information about demand faced by retailers, and here typically you want to have point-of-sell data. The problem that this company has is that most retailers refuse to give them point-of-sell data, and so here we use the third-party data from companies like IRI and Nielsen, companies that we are all familiar with. The third was all about micro-economics data inflation, unemployment because this helped us understand customer demand. And the last element was all sorts of Google trends, social media information, social network information that allow us to better understand demand, especially for a new product introduction. It's these four sources of data that allow us to start by predicting what the retailer will face in the market. Think about this this is the CPG generate its own prediction of what retailer will face in the market, retailer by retailer, skew by skew, week by week in the case that we are focusing on for the next 80 weeks, and this prediction was an input into an engine that generate a forecast of what the retailer will order from the CPG. So the forecast of what the retailer will order from the CPG is driven by two important elements the historical orders from the retailer to the CPG that's obvious, that everybody's using, together with forecast of what the retailer will face in the market, skew by skew, retailer by retailer, week by week, for the next 80 weeks. And all of a sudden a process that was very manual and involve human making adjustment to the forecast became much more automated, where the focus of human is to understand the data and make sure that we agree on the data that we are using. And one element that I'm not sure always follows, but one element that was part of the process was don't sell though anybody else don't come and say, hey, I don't like the forecast increasing by 5% or decreasing by 10%. We had an incident when sales came and said look, the forecast in the West Coast is way higher than our intuition and experience suggests. That's all good, let's look at the data and see why we are wrong. And in that case it turns out that the number of stores that was documented in the data set where the product is being sold in the West Coast was way higher than the numbers that sales people knew were product are selling. Adjusting for the numbers reduce the gap between their intuition and what came out of the tool. The point is it's all the elements that you mentioned. It's automated, but it involves people in the process, human in the process to make sure that we use the right data to generate a focus.
Joel Beal:
I love that story that you just shared because I've certainly seen this in my career. I imagine you have a lot when you you get into analytics, data science, whatever you want to call it, I guess these days you bring the data and you run into, I guess, that intuition, as you call it. You know people who say I've been selling this product for 30 years. I have a pretty good gut reaction and it can't. There can be a lot of conflict there where people you know they'll try to Identify problems with the data and so let's just throw that all out, let's rely on my intuition. And and I loved your example of that hey, instead of saying let's bump that number five percent or let's change it, really trying to root cause what is? You know? People have a lot of ideas. That intuition comes from somewhere and sometimes people know things that aren't reflected in those models. But let's bring that intuition into the models, rather than kind of throwing it out and just saying let's just arbitrarily, you know, make changes here without describing why they are.
David Simchi-Levi:
So I think that Example of the number of locations a really good one and and maybe to support your point, one thing that I typically highlight in discussion around data science, analytics, unified view of demand, supply chain digitization, is that zero to success is not about just about science. When we see a killer opportunity, huge impact is when we bring together science and Art. Science is all the analytics and the data and the algorithms that we are talking about. Art is the intuition, the experience, the domain knowledge that business as supply chain professional have. Where I see biggest impact is when we combine the two into An important opportunity. The second thing, just to continue along the lines that you highlighted, is this is not just about science and art. This is not just about developing Effective focus. This is also and I'm sure you have seen this before this is also about being able to explain why is the focus, is the way it is. Nobody put in a different way and no decision-makers that I'm familiar with is going to look at the forecast coming out of a blackboard, say, hey, this is great, let's use it tomorrow. They want to understand what's happening. And when you think about explainability, there are multiple levels of explainability that you need to address, and part of the capability today is the ability to provide that insight. The explainability has three dimension. Let's enumerate them to highlight what is important and what can be done. The first people want to understand what drives my forecast. They look at your forecast, the forecast you generated or I generated, and say, hey, I see that this skew or this product family is going to do very well in the Midwest In the summer. Explain to why? Is it? Because our pricing, our marketing or just the product itself. So Identifying, explaining the driver of the forecast is one part of the story, but this is not the only one. The second one and the second one I'm sure again, you and your client have seen this before Is a comment coming from finance. You generate a forecast, you get a call from finance and and the father, the person on the other side of the call, is telling you hey, you gave me a forecast for December. That is, forecast for December is different than the forecast for December you gave me last month. What change? Why do I see 5% in December? It should have been exactly the same as the forecast you gave me a month ago. We need to make sure that we understand what drove the chain in the forecast that we are, because if if we cannot explain it, nobody will use it. I Right. And the last element is to be able to explain the difference between the forecast and reality, between the forecast and what we observe in the market. These are, to me, the three important capabilities that any effective Demand forecast tool needs to have to make sure that people trust and use the forecast.
