How Close Can We Get to a Self-Driving Supply Chain?

It’s never a great sign when people are talking about the supply chain. With the pandemic stress-testing it to new levels, however, it’s been the hottest topic of conversation for nearly 18 months. As more consumer goods brands turn to tech to maintain agility and transform their planning process, a recent webinar shared insight from IDC about when we can expect things to right-size themselves, the valuable role AI can play in forecasting, and why tech can’t solve all our problems. Read on for the webinar transcript and all of the presentation slides. Lisa Johnston: Hello, everyone. Welcome to “Redefining Consumer Goods Demand Planning.” My name is Lisa Johnston. I'm the managing editor at CGT and I'm excited to welcome you here today. Joining us today are Simon Ellis, program VP at IDC. He currently leads the global supply chain strategies practices at IDC Manufacturing Insights, specializing in advising clients on supply chain digital transformation, network and ecosystem design, supply chain planning, global sourcing, transportation and logistics. He has almost 35 years of experience in manufacturing working across all major areas of the supply chain. He previously was also the supply chain strategy director for Unilever, North America. Next, we have Ben Hosseinzadeh, VP supply chain at Caulipower. Ben's responsible for the company's strategic supply chain planning and execution, and he and his team manage all aspects of the supply chain from demand planning to delivery of goods to customers. Finally, we have Shastri Mahadeo. Shastri's a serial entrepreneur who's been building companies in the food and beverage industry since the age of 18. He's now co-founder and CEO of Unioncrate, an AI-powered, integrated business planning platform built specifically for the consumer goods industry. Prior to founding Unioncrate, Shastri was the founder and CEO of a specialty beverage brand and he founded Unioncrate based on the struggles he faced operating that business. Thank you, everyone, for joining us. Now this webinar is an extension of “Not Normal Just Yet: Demand Planning in 2021,” which is the recently published special report by CGT. The truth is, and this likely comes as no surprise to anyone here, that while optimism is certainly growing as we all look forward to putting the pandemic behind us, the consumer goods supply chain will have to deal with its ramifications for quite some time. One executive we spoke with while preparing the report compared the disruption with 9/11, noting that inventory levels in that case took about 18 months to reset. As part of the report, we received insight from Lora Cecere of Supply Chain Insights. She provided details from a study she conducted last fall with supply chain leaders, which revealed that just 13% of companies strongly agreed their supply chain was working well during the pandemic. This is down 18% before the pandemic. During the pandemic, many companies were unable to use their demand planning solutions and many of them simply just turned them off and kind of white-knuckled their way through it. Now the issue with this strategy of course, is that order and shipment data was out-of-step with the market. The models weren't flexible enough to use consumption data to understand the changing market conditions. Companies are now rethinking their demand solutions in order to be more nimble in the use of channel data and flexibility of models. As a result, Lora tells us to expect the step change in the building of outside in processes and improvements in modeling. A recent article in Harvard Business Review meanwhile, noted that while many consumer-facing companies responded to the degrading forecast accuracy by trimming their production and marketing, the ones that pursued new data sets, simulations and model developments actually had initial success in better predicting demand. As a result, these companies are expected to be in a stronger position once the market stabilizes and other companies are being strongly encouraged to follow suit. Now Simon was also a contributor to this report and I'm really excited to have him here to share his insight. He's going to share some additional insight today that's not in the report. Simon, welcome. Ellis: Thank you, Lisa, pleasure to be here today. I work for IDC. We always joke that data is our middle name because actually, it literally is International Data Corporation. I think we saw 2020 was a crazy year for everybody for many, many different reasons. I always used to say that when country leaders, prime ministers, presidents, whomever talk about the supply chain, that's not a good thing. It's not because the supply chain is really well, it's because we have issues. At the same time, I actually think in many ways that the supply chain performed quite well. Let's drill into the data a little bit. I think there's a tendency — and I fell into this a little bit last year as well — to think of COVID as sort of a disruption. But it wasn't actually a disruption. It was a whole bunch of individual disruptions. Factories in some parts of the world closed, where in other parts of the world, it didn't. When those factories reopened, perhaps factories in other parts of the world then closed. Or social distancing, staying at home, closing of businesses meant we saw demand shift in different timings. We saw impact and demand earlier in Asia, a little later in Europe and North America. So