Data Rich But Insight Poor: How to Optimize Multi-Echelon Inventory

Speaker A: You really need to think about the fact that the faster and faster you get, the more you're going to have to invest. And that's something that those of us who work in supply chain understand. Data collection is so much cheaper and so much faster than it was even when I started in this field in 2013. I've realized I'm now on the third or fourth career in my time working. And so I started off as a Marine Corps officer on active duty. I see that the world is in the midst of a pretty significant geopolitical shift, and supply chains are part of that geopolitical shift.

Speaker B: Welcome to Supply Chain Optimizers, the show that uncovers the controversial strategies and candid stories of innovators and disruptors from some of the world's largest supply chain operations. Let's cut through the noise and optimize your logistics and supply chain one bold idea at a time. Welcome to Supply Chain Optimizers. I'm Diego Solorsano and today it is my pleasure to welcome Ralph Ascher, founder of Data Driven Supply Chain llc. Ralph combines a fascinating decade of experience in network design and supply chain analytics to, together with decision science and AI, everything in the mix to tackle some of the most complex supply chain challenges across industries. He's done it all. Cpg, automotive, agricultural, restaurants, everything. He's also a Marine Corps officer and an adjunct instructor at the University of St. Thomas where he shapes the next generation of supply chain leaders. In this episode, we'll dive very deeply on how data AI, uh, decision science and others are being applied to solve sticky supply chain problems, the role of, uh, systems thinking in modern supply chains, and how organizations can future proof their operations in an increasingly complex world. Ralph, welcome to the show. Great to have you here.

Speaker A: Thank you. That's great to talk to you, Diego.

Speaker B: Ralph, you have a pretty unique blend of experiences, right? You're serving in the Marine Corps, you are a startup founder, you know, data driven supply chain. You have a very unique career trajectory. Right. What inspired you to take this focus on decision science, AI and supply chains?

Speaker A: Yeah. Thank you. When I've looked back at it, I've realized I'm now on the third or fourth career in my, in my time working. And so I started off, yeah, as a Marine Corps officer, uh, on active duty, now as a communications officer. And that was kind of the seeds of getting into the supply chain analytics space because as a communications officer, I led Marines who were setting up, uh, technical systems like radios, telephones, computer systems that were support of, you know, broader Marine efforts. And it was a really fascinating position to be in as a young junior officer, you know, 22, 23 years old, because I was both leading Marines who were doing very technical work and using very technical equipment. And I needed to understand how to employ that kind of technical equipment. But then I also needed to understand how that all fit into the much broader picture. Right. And be able to communicate with people who are not doing the technical work, but need to understand how we're all going to work together towards the bigger picture. And so that has really carried me forward into the supply chain analytics space because all my corporate experience and then I was uh, as a consultant is working primarily with supply chain leaders who have these supply chain objectives. And they may not know or care really about the math that's going on under the hood. They really just care about, hey, how am I going to help them with their objective. And that was a kind of a trait that I took directly from the Marine Corps. And how did I land in supply chain analytics specifically while I was uh. It's an interesting story because I think a lot of careers are not necessarily defined by the intentional twists and turns and decisions you make, but rather just by happy accident, so to speak, and you recognize the opportunities there. And so I was deployed to Afghanistan in the winter of 2009, 2010, with an, uh, Air Force officer who had gone to the Air Force Academy in Colorado for his undergraduate uh, degree. And he had uh, actually majored in operations research. And I did not know what operations research was at all. I'd never heard of that. I majored in physics in my hometown in Indiana. And I started talking to him like, oh, this is pretty neat. You know, it's basically like a bunch of different math techniques and, but you know, you use them all to solve real life problems. And that was the real life application of the math is what appealed to me as a physics major. And I thought, hey, well, maybe I could study this, you know, after, uh, after the military and go back for a graduate degree. And so that's what I did. I, uh, went back to school for a master's in operations research and did that while my latter years in active duty in the Marine Corps. And then at about the same time I graduated, I also, uh, lacked active duty, moved to Minnesota, and shortly afterwards began working in supply chain design at General Mills, the big CPG manufacturer here in Minneapolis. Uh, worked there for a couple years and then I was recruited to work at Target in their network design group, whereas mostly in E Commerce. And I can go into more detail if you want to know about that? And after about six years of that decided to uh, start my own company and data driven supply chain and started with me and now I'm fortunate to have a small team that works with me to help our clients across those kind of industries that you described.

