AI Reshapes Supply Chain, Retailers Like Walmart and Levi’s Jump on Innovative Tech

The supply chain has only grown more complex over the years, with COVID accelerating innovation within the space in order to meet rapidly changing consumer demands. But retailers struggle to cope with supply shortages, rising costs, uncertain deliveries, and more amid this volatile environment. RIS' new supply chain report shows there’s a transformation happening despite these obstacles. Agility and speed are at the forefront, and technologies like artificial intelligence and machine learning are leading the way. Take a deep dive into RIS’ research and case studies that showcase the industry’s most innovative uses of AI in the supply chain. Learn how companies like Levi’s, Mulberry, and Walmart are rising above existing challenges to flip traditional approaches to supply chain management and pivot toward intelligent automation to increase efficiency. In order to find a better way to navigate the retail supply chain landscape, retailers have turned to artificial intelligence and machine learning to help deliver the right products to the right place at the right time. In RIS’ latest webinar, we took a deeper dive into how innovation can accelerate intelligent demand planning, forecasting, and inventory visibility, as well as how to optimize daily replenishment decisions and align with service level targets with minimum inventory, using RIS’ recent report “Unveiling the Transformative Potential of AI and Automation Across the Supply Chain” as a backdrop. Albert Guffanti: Welcome to our webinar entitled “Predict the Unpredictable Supply Chain with Artificial Intelligence,” which is hosted by RIS and presented in partnership with Oracle. I'm Albert Guffanti, VP and group publisher of RIS. Thank you for joining us today. We are all woefully aware of the supply chain struggles the industry — and quite honestly, the world —has endured over the last couple of years. Retailers had to cope with supply chain shortages, rising costs, uncertain deliveries, and new demand patterns from consumers. The laundry list of challenges led forward-thinking retailers to ask, "Is there a better way to navigate the new normal while increasing our supply chain speed and agility?" Today, we’ll tackle that as we discuss how retailers have embraced the power of machine learning and artificial intelligence to not only meet those immediate challenges, but also set their organizations up for prolonged success. Standing still is no longer an option. Retailers must accelerate the supply chain evolution, and that evolution starts with machine learning and artificial intelligence. With that, I’d like to introduce our speakers for today. Joining us are David Fleischer, senior director of strategy and solution management for retail supply chain at Oracle, and our very own Jamie Goodman, senior editor of RIS News. Let’s begin by introducing yourselves — what your role is, what your focus is, how it relates to the webinar today, and some of your background. David Fleischer: It's great to be here today. My role at Oracle is to lead an incredible team in what we call retail supply chain, which covers demand forecasting and fulfillment and allocation. What I personally do is help develop the strategy and lead the product development teams in those areas. I'm not going to fancy myself a data scientist, because I'm not, but we have wonderful teams of data scientists. We're not going to go too far into the ML/AI world, but certainly, we work with it every day and how to incorporate that into these business processes which are key to demand planning and fulfillment. Jamie Goodman: Hi all, I'm happy to be here today. I'm the senior editor of RIS News, covering a lot of retail technology, all different facets of the industry. I'm up-to-date on all of the news from all of the top retailers and we're diving into it every day, different studies, surveys, etc. I've been with the company for seven years and also worked on CGT with the consumer good side. Prior to RIS, I worked in private label for many years, so I have a long history of retail and consumer products background. Guffanti: Excellent. Many of the topics we are going to discuss today were covered in RIS' recently released report called “Unveiling the Transformative Potential of AI and Automation Across the Supply Chain” We're going to explore some of the key themes from that report, but if anyone wants to take a deeper dive into that report, you can find it on risnews.com. Today, we're going to dive into four main sections. The first section that we're going to discuss is around demand planning and fulfillment forecasting. We all know that consumer expectations are nothing. They don't grow, consumers aren't getting less demanding with how and when they receive products, and retailers need to act with lightning speed regardless of the channel. This requires the ability to predict and plan. We've been speaking about this for a long time, but it's only getting more and more intense. In addition to anticipating a demand, retailers must quickly predict fulfillment options, how to leverage them quickly, and be prepared to execute quickly. There’s a lot to dive into around demand planning and fulfillment forecasting. Jamie, can you broadly discuss the critical role fulfillment forecasting plays in the modern retail environment? Goodman: This