it was this sort of weird year where it wasn't just one disruption. It was a whole bunch of disruptions. Given that I love data, and I had an opportunity last year to do a couple of surveys — one in March and one in December — what were some of the things that we heard? Clearly demand has declined. That actually was not in all industries because at the bottom you see demand has increased. If you happened to be, for example, a pet food manufacturer, your business probably did really well. If you were in the novelty or impulse business (candy, gum, movie theaters were closed, airports), you dramatically reduced foot traffic, so people aren't walking by the shops in the airport and impulse buying the things that they normally would. So we saw this interesting demand decline certainly, but also demand increased in other cases. We also saw some hoarding behaviors. Anyone who tried to get paper products in March and April, they were hard to come by. Not because people were necessarily using more, but there was certainly some hoarding behavior going on. We saw transportation delivery delays and that impacted demand. It impacted the ability to deliver orders, to deliver products. Suppliers were unreliable or unpredictable because suppliers have their suppliers, and their suppliers have their suppliers. So we saw this knock-down effect of supply. We still see a little bit of that, not in the consumer goods industries necessarily, but the automotive manufacturers are struggling because there's a silicon chip shortage. So that will have to play out over the course of 2021, as well. But again, a mixed impact over the year. Certainly demand's a big part of it. It's a bit of a pet peeve of mine that when folks talked about the supply chain, the focus was always on the supply side. I get it because of “supplies” and the term “supply chain,” but in many ways, the demand impact and demand implications were broader and deeper, and I think will be longer lasting. Lots of things that could help with demand planning, certainly we're seeing improved demand focused on algorithms. We're seeing companies use data better. The ability to use predictive tools for both today and tomorrow helps a lot. Visibility — we've been chasing visibility in the supply chain for a long time. Some companies have done better than others, but we're finally to the point where there are now visibility tools that can give companies a much better view to do inventory certainly because inventory was a real challenge. A lot of companies found that as we've historically had, we've had the wrong things sent to wrong places invariably. So visibility into better inventory or smarter inventory and data is really important. Then it's easy, maybe even a bit lazy of us to say, well, technology will solve that problem — sometimes it can, sometimes it cannot — but I know many companies that are starting to use artificial intelligence, cognitive computing, whatever term you'd like to use, to start thinking about how to better manage forecasting. How to balance responsiveness and forecasting. It doesn't always have to be a forecast; it may be that I have very short-term real-time data and I simply react to it. Artificial intelligence, cognitive computing can play a big role there. I've touched on some of this, but we see a lot of focus around this notion of an automated supply chain. Somebody the other day mentioned the term self-driving supply chain, which I thought was quite good. This ability to leverage people where people excel, and leverage technology where technology excels. A lot of the technology that we're starting to see can have a massive impact on, or substantial impact on, forecasts and forecasting improvements. From a demand challenges perspective if you focus solely on the demand side and set aside supply, set aside transportation, set aside inventory. A lot of companies saw demand increasing. Much of it was consumer hoarding behaviors, but not all. If you were in the barbecuing business or pressure treated lumber, it was real demand increases. Other companies say demand has declined significantly. Again, there was an element of dumb luck. If you happened to be in any services, true. If you happened to be a restaurant in the city, your business got crushed through no fault of your own. Again, some companies saw an actual increase in demand, and then almost 20% say, "Actually, we haven't had any demand planning or forecasting issues." That number will always be considered in this sort of 25-30% range. Now it's down to 18, so clearly, the broader notion of demand forecasting declined and deteriorated in 2020. Be clear about what happened in 2020: What went right, what went wrong. Where were the issues? How do you start to think about managing demand volatility better? There's one company that I work a lot with who said, “The only thing true about any forecast is it's going to be wrong.” It doesn't mean you shouldn't forecast or try to improve the forecast, but I do think you have to recognize there will always be times when the forecast proves to be wrong. Then you have to start thinking about the agility of your supply chain. How can I quickly respond? Maybe demand went up, maybe demand went down, maybe it went up in the very short term, or maybe it went down in the very short term. How do I start thinking about agility? Is it about being more flexible in terms of how I think of my inventory? Is it better to pull levers in factories and do line changeovers in