Speaker B: Pretty cool story Arch. Particularly enjoyed you know, hearing about your officer. Right. That kind of like. Yeah, I don't know. There's something about mentors isn't it that really can change the trajectory of your life? I really, really do believe it.

Speaker A: Oh yes. I can look at two or three of my real of the mentors that I had as a marine officer just really changed my life. Even if they're not technical people, they're just giving me great piece of advice. I'm like hey wow, that's, that makes a ton of sense and went with it.

Speaker B: Yeah, I have an outside of the armed forces, uh mentors who changed my life trajectory. So tell me a little bit about Target network design and General Mills supply chain. What were you doing there and what was it that sparked the curiosity to start your own company, helping these companies solve some problems? Maybe what were those problems that you were working on that you said hey maybe there's an opportunity here for launching my own company?

Speaker A: Yeah. So at General Mills I was in the network design group which uh, had a few different names over the course I was there but uh, basically I was doing distribution network design uh for different temperature channels. So General Mills has food at both the ambient temperature we think of like cereal, right. That's most commonly known snack foods but then also refrigerated food, uh and then also some frozen foods that have different distribution channels for those temp channels. And then I also at that time worked on a couple manufacturing network design projects for uh, international expansions. They were uh, expanding quite a bit and uh, acquiring companies outside the US so was helping those acquisitions with that. And then when I was at Target I was primarily working on the expansion of the E commerce network, doing the supply chain models, building those and then uh, helping guide the strategy and inform it from a modeling perspective on expanding out first the ship from store network and then ultimately my latter couple of years that was there on the sortation center network which uh, if you see anything in uh kind of their news releases and the business process. Essentially the idea that in a city that has a bunch of different Target stores, if you as a guest, as a customer can order something from the website, we don't necessarily need to take it pick and pack at the store and then Give it to a um, like a UPS or a FedEx to deliver to you, but rather can control that middle mile, bring it to a centralized location called a sortation center, which it then goes out on a gig worker route so that you and your neighbor both shopping at Target are getting deliveries on the same route, even if the products that you're receiving did not come from the same store. So it's kind of like using that whole metropolitan market as a single inventory source to be able to drive efficient last mile routing.

Speaker B: Yeah, and that's a lot of what you, yeah. You even posted recently on LinkedIn, right. You, you for, specifically for the Minneapolis and Paul area, basically you had an article on design on last mile delivery specifically there, right. For a 30 minute delivery. And you came some very interesting conclusions.

Speaker A: Yeah, and that was, it was interesting because when I was thinking about, hey, uh, what can I tell to uh, you know, broader audience about last mile delivery and it's that you really need to think about the fact that the faster and faster you get, the more you're going to have to invest. And that's something that those of us who work in supply chain understand is that, you know, once you get up to a pretty high level of service, however you define it, that next incremental improvement is going to not cost. A 10% improvement in your service is not going to cost 10% more. It might cost 50% more or 200% more. And that's a trade off supply chain leaders need to take into account. And that's something that people with backgrounds in advanced analytics data science or can help guide.

Speaker B: You also chat, uh, quite a bit about systems thinking, right. And going to your webpage, et cetera, it's very central to your approach. So how is thinking in systems rather than silos, how has this changed the way in which you perceive supply chains and in how supply chains are designed and managed? Maybe if you can share an example, right. Where this approach has actually dramatically improved supply chain performance, that would be fantastic.

Speaker A: Yeah. So one of the reasons why I emphasize systems thinking is that most people who work in supply chain, uh, you know, at least if you're below say the executive level, you're focused on one function, right? Like you're either in procurement, you're in sourcing, you're in transportation, you're in warehousing, and you know that part of the business very, very well, you know where the problems are, you have really good ideas to improve it. You know that cold. But one thing that when we look at supply chains from a Systems perspective, we realize that each one of those silos has its own goals. But oftentimes the KPIs and the goals and ways to approve those silos can be counterproductive and actually hurt performance with another silo.