is becoming a huge topic today with multiple points of purchase and expanding buy online pickup in-store and curbside pickup. There's so many factors and so much data that now goes into where and how an order will be fulfilled. If retailers can't get this right, it's going to cost them. Just this week, we saw Walmart and Target’s earnings — these big retailers are still struggling. I pulled this from the Walmart earnings today: CFO, Brett Biggs said Walmart gross margin rate was down with more than three-fourths of decline related to higher than expected supply chain fuel and e-commerce fulfillment costs. Then, he noted that while they saw some supply chain improvement in the quarter early, the war in Ukraine and ongoing COVID impacts, including China, are leading to more challenges. As we can see, everybody's dealing with this. Target posted an earnings upset this morning and CEO Brian Cornell said that they faced unexpectedly high costs throughout the quarter, driven by a number of factors. It’s clear that everybody, even the huge players, are struggling with this. Any size retailer, no matter how big you are, the foundation for good fulfillment and fulfillment forecasting is going to start with knowing where inventory is needed in the first place. One example that we gave in the report was fashion brand Mulberry. This retailer is expanding, they're currently in 120 stores and online, but when they realized they were expanding internationally, both digitally and in-person, they needed to establish a unified view of inventory to know where stock was in stores, distribution centers, and other places. Mulberry deployed new order management technology, which helped to manage a wide range of fulfillment scenarios — including buy online, ship from warehouse; buy online, ship from store; and in-store pickup — whether the product is being fulfilled from the warehouse or the store's inventory. They've also gained capabilities for enterprise inventory visibility, order orchestration, rooting customer service, and store fulfillment. The technology also served a critical need when stores were closed during the pandemic, with the technology and customer experience director noting that the ability to sell through store inventory was a strategic advantage to the business during the pandemic. Another example of this, Finish Line has been working on applying AI to determine the best location to fulfill an order, which isn't always decided by geography. I'll talk a little bit more about them in a minute, but when it comes to demand forecasting, I reviewed some of the upcoming RIS and CGT Analytics Study. On both sides of the house, consumer goods companies and retailers are laser focused on forecasting demand. Study details will be shared at our Analytics Unite summit in June in Chicago. David, can you speak a little more in-depth on demand forecasting and how important it is. Fleischer: With new machine learning and AI models, the ability to use short-term demand-sensing forecast models is a huge opportunity for helping retailers to understand the trends and demand patterns. Shopping patterns are ever-evolving and the need to be able to pick up on trends even before they become a true trend is important. We call these micro trends, and they include all types of things. No longer can you wait seven weeks, nine weeks, or 12 weeks in a traditional retail planning methodology to analyze trends. With modern technology, you have your ML models that nightly or weekly go out, look at trends, and decipher what is going to be your best forecast going forward. In addition to that, in terms of demand forecasting with the introduction of AI and ML, the low-touch exception planning is key. Utilizing an enterprise-wide forecasting engine to automate processes is critical as retailers are faced with smaller teams. Retailers have always had to do more with less, but now it's even more difficult. Also, the one version of the truth that we all used to talk about data: How did we do last week? What's our gross margin here? Now, this one version of the truth has translated to forecasting as well. No longer in retail are we seeing different forecasts at an item level versus a category level versus a store level with these incredible advancements in retail demand forecasting and AI/ML powered forecasting. We can have one forecast that encompasses all of those different segments, and that's a huge leap forward in terms of forecasting. There's also no need for a large team. For instance, for short life cycle items and long life cycle items, and this and that, you used to have different teams that had to segment them and decide what the right forecasting methodology is. That's what we're looking at — how the ultimate forecast is a combination of the demand, as well as the fulfillment, what is going to be sold, and then where the customer is going to want to pick it up, by what method, and with what speed. Guffanti: Very interesting. Jamie, how is this technology being used to avoid other negative things such as markdowns and driving consumer purchasing decisions, and things like that? Goodman: Predicting demand properly is a big step in the right direction for retailers who want to avoid markdowns. I mentioned Finish Line earlier. They're deploying technology in partnership with Northwestern University's Retail Analytics Council, AI Lab. But they are testing the