a more nimble way? Whatever it is, in some ways, it's the flip side of the forecasting coin. If demanding forecasting is heads, then agility and responsiveness is tails. Here are a couple of slides that I was going to touch on, that actually weren't in the report. We did a survey in March and in December. By December, I think the significant negativity that we saw back in March had been tempered a little bit. For many companies, the impact on the supply chain wasn't quite as dramatic as they thought so people who said there would be a significant negative impact declined dramatically in December. But people who said that there would be somewhat of an impact increased. We saw it maybe not quite as bad, but still had an impact. I've been talking a lot over the years about supply chain resiliency. It was both rewarding and frustrating to me that resiliency became the overused term in supply chains last year because I think it's not a jump on the bandwagon thing. It's really important. Here you can see when we asked companies about what were, for the future of their supply chain, the gaps that if they didn't fix, if they continued to have that gap, that would be most problematic. The two were: supply chain resiliency and the digital competencies in the supply chain to pivot the business model if necessary. Those that could reposition themselves as local markets and you have to have the supply chain capabilities to be able to pivot, to do that. I won't linger on this other than to suggest that we saw, in a way, three major impacts on the supply chain. We're talking mostly here today about demand. Certainly, there were supply issues, right Factories closed, tier one supplier problems. Maybe even below tier one. It's very common in the supply chain to discover to your horror that all of your tier one suppliers are dependent on a single tier four. If that tier four goes down, all of your tier one suppliers have a problem. A lot of supply impact, logistics as well. Massive increases last year in e-commerce and direct-to-consumer. Probably much of that will stick, but that means shortage of capacity and shortage of capacity in the classic Keynesian economy, the very means, the rates will climb. So we saw issues around supply logistics, certainly. I've focused mostly on the demand part because that's the one that's the hardest to deal with. Johnston: Thank you, Simon, very much for sharing all those details. I also like the idea of the self-driving supply chain. I think that's a great one. Now we're going to receive the brand view when it comes to demand planning. I'm excited to talk with Ben Hosseinzadeh from Caulipower. Can you get us started with just some quick background on Caulipower? Hosseinzadeh: Sure, Caulipower is a growing CPG brand known for gluten-free pizzas. Really, our mission is to use the power of veggies to bring about a healthier and more convenient version of the foods you crave, that actually taste like the foods you crave. So you can find us down the freezer aisle in every major retailer and I'm very excited to be part of this group. Johnston: We're very excited to talk with you today. When it comes to demand planning, tell us a little bit about some of the pain points you were experiencing. I know this leads you to bring on demand planning technology, in your case, an integrated business planning platform. So we'd like to get some background on that. Hosseinzadeh: We are a high-growth company and as we started to continuously grow rapidly, we found several pain points in connecting the dots between making sure that we had the right real-time information from the shelf, down to the supply side and demand generation. Bringing about accurate forecasts and timely forecasts to be able to make the right decisions. As the company continues to grow, we needed a tool that allowed us to have a little bit more consistency and create a baseline where we can build both long-term and short-term planning from. That's the main reason why we brought on a technology that allowed us to be able to do that. It supports the functions between operation, supply chain, sales, and finance to be able to come together with a set of data that is consistent in allowing us to be able to make the decisions we need to make to be successful. By implementing a tool that allows us to do this, we have decisions that will be made that allow us to have a higher-than-standard fill rate. Both in general, but even during the pandemic, giving us the foresight and the insight to be able to build inventory levels. Not to carry too much or too little, and making sure to get the product to the consumer on the shelf during the highs and the lows, both during panic-buying and when it's stabilized. Then up and down again. The benefit was seen immediately. Johnston: I want to dig into the tech a little bit because I know that's a key part of helping deal with those highs and lows. Shastri, can you give us a quick intro to Unioncrate? Mahadeo: Unioncrate is an integrated business planning platform that unites artificial intelligence with human intelligence so that brands can plan and execute independently of their supply chain strategies by clicking buttons on the platform. Johnston: You've talked about the industry needing a new benchmark for accuracy, what does that look like in today's consumer goods landscape? Mahadeo: The benchmark for accuracy now, depending on the size of the company, is around maybe 40-50%. The reason why that benchmark needs