Speaker B: Can you give us an example of that? I agree. A fascinating topic.

Speaker A: Yeah, so like for example supply chain network design, which is uh, what we primarily do in our consulting business is the question of asking where should my facilities be. Whether I'm talking about my plants, my distribution centers or something like that. And let's just focus on the last leg from a distribution center to the customer. Whether we're talking about a business to business or business community, that last leg, the closer you are to your customer, the more distribution centers you have. Generally the closer you are to your customer that reduces that last mile cost on average. Right. Because you're closer. And however the more facilities you have, the more you're going to pay in capex, the more you're going to pay in opex, the more you're going to have inventory bloat. We saw that in the post I just made. And so as we think about hey, the last mile cost going down, well warehousing cost and um, our inventory cost is going to go up and it's not a one for one trade off unfortunately. So it's not like we can just say hey I want this much inventory and this much last mile cost but rather we need to think about it in a system and that's where things like optimization, modeling, which is the math that underlies a lot of ah, all network design models including commercial and open source approaches that allows us to think in systems and say hey, let's identify how these decisions are related and then let's try to quantify the impacts of these decisions and then we can use an optimization model solver, give us a suggestion on how to move forward in a way that balances the trade offs between the different silos. So that's what I mean by systems thinking, trying to understand how all these things are connected and how we need to make a holistic decision about it.

Speaker B: Yeah, it's interesting and I appreciate the example. Also really begs the question, right, okay, so you have a bunch of factors to make a decision. Okay, if I increase the number of warehouses then my last mile cost does dramatically drop but maybe that comes at the expense of more capex and more inventory.

Speaker A: Right.

Speaker B: Because I have to fill those warehouses. How do companies collect these data to make these decisions? And these models, right, there's a lot of data Collection is very cheap. Cloud computing now is extremely cheap as well. It makes these large scale supply chain models easier.

Speaker A: Right.

Speaker B: Than it was certainly a decade ago or even, you know, three, five years ago. How do you go around collecting data to make these decisions and these models?

Speaker A: That's a great question. And I think that this is where it's always a little eye opening going into every new client because oftentimes our clients have never done a full supply chain optimization model. And so they, you know, they've probably been making very good decisions with their business so far, but no, now they're wanting to pursue, uh, doing something like a network design. And they've just never had to have all this data in one spot. Like they've never had somebody ask for all these different things. And it is absolutely true that data collection is so much cheaper and so much faster than it was even when I started in this field in 2013. But still, even businesses that are data rich tend to be insight poor. And oftentimes the key inputs for these kind of models are not things that are being tracked. So some of the key inputs are forward facing demand. Okay, well, how are you doing forward facing demand forecast at a granular level? Or are you just basically saying the business is going to grow by x percent year on year and a rising tide lifts all boats? Well, that's a big input. And it's not necessarily being produced at that level of granularity that it's being collected. Right. The level at which we're collecting data and the level at which we're making decisions in forward facing projections. Not necessarily the same level. Things like cost structures. Very difficult in practice to actually, uh, tease out. You know, if you go into a manufacturing plant and say for this individual skew, what's the cost per unit to produce this? That's going to be a real intense discussion with that financial analyst. Right. Because it's like, how are we going to allocate raw material cost? How are we going to allocate electricity and labor and overhead? It's oftentimes comes down to, uh, an educated judgment call. There's not necessarily a definitive way and established process for doing these kinds of things. So it's a mixed bag. No, for sure.

Speaker B: And I want to stay a little bit here on. You made a couple very good points that I want to double click on and I think our audience would appreciate. First one, you said so far, companies, you've made very good decisions. Right. And the question here specifically is, do you think if you go to any company, pretty much in the world. Right. Any company that has a complex supply chain or even not that complex. Right. Any company that has some leveling of decision in any company. Right. They've so far been successful, they've made good decisions, as you put it. Is there optimization opportunities for any company that you touch?