effectiveness of AI technology to optimize markdowns and improve margins by predicting customer demand. Collaborating with the Lab, Finish Line tested different algorithms to cluster, classify, and predict demand. Then, compared the markdown cadence it had historically used for selecting products to a different markdown cadence determined with the AI analysis. It found that the model anticipated a potential 8% improvement in gross margin over the historical cadence. There's some work being done here to test how AI can help with this. Levi's is also working on leveraging AI and demand forecasting. The results from their first wave test last year demonstrated that AI-driven demand forecasting improved accuracy. The CEO said at the time, “scaling it should enable more precise inventory investment, lead to less markdowns and clearance, prevent waste, and enhance sustainability, all of which improve our margins.” This will be powerful in combination with ongoing work AI has been contributing to pricing and promotion. Retailers are starting to dig into this, but there's a ton of room for the technology and benefits to grow. Fleischer: I'd like to add to that, as well. It all goes back to the forecast. Better plans and better forecasts reduce the markdowns, but retailers are always going to take big bets in fashion. Fashion isn't always something that even with the best forecast is going to hit. That's when you have to go ahead and start to promote things, no longer is the need to wait until your plan is marked down in the ninth week or 12th week, or something like that. With advancements in ML and AI, we can look at those micro trends — when something is accelerating in one area, in a few stores, in a cluster of stores or with a cluster, let's say of a customer type and we can localize promotional and markdown offers to be targeted, whether it's by product attributes, customer attributes, or a combination of that with pricing. For instance, Rolex watches. We are working with a jewelry chain where they've implemented AI and ML, and we could find through these micro trends little niche, customer segments, communities, different cultures, and things like that. Around Mother's Day in certain communities that are sometimes not as affluent as others, you wouldn't think to sell Rolex watches, but lots of Rolex watches get sold around Mother's Day in certain communities. By knowing that, and being able to adjust to that quickly, you can accelerate inventory to those places and satisfy customer demand that normally wouldn't be able to be satisfied in that brick-and-mortar store. Guffanti: We hear it all the time in our conversations at RIS News, how important identifying micro trends is. To your point, it takes AI and machine learning to make that happen. We're going to leave the demand planning and forecasting and advance to our next topic of discussion: inventory optimization and visibility. Obviously, it all ties together. As consumers do more online shopping, e-commerce becomes more and more prevalent. I know anytime I buy something online, if I don't have immediate visibility into where that is, even seconds after I've bought it, I get a little anxious. Visibility is equally important to retailers. They need to know where things are. If something is not accessible, it almost doesn't exist — we need that visibility. David, can you talk about what retailers can do to optimize visibility with inventory. Fleischer: That's a huge question, there's so many facets to that. It all starts with data; you have to have good solid data. It's a fundamental that many forget, but if you're not capturing data appropriately, successful planning becomes difficult. If a retailer fulfills demand from the store, that fulfilled demand needs to be separated from the overall store demand. You have to segment those out. It's interesting how many retailers are still struggling to have that fully mapped out. It would be surprising as to how many of the nation's biggest, publicly traded retailers still have challenges around separating stock to be sold in-store versus backroom fulfillment in-store, and being able to share inventory dynamically when they need to between those two channels. Also, it's understanding the demand regardless of channel and the local trade area demand. This varies greatly by the type of product. For instance, for grocery and pharmacy customers, there's this incredible enhanced need to have items close by. Whether it's the customer going to the brick-and-mortar store, buying online pickup in-store, or buying online and having it delivered, for grocery, pharmacy, and items like that, it's important to have them close by. Now, there's other areas such as appliances, furniture, fashion, which customers are a bit more flexible on and will travel a bit further to buy those things or wait a day or two longer to receive them. There isn’t a one size fits all model for optimizing where your inventory goes, it very much varies not only by customer, but by product segment as well. To me, it all comes back to forecasting. You can see I’m passionate about demand forecasting and about how it leads to all these decisions. Optimizing inventory starts with an AI/ML-powered, statistically-driven forecast. This also includes forecasting the method and speed of fulfillment. Now that we're capturing this customer data, we know in a particular zip code or market