to be reevaluated is because that benchmark completes calculated debates on a brand forecasting at the brand level, or forecasting at the customer level. Not being able to go down to the granularity of where they're shipping to because instead of just forecasting to Walmart, we want to forecast to every individual Walmart's distribution center. When most brands calculate that forecast accuracy, they're not able to go down the granular level. We need to reevaluate how we're calculating accuracy and what level that accuracy is being calculated, because if you're not calculating it based on the individual SKU and tool level, there's a lot of money being left on the table when it comes to logistical costs and so on. What needs to be done a bit more is understanding how the accuracy affects the entire business. Most of the time, accuracy is just talked about from the sales perspective: How am I forecasting my sales? How am I forecasting the inventory? But how that actually affects the fill rates, the shelving cost, finance, marketing ﹘ understanding how accuracy affects the entire business needs to be reevaluated. Johnston: I'm actually going to welcome the whole group so we can talk more about how it affects the whole business. Ben, I know everyone has hope that 2021 is going to be better than 2020, but I think we're realizing that might not necessarily be the case for everyone. What is something that you learned from 2020 that you think you're going to bring into 2021 in order to keep thriving? Hosseinzadeh: More communication, more collaboration, and more often. What I mean by that is typically, we would have a handoff between when a forecast was complete, it would go from marketing to sales to supply chain, and then we would communicate back to our manufacturers to be able to get the process going. What we've learned is that having the collaboration often and early between the cross functional teams allows us to be more nimble, more agile. It gives us the opportunity to make shifts where we need to. Now layer into it the use of the technology that gives us a fantastic baseline to build off of. That interaction, providing the inputs that are required to fill any gaps that the platform has, gives us that success factor. I don't think that's ever going to change moving forward — certainly not for us. Johnston: Shastri, can you talk more about the tech perspective? What do you think brands need to keep doing in order to survive and thrive for 2021? Mahadeo: I think Ben's point and Simon as well, brands need to think about demand planning and supply chain planning as a company-wide activity and a holistic activity. The days of forecasting and silos of sales doing their own forecasts, and marketing doing their own, and the supply teams doing their own has to be gone. Everyone's remote, there's too many different factors at play for everyone to be able to forecast individually. As Simon mentioned, the ability to be agile. In order for a company to be agile, you can't have a forecast of good sales and then wait for a meeting with supply, then wait for a meeting with the manufacturers. You need to be able to have all this in real-time. In order for a company to be successful moving forward, they need to think about supply chain planning as a holistic approach, and realize every department affects the one goal of the company, not forecasting in silos. Johnston: I'd like to hear the perspective from all three of you. If you had to list the No. 1 thing brands should start to do if they aren't already, to compete in the new normal in terms of demand planning, what would that one thing be? Hosseinzadeh: Good question. I think you should spend the time that's required to meet often, even if you can't afford to. It'll save you in droves later on by dedicating that time to be able to collaborate. Ellis: I don't know that my view today is necessarily all that different than it was before 2020. I think that forecasting is one tool for the supply chain. As I mentioned before, the only thing that is universally true about any forecast is that it's going to be wrong. Hopefully it's not really wrong, but it's going to be wrong. So as a business, as a supply chain, if you calibrate too rigidly to a single forecast, when it starts to divert or diverge from that number, that creates problems. For me, more so now than ever before because we are seeing forecast accuracy challenges across many industries, not just brands. This view that forecast is just one part of the whole, delivering again, it's your supply chain obligations. You also need to be able to be responsive. It's a supply chain planning problem, too, but build the capabilities into the supply chain so that when the forecast proves to be wrong — I thought somebody was going to order 1,000 and they ordered 5,000 — okay, that's a good problem to have, but it's still a problem. You still have to be able to find that additional 4,000 units. Do what you can to forecast better, use technology where technology makes sense, but don't only rely on forecast. Focus also on the ability to react when the forecast proves to be less accurate than you might've ideally hoped. Mahadeo: I was going to give one answer, but as Simon spoke, I changed it a little bit. The forecast, even if you're 80% accurate, you're still 20% wrong — you're going to be wrong at some level. One thing that brands need to focus on, or they should, is understanding why it's wrong. Right now, it's very difficult to