Speaker A: Yes, I think if they're willing to make the changes, absolutely. Something that I have found, especially with mature companies, because we tend to be working with larger companies, uh, just due to the size of the problems. And so with larger companies they're usually pretty mature. And so the big inefficiencies have usually been found and stamped out long before we get there. And so the question is, well, if we make the decision, how can you help us? How can you make things a little bit better? And one thing I have found is that the first step is always just to talk to the subject matter experts and the supply chain managers, the operations managers who've been working there 5, 10, 15, 20, 30 years, because they usually have pretty good insights about, oh, this is what's going right, this is what's going wrong. And when you see, how are they doing things right now, say, okay, from a scale of 1 to 100, let's say that the way you've got it right now is like a 92. Like we'll find how much of that remaining 8% we can try to, uh, close that gap. And especially in very mature companies, if you can improve their operation or suggest ways to improve their operations by 2 to 3 percentage points, that, that can be a huge amount of, uh, margin. Right. I mean, 2% of a big number is still a big number. Oh yeah.

Speaker B: The other concept that you brought up, which I really appreciate, is data rich inside poor. What do you mean by that?

Speaker A: So a lot of companies are collecting data at extremely granular levels, particularly around sales and inventory. You know, if they're running an ERP or CRM, they probably are, but they aren't necessarily using that data to make even tactical decisions. One thing that I like to, um, talk about when I want to give kind of introductory presentations to audiences, talk about the supply chain, uh, analytics pyramid. And this might be a little bit outdated of, uh, of terminology, but I do like to use that, uh, terminology of descriptive analytics, diagnostic analytics, predictive analytics, and then prescriptive analytics. I like to break that down because most companies are still just trying to get the descriptive analytics in place. Right? They're trying to get dashboarding in place that will help them make good decisions. And I, for better or worse, uh, things like Data management, like what you all focus on and trying to get, uh, that's not, that's not as high of a priority as it needs to be. You know, I'm a, I like to tell people I'm a lifelong Midwesterner. Outside of being in the Marine Corps for uh, years I've lived in every, every house in the Midwest has a basement. Why is the basement there? It's to hold up the house. You don't see it, but you're not going to be able to have strong analytics that drive way really good decisions if you don't have that data structure in place. And a lot of companies, you know, they want to say they're using artificial intelligence, right. And want to jump to these proof cases. But in reality vast majority of companies large and small would be best focused on just saying how can we get the data? We have to get actionable insights, both tactical and strategic levels.

Speaker B: Which maybe the follow up question to that. All right, you brought it up right Yourself, right. So now you're in the hook for the word AI in this context. Right. And uh, thinking also the pyramid of data analytics for supply chains. So what is AI and what isn't AI?

Speaker A: That's a great question. So one thing I would like to say is I uh, do like the definition that Mike Watson who's in previous guest years, he's saying that it's like practical, I think he's used the term practical AI. It's basically saying how can we use mathematics and everything under the hood to make better decisions? And most traditional operations research like optimization simulation, I would count as that. Nobody's going to take out a piece of paper and a pen or a spreadsheet and build an optimization model with 20 million decision variables and 70,000 constraints. Right. But that's extremely common in a supply chain model. Right. So in that sense I would say it's artificial intelligence because it's a computer doing something there's no way I'm ever going to be able to do. And I do think that term is broader than just gen AI, but I think that it can get way too wide. Like I saw recently a uh, RFI from um, a company that was asking for data visualization. AI? No, that's data, uh, visualization is, is not artificial intelligence. Right. Like that's just being able to present your data in a way that people can, can make insights for. I like to keep a pretty large definition but not, not a big tent. But it's not a, it's not a circus tent, it's More like maybe a ten man tent.

Speaker B: Yeah, it's, it's interesting on, on the, especially in the concept of a large scale optimization model in which 0% chance any human could, you know, incorporate as many variables, et cetera. Well that looks pretty smart, right? Like the machine's doing something pretty smart. Right. So maybe that is also uh, artificial intelligence. Right. Maybe kind of like a controversial opinion to the uh, like very purist AI practitioners.