area with this type of product, how customers are typically and predictably going to want to receive the product. Again, it's combining those two things. We're back to where we were before, combining the traditional demand forecast with an optimized inventory plan. The question was about optimized inventory planning and to us, that's your demand forecast optimized across a time-phased inventory plan. Guffanti: Jamie, at RIS, we hear all the time about tying consumer demand with back office and the operations, and so on. Can you talk a little bit about inventory visibility, how it relates to personalization? Goodman: In our last supply chain tech study, we found that retailers named lack of real-time inventory visibility a top obstacle to overcome to prevent efficiency and productivity. This is because if you don't know what you have and where it is, then you can’t personalize your communications, you can't offer someone a snowboard they were looking for in green that you just got in-stock at their local store because you don't know that you got that snowboard in-stock. I said “snowboard” not “Burton” because I talked to them last month about the challenges they're trying to overcome with inventory visibility, knowing what inventory is available, and who's purchased what. This is empowering Burton's associates to give high touch, personalized service. They've been testing and sending text messages to communicate with their best customers in an effort to be more proactive with their needs. An example might be when a hot new product comes in, Burton can say, "The customer came into that location before looking for something similar, and the store didn't have it at the time," but now, they can communicate with the shopper that they have it. This is one way they're trying to overcome the challenge on how to personalize, but it all starts with inventory visibility. Sometimes when you can see that you have the stock or you don't have it, stock outs are going to occur, it's going to happen. One thing that Walmart's been working on is AI for grocery substitutions to ensure that shoppers are still getting what they want, even if it isn’t in-stock. Walmart has said that customer acceptance of substitutions increased 95% since they deployed the technology — that's pretty good. The EVP of Walmart Global Tech said in a blog that the decision on how to substitute is complex and highly personal to each customer, if the wrong choice is made, it can negatively impact customer satisfaction and increase costs. They're trying to overcome that challenge by developing their own deep learning AI technology solution, which they've said considers nearly 100 variables to determine the next best available item. It's looking at things like size, brand, price, shopper data, customer preference, current inventory, and doing it all in real-time. This is highly next-gen that they’re working on – the technology preemptively asks the customers to approve the substituted item. That signal is then fed back into the learning algorithm to improve accuracy of the future recommendations. Walmart says they're continuing to enhance the technology, but the cool thing is, with AI, the system is learning and getting smarter based on customer input and actions as time goes on. Fleischer: It is. Jamie, my family and I, we have three boys, we use a lot of groceries and are Walmart delivery shoppers. We started with it early in COVID and I'm familiar with the evolution of their substitution methodology — it has gotten a lot better, and has been noticeable as a consumer. Early on, I would get upset with the substitutions that were given and it would make me not want to order again. I’d order from Costco, or a different DoorDash or Instacart provider, etc., but now that they've improved, it has led us back again. It is the real world and I see a difference in it. Guffanti: We've clearly hit a nerve with this topic. Our next question revolves around predictive analytics, AI and machine learning. David, I'd like to ask whether or not you're seeing retailers collaborate much more closely with consumer goods brands in order to achieve this inventory optimization and enhanced visibility. Do you see a lot more collaboration going on with regard to demand planning? Fleischer: Yes, absolutely. It started before COVID, to be perfectly honest, with the acceleration of cloud-based, SaaS-based software and architecture. Retailers were able to allow CPG manufacturers to have access into their forecast very selectively, without sharing any sensitive data. They could only see their forecast, their brands, and things like that so that they could approve, deny, or make suggestions on the forecast. We'll get to the AI/ML part here in a second, but it's also from a manufacturing point of view. The sooner and more collaboratively you can plan with that manufacturer, the better percentage you are going to receive and timeliness on your orders. That was the first stage. Then, secondly, we are able to share — and the manufacturers are able to share — tokenized or anonymized consumer carts and consumer buying preferences. They have strong data analytic teams, and retailers have strong data analytic teams, and collaborating has definitely been a strong suit. No doubt about that. Guffanti: We are ready to move on to our next topic of discussion today, which is fulfillment and warehouse automation. For a