Joel Beal:
So let's change gears a little bit and let's talk about what you mentioned at the beginning around supply chain resiliency. You know a topic that you know. I think you mentioned this in your article. You know these competing interests between that kind of efficiency, I think for the last couple decades. You know, with globalization, every company how do I, how do I squeeze down costs and you know that can compete to some degree with Resiliency, making sure that I can weather. You know, anything that comes up. There have been supply chain disruptions forever, but they probably haven't been in the news like they have the last couple years. So curious to get your your thoughts here on the right balance between those two and how you know companies what, what steps they should be taking to make sure that they have more resiliency in their supply chain.
David Simchi-Levi:
So. So let's start with what are the key capabilities in supply chain resiliency that we emphasize, that we found to be important when Companies is focusing not only on efficiency but also on resiliency, and there are three capabilities. The first one is planning. I want to be able to model my supply chain, analyze the supply chain and identify hidden risk and develop integration strategies Right. That's one important capability that now is recognized by companies as an important, as an area of Focus, and I've seen this surprisingly CPG high-tech automotive it's now a cost the board Because of what happened over the last couple of years. The second one is Is monitoring. Nothing is wrong, but I want to understand how much exposure to risk I have in my supply chain so that something goes wrong, I will know where my potential problems are. Think about this much like financial Investment, companies are measuring their exposure on a day-to-day basis for two, financial risk. Why exposure to supply chain risk is changing from week to week. Lead time is increasing, inventory depleted. I want to understand how and where there are changes in my exposure and, if it's below certain threshold, maybe expedite or maybe change my inventory strategy. And the last element is responding. Something happened. There is an earthquake in Thailand, one of my supplier facilities is affected. Right now I have enough for material to feed my assembly line. However, because of these disruptions that happened yesterday, six weeks from now I may not have enough for material to feed the assembly line. If I know this today, I will be able to respond effectively. And if you think about the story I told you about the beginning of the pandemic, that's actually what I did. I Basically use the technology to understand hey, something is happening in China, when is it going to hit the supply chain in North America and Europe? And if you did the same thing at the beginning of the pandemic, you would have secured raw material, a component for your supply chain, to make sure that you can feed the supply chain on an ongoing basis. All of these capabilities the planning, the monitoring and the responding rely on Data and Analytics. But one important element that we found in implementation of this across multiple companies even before the pandemic supply chain resiliency, surprisingly, is not necessarily going against Efficiency or cost cutting. So in the implementation at Ford, it's all in the paper, so there is nothing that that is confidential in the implementation at Ford, we found that for knowing much through a lot of inventory on the component Because we had a way to measure the resiliency of their supply chain. We identify cost savings opportunity. We were able to reduce or Ford was able to reduce inventory by certain percentage point and still maintain a resilient Supply chain. So supply chain resiliency is definitely about identifying hidden risk and developing mitigation strategies, but it's also about Identifying where they built too much inventory and where are the opportunities to cut inventory While still maintaining a resilient supply chain.