Speaker A: Yeah. And I think that that's where if you want to get like super math geeky about it, an optimization model, for example. If I build a really big optimization model and I send it to Grobi as a solver, I see the progress report. Do I know, really know what's going on with it? No, there's, there's a level of trust that, hey, I know of the broad strokes of what's happening and I just trust that the good people at Groby are uh, building great products. Same thing. If I use an LLM like ChatGPT, I know there's a probabilistic model under the hood, but I certainly don't know the specifics of it. So there is a level of you got to trust and be able to some uh, level of black box I would say with optimization it's more like a gray box through a glass darkly.

Speaker B: Uh, let's switch gears. Well, not that a little bit more here on this topic of AI, but also introducing the concept of decision science. We often discuss those two together in your work and your posts, everything. So how do you combine these two disciplines and how does that translate into real world models, insights and solutions for your customers?

Speaker A: If you use the definition of artificial intelligence that I did, which is more of a broader tent of analytical approaches in a way that human brains can't do, you know, at least not at all at the scale and speed that uh, the models can. If you use that, the decision science is. Well, how do we use those approaches towards making good decisions? Right. Improving operations. So, and this is very um, context specific, you can know optimization model in cold, but if you don't know anything about supply chain, you're not going to be a very effective decision scientist in the context of supply chain. I know virtually nothing about marketing analytics. So like even though I know enough optimization, you put me in front of that uh, marketing, I'm not going to be as a effectively somebody who uh, does it right. So I think decision science is knowing when and where to apply those techniques. And that involves not only the analytical knowledge but also a pretty significant level of domain knowledge to what you're helping.

Speaker B: Let's try to make it specific. Right. To the extent that you can talk about a real life project. Can you walk us through a recent project where AI or decision science or the two combined actually solve the problem?

Speaker A: One project that I'm very um, proud of is we did a project for a national uh, auto parts distributor. So I'm going to keep it to level where I can, can uh, but basically their business model is their B2B and their customers are local auto repair shops. So with both chain and independent ones. And so the, the idea is that they have like a hundred distribution centers across the United States. So you know, pretty much any city's got one fairly pretty close to them. And at the end of the day a auto uh, shop will call them up, uh, that local DC and say hey I for tomorrow I need these 20 different SKUs. And it could be you know, any, anything pretty much. Right. 20 different skis and this distributor. Their business model is that they're going to have uh, everything you want. Right. Because they want to be that one stop shops. Because otherwise that auto shop's not going to call up three or four. Yeah, you're going to lose the customer. Skewed. Skew breadth is really there and speed being able to get that stuff to them early in the morning so they can work on those cars next day. And so when you think about an operating model that has extremely wide SKU breadth and a lot of distribution centers, proximity to the customers, it's like that post I just did a lot of SKUs really, really close where you got a lot of inventory. And so I was approached by this company saying hey, we know we have a lot of SKUs that rarely, we rarely sell, you know, and when you look at it some of them are like twice in the last three years they sold. And so we want some help in identifying what SKUs we can potentially move up to an upper echelon. So what they're thinking is that still on those hundred local VCs, but then they also would then establish somewhere between five to ten regional distribution centers which you would then push some subset of the lower selling SKUs, not necessarily just an ABC kind of thing, but some subset of the lower SKUs to the regional ones. Then when they are needed you can push them down to the local DCs and you get the, you know, the inventory pulling effects by having that in fewer locations. So it came to us and said hey, first off, can you give us some recommendations on where we should be looking to build these regional DCs in terms of, you know, kind of a standard network design model. And then once you've done that, can you give us some suggestions on what skus to put into it? And this was really, really fascinating because it was both an inventory question and a network design question. And I find that a lot of times with clients they say they have an inventory question or they say they have a network design question, but really they have both because you can't separate those two questions. It's mostly just what's more important to you right now and in the long run. And so with this client we did uh, again first, uh, kind of a standard network design model and came up with certain number and location of these regional DCs they'd recommend and say, okay, now if we're going to proceed with that plan, we'll build an optimization model that basically says what should you be pushing up? And this was a really fascinating kind of question with the inventory because it wasn't as simple as just saying Pareto curve chop off here and push everything up. Because with some of these products, first off, by pushing the product up to the regional distribution center, you are adding some risk of a lost sale, right? You know, what if the truck from the regional distribution of the lower drops breaks down or something? They can't get it. So there's some goodwill loss there. But more importantly, with some of these products there was a significant rebate if you bought them in bulk. And to the extent that where if you bought 99 each of an item it cost X, if you bought 100 each's it cost 80% of X, right? It was actually less total if you bought more. Which kind of throws the whole economics of things off. Right. And so it was, we built basically simulation model says, well, taking into account inventory costs and lost sales risks and rebate considerations, what's the benefit of keeping something at the higher level versus the lower level? And then once you have that, let's build a constrained optimization model that says let's push what this up here, this down here. Because we can't realistically have uh, some limitations. We do have some limitations in terms of uh, storage at the regional level. So built that and the end result was uh, a SKU by sku, uh location by location recommendation on what to hold at what level. So it was really cool because again it's an inventory question and a network design question. And we use those AI approaches, classical uh, operations research, optimization and simulation combined with expertise and splatter knowledge Very interesting.