lot of the reasons that we've already articulated, the modern supply chain needs to be capable of delivering products to and from warehouses, DCs, micro fulfillment centers, stores etc., quickly and seamlessly. David, how can artificial intelligence and machine learning help retailers streamline fulfillment from multiple channels into a single work stream? Fleischer: That's a great challenge that we're hearing about all the time. It's the same concept that we have been discussing, particularly around optimizing that inventory fulfillment plan. The better plan you can lay down a year ago, six months ago, 90 days ago, as we old planners know, you are always working on those different timeframes at the same time. The better plan you can lay down, the more streamlined fulfillment you're going to get. Again, this is fueled by being able to process huge data volumes. No longer can you pivot if you are in the spreadsheet, in the Excel planning world anymore, and process the amount of data coming in quickly enough in order to make all kinds of what-if fulfillment and automation planning decisions. You need to have a more advanced AI/ML-driven forecasting system to fulfill in a single work stream. Then, when you start layering on sustainability, which is definitely a focus of Oracle as well as our customers right now, in both the ordering and the return process, that adds a whole new layer onto this. Guffanti: Thank you for that incredible response. As you can all see, everything builds on everything else. Jamie, we all saw buy online pickup in-store and curbside pickup accelerate rapidly when COVID hit, to the point where stores were kicking the can down the road for years, then when they realized it was necessary, they implemented it within weeks. It's somewhat surprising that it continues to be such a popular option among people, even after the pandemic subsided. Can you talk a bit about that and how this is continuing to remain popular? Do you see it continuing to be a primary source of customer interaction, and if so, what can retailers do to enhance that experience? Goodman: I would say definitely. Store fulfillment is one model that every retailer should be considering to keep up with the digital demand. When it comes to store-based fulfillment, the most popular options are both curbside and ship-from-store. Our studies have found that lockers, self-serve pickup, are still lagging behind everything else quite a bit. We want to look at store fulfillment in the future. Walmart increased the number of digital orders coming from its stores by 170% in the last year to keep up with digital demand. This is leading it to scale its automated market fulfillment centers, tap machine learning, and test autonomous delivery. To support that, they built a tech platform to power the last mile delivery ecosystem. It's agnostic to supply and demand, built around its own marketplace and uses automation and machine learning to take the near-infinite number of factors and turn them into useful data. Walmart announced that it’s going to add more MFCs to stores, which are compact module warehouses that are built inside a store. These mini warehouses can store thousands of items — even fresh or frozen grocery — and then rather than sending a store associate to get items from store shelves, automated bots retrieve them from within these MFCs inside the stores. They recently announced they're planning dozens of these with many more to come. They're moving toward automation and keeping it closer to the store. That's one way to keep this in mind and change the experience that customers are having. To back this up, RIS research found 15% of retailers and CGRs are up-to-date or rolling out robotic systems for picking and material handling. Keep in mind, with labor being an issue and order fulfillment just needing to be faster and faster, we'll expect investment in automated systems to fulfill orders to rise, not just in warehouses, but in stores as well. Guffanti: We are entering our last topic of the day, which is order management. Again, this all ties together with our conversation of e-commerce — computer expectations accelerated, the need for flawless execution and demand planning fulfillment. This is no different from order management where any hiccup can create massive delays in profitability, negative impacts for retailers, and consumer dissatisfaction. Given everything we've discussed, Jamie, can you talk about order management and how it's evolved over the last couple of years? Goodman: Distributed order management is a top investment today, according to our tech study, followed by returns management, real-time inventory visibility, and e-commerce has taken on a whole new life in the past few years. We didn't even have curbside pickup pre-pandemic. Order management is going to remain important because customers want to be able to order online pickup — have it shipped the next day, within the next hour — the demands are getting bigger and bigger. If a store is wrong and the items are out-of-stock, they want to have it replaced. This is going to become very important. They also want to be able to do returns to a store, in an even easier way — you see Amazon partnering with Kohl's to let Kohl's customers return Amazon packages in the store in-person, and Walmart's InHome service, lets you leave the return on the counter, and they come in your house and