Joel Beal:
Yeah, I appreciate you sharing that, because You're right, and even the way I frame the question is that those things are at odds. You get efficiency or you get resiliency, and so that's very interesting. You know that example at Ford and presumably at many other companies, there would be opportunities to really do both. So fascinating. I also I liked your analogy of talking about finance. I previously was in financial technology, so I have a little background there and I do think of all these firms that have risk management functions and they're obviously constantly monitoring the portfolio, what the risk is. Doing that much more real time and if you're running a supply chain, you should be doing the same thing. But I want to go back to something you mentioned at the beginning around when we started talking about supply chain digitization, and you mentioned that it can be intimidating if you think of this as being a five, six year project, millions and millions of dollars that can be overwhelming, and that's something we hear a lot, you know. Just, it's like how do I get started? And you mentioned that you know getting started might be a lot easier, faster and cheaper than companies expected, and so curious to get your thoughts on how companies do start down that path.
David Simchi-Levi:
That's again a very good way to discuss my experience, at least in companies that focus on supply chain digitization, and the experience that I have is that you don't start with full fledge implementation the way I just described here. The focus that I've seen so far a lot of time is all about a proof of concept. That takes six to seven weeks that allow you to demonstrate the value, illustrate some of the capabilities and maybe convince yourself and your colleague that you should go in a digital transformation journey. And let me highlight what I think the key element in such a proof of concept. So a proof of concept is all about assessing where you are Remember I told you the story about the CPG company, some of the challenges that they had. So assessing where they are, identifying where you want to be what people in industry call the north star and develop a plan for closing the gap. In this plan, you want to demonstrate value. You want to illustrate with a small data set what is the improvement you can achieve on focus accuracy, and we never do focus accuracy improvement for the sake of the focus. We do it to reduce inventory, to increase service level, to increase revenue, to cut costs. If you can demonstrate that then you have an opportunity to open the door for a digital supply chain transformation in this process. You also want to develop the operating model, suggest how are we going to execute? Execute the digitized strategy. How do we make sure that the process sticks? Remember unified view of the end. You want to make sure that at the end of this process, the different functional area do not go back to their own department saying, ah, the old guy is wrong. We will continue using the focus that we generated using consensus model, using the old statistical model.
Joel Beal:
So identifying process and identify organizational structure that make sure that this new strategy is executed effectively is part of the six, seven weeks initial exercise now, okay, thank you, and I like your comment about because we hear this a lot people saying I want to be more accurate with my forecasting and it's like, well, yeah, we all do, but for what reason? What are you going to do with that? How's that going to, you know, reduce inventory costs, or, you know, increase sales, or whatever it may be. So I think, always aligning it to where the dollars, so as we close, can you tell, tell me and and our listeners, a little more about the data science lab at MIT the data science lab, as I mentioned, is a partnership with 25 companies.
David Simchi-Levi:
The areas that we cover start with supply chain resiliency we talked about it earlier supply chain digitization but we do way beyond that. We do a lot of work in the space of price optimization. We did work of price optimization for one of the largest online fashion retailer in Europe, a company called the lando, where every week, our analytics what comes out of the data science lab price 1.5 million different skills in 23 different countries. We do it for online platform and we do it for brick and mortar retailers a company like couple, which is a large retailer in Mexico, companies in the Middle East. We do a lot of work in personalized offering and let me explain this with an example. If you fly today, joel, from London to Paris, when you purchase your airline ticket, we are not involved, but once you purchase the airline ticket, our system kicks in and offer you an salary, product, priority boarding, current hotel, and the offers that you will get, joel, may be different than the offers that I will get, because the system is learning about individual preferences. We have done a lot of work on inventory, transportation and procurement, especially last mile delivery for a software company like Blue Yonder. So some of our partner are software companies that collaborate with us in a way that we develop the engine and they implemented as part of their tools excellent, and if someone wanted to get in touch with the data science lab or with you, what would be the best way to do? that just send me an email, dslavymitedu and I can provide more information about what we do, about our partners and about the impact of our technology on a variety of companies excellent.
Joel Beal:
Well, david, I so appreciate you joining us today and all your insights. Thank you, you've been listening to professor David Simchi-Levy, professor of engineering systems at MIT and head of the MIT data science lab. That's all for this week. See you next time on Shelf Life.