Speaker B: Thank you for sharing that couple of closing questions. Rolf. You know a lot of organizations struggle with implementing right really uh, translating these models, data analytics, AI into real world execution. And I know from your role you're not exactly or at all responsible for the execution of it, right? Like hey, here's the model and like you do you company uh, right. Have you seen some of this actually turn into real life execution? And do you have any recommendations for companies to actually push through in the implementation of these types of uh, operational changes?

Speaker A: Yeah, it's a good question. Yeah. As a consultant. Yes. Or uh, we're not seeing through the execution but I have uh, I have been part of seeing my clients start to execute this and of course in the corporate world I saw it executed as well. I mean I think the biggest thing is you need to identify everybody that would be involved in the execution side early on, like before the modeling begins. You need to have ideally project management office already have somebody allocated, you know, dedicated to it. You need to have you know, the finance people, the construction people, commercial real estate. You know, there's, there's a whole laundry list of people needing about this team and if you don't then you're basically just going to have a nice PowerPoint presentation and then like well how are we going to do this? We forgot to notify you know, these six key individuals, you know and that's, that's not a good way to execute anything. Right. The whole idea around or and decision science is we want to make better decisions and execute on those better decisions and just planning and project management upfront is key from what I've seen.

Speaker B: Final question, looking ahead, what excites you the most about the future of AI data, decision science and supply chains? Maybe like, yeah, like also like a non obvious technology that's, you know, you think we'll, we'll hear a lot about in the next few years.

Speaker A: Oh gosh. Um, actually what excites me the most, and this was definitely the hardest question to prepare for, I'll tell you that one. That was a really good question. You know, I think that what excites me and really makes me curious for the future is not really the technology side. It's much more that I see that the world is in the midst of a pretty significant geopolitical shift and supply chains are part of that geopolitical shift. Right. And you know, no matter what's going on in Washington or whoever's in uh, power, we're going to see some really significant changes. And I think that that will most affect the world of sourcing and procurement. From a technical perspective, I'm most interested in seeing some of the technologies that are coming out to improve sourcing analytics, especially origin, and knowing your tier 2, 3, 4 suppliers above it. Because a lot of people just, it's oftentimes a mystery. Blanking on the name of the book. I can probably find it afterwards, maybe put in the show notes. Ah, there's a book pretty recently talking about the different materials that make up the world and it's like copper and lithium, um, and silicon. And realize like for all, even people who work in supply chain management don't necessarily really know where this core stuff is coming from. And I think that's going to be the biggest point of opportunity for supply chain managers to kind of future proof their supply chains.

Speaker B: Man. Wow. That's a very unique answer. I appreciate you taking the time to prepare for that one. It really shows up. Ralf, thank you so much for doing this. I really, really enjoyed the conversation.

Speaker A: Thank you very much, Diego. It was a real joy to be on here.

Speaker B: That's a wrap on today's deep dive with supply chain optimizers. If you found value in our controversial tactics and data driven stories, don't forget to hit, follow and subscribe. So you never miss an episode, have a burning question, or want to share your own optimization success. Connect with me, Diego solors, uh, on LinkedIn or for more information on how we can help you transform your supply chain and logistics operations, visit the steaah.com. thanks for listening.