pick it up when they deliver your groceries. We're seeing people testing new things and up-to-date, order management and store fulfillment is needed to keep up with this buy anywhere, pickup anywhere, return anywhere mentality. This shift in retail came relatively quickly. It picked up drastically since the pandemic. Within the next five years, everybody should have serious order management, keeping up with the new landscape, and doing store fulfillment — and doing it well — it shouldn't take that long to stand these things up now that we know how to do them. Guffanti: Easier said than done, though. David, where are you seeing retailers struggling and what should they be investing in to improve their order management capabilities? Fleischer: Jamie almost stole my thunder earlier. I would say you wouldn't believe it, but you would from your research and talking to the retailers, but even the largest big box retail chains here in the U.S., their store systems still aren't connected to the online systems necessarily — they're not connected to their merchandising and inventory management systems. In a lot of places, these are three systems by three different companies, of all different ages and generations and things like that. When you go to the counter and say, "I'm here to return my e-commerce order," and they can't find it or it takes them 5-10 minutes to find it because their store systems are different from their e-commerce systems. Many big, big retailers are still struggling with that. Since we're talking about order management, before COVID, we used to try to do order management almost at all cost to get it done, get that order satisfied. But here with the cost associated with labor shortages, with transportation. Again, Albert, how you're talking about it all, everything builds on each other, there's no doubt about it. Now, order management needs to be done profitably, not at all cost margin, AOR, etc., how many splits to fill an order, labor shipping, all these costs need to be taken into account. Again, back to AI/ML, the theme of the whole day today, which is that nothing can crunch this amount of data that quickly and give you so many iterations on a nightly basis of what's the best plan going forward as that customer transaction inventory and sales data that you need to have. That's what retailers are struggling with and that's where investments need to be made. Guffanti: David, for a retailer that's not using AI and machine learning in their supply chain, where is the best place to start? This is a great question, what's step one? Fleischer: Forecasting, it's a one-word answer. Demand forecasting and modern demand forecasting tools from Oracle, and from other vendors too, but Oracle AI and ML is embedded into forecasting. It's a cloud service, not a heavy lift, there's no strain on the IT and enterprise department, but we can get the data and give a great demand forecast to work on. That's got to be the first investment because everything we've been talking about is forecasting today. Guffanti: That was a great mic drop question because that answered a lot of the other questions that were asked, but yes. Let's start with forecasting. That's a great way to summarize things here. How about you both provide some wrap-up thoughts on today's webinar — give some food for thought as people go back to their daily jobs. Goodman: Keep in mind that you want to consider putting AI into substitutions to make sure that orders are being met regardless of whatever stock you have. This seems to me to be a no-brainer: if you have the order, you don't want to lose that order. That's easy. It's there, they want to purchase it. If you're looking for somewhere to start with AI, that's where to consider. With the demand forecasting and all of that, you want to have your inventory, but that's something that we should be exploring more going into the future and fixing the inventory visibility issue. Now is the time to experiment with automation, artificial intelligence, and machine learning. Even the big players are still working it out, so you have time, get in there, start exploring your options and testing things. Fleischer: Bringing it back to the theme of today's webinar (“Predict the Unpredictable Supply Chain with Artificial Intelligence”), in order to achieve that, your forecasting tools need a very granular level of data. They benefit all of your AI, ML — it might be doable at a higher level of data to some extent, but to get your best results, you need that customer transaction SKU-level data. That nightly daily data is going to be what's used to make your ultimate AI/ML-driven demand and fulfillment optimized forecast going forward. Guffanti: Great final thoughts to leave us with. Thank you both for providing such great insight into this very important topic. A lot of this discussion was based on a report that RIS recently released called “Unveiling the Transformative Potential of AI and Automation Across the Supply Chain.” David, Jamie, thank you for this great and important conversation, and thank you all for attending this webinar. We hope to see you soon. Have a great day.

You May Also Like

Starbucks Continues Five-Step Reinvention Plan with Focus on Rewards Program, Expanded Delivery Offerings, and Equipment Upgrades

RFID’s Evolution – From Addressing Post-Pandemic Supply Chain Issues to Where it's Heading in The Future