understand why the forecast is wrong. You may go back and say again, we have good problems like distribution, or we had a manufacturer that ran out of oranges and we couldn't ship out the product on time. But if you understand why the forecast was wrong, why there was a 20% error or a 50% error in the forecast, then you can make changes to fix that. Leveraging technology is yes, a solution. It doesn't have to be something like ours, it could be something as simple as just the AI tool, to help you to see the data holistically. I know you mentioned a self-driving supply chain, I have a little disagreement with it because the self-driving supply chain will say that technology takes over everything, but you can't ever replace human intelligence. You can't replace that completely. You can leverage technology. If you've ever seen Iron Man, you have Jarvis, right? Jarvis assists Iron Man, but he's not taking over everything. It's probably a good analogy that people understand. Leveraging technology to then understand why the forecast was wrong so you can make changes to it, is probably the No. 1 thing I would focus on. Johnston: Let's talk about AI or Jarvis. I want to hear your perspective on AI. Of course, it's important, but what value is it bringing for demand planning in the supply chain? Mahadeo: AI, in a very general sense — how it's been treated in the past couple years, is a very general view of AI. It's very rare that a company will take AI and make it specifically applied to an industry, which is what we tried to do. We've tried to take AI and train it specifically on consumer good trends. That allows us to be a little bit more accurate when we're forecasting for some of these companies. What AI can do is automate all of the tedious tasks across the department. So Ben mentioned being able to communicate between departments, if sales makes a change, it says here's a new distribution. That should automatically trickle down and change the financial forecast, it should automatically change the supply forecast. If supply says, "Hey, we can't get you oranges," then that should automatically change everything going backwards. It takes away the manual work that causes a company to not be agile. It saves them time in between. AI, when it's applied to specific industries, can remove a lot of the time it takes to understand why something's going wrong or why something's going well. AI alone can't solve all of our problems because no one, again, taking COVID as an example, COVID is unpredictable. We would have never anticipated COVID to happen, but we can actually see signs of it. We can see signs of its impact along the way and AI can say, like Jarvis would, "Hey, you're flying into a place and you're going to run out of fuel." But the human can say, "Well, I think I can make it this much further with this level of my foresight." Collaborating with an AI to automate a lot of the tasks that we have will help us to use technology and AI at its full potential. Johnston: Again, I'd like to hear from all three of you. What's your best estimate on a return to more normal forecasting accuracy? Ellis: I mean, what's normal? We've been seeing a decline in forecast accuracy across many industries now for a number of years. It's partly driven by SKU proliferation, partly by declines in consumer brand loyalty, partly by companies like Caulipower. Lots of the big brand companies have been pecked at by lots and lots of small competitors who outsource supply chain capabilities. So you have many more choices available to consumers. I think, and this is purely speculative, I think by the end of 2021 we'll see some general return to pre-pandemic volatility levels. Johnston: Now Shastri, beyond improving forecasting accuracy, what are some of the business benefits of an IBP? Mahadeo: Time savings. That's a huge benefit and being able to be agile. The No. 1 benefit is being agile. Just take this scenario, for example, that Ben mentioned. If sales generates a forecast, he has to email or share that forecast with his different teams — logistics, supply chain, and so on. That can take a week or two, and then you have to get a meeting, they have to get in a room and say, "Okay, why did you do this?" and I approve of that or I disapprove of that. Ben now has to go to finance and say, "Okay, do we have money to buy this?" and then that's another meeting, another set of time. IBP allows you, outside of forecast accuracy, to be very, very agile. You can click a button within the platform and say, "Hey, I'm making this change to the forecast because of this reason." A manager goes in, they get prompted, they can approve that change. If they approve that change, it can affect supply, it can affect finance, or it can be treated as a one time occurrence. But that time in between, those weeks of time to have meetings and waiting for different forecasts and having to change and make errors, you can be more agile and save a ton of time. Johnston: I have a question that's related to this. Do I need to rip out my current ERP system to bring on a demand planning technology? Mahadeo: Not at all. I'll speak specifically about our platform because all of the private firms do it as well. We spent a very, very long time building a very complex data injection structure which allows us to take any unstructured information from an ERP and structure it into a database. Find errors within that data for the client. You don't have to rip out any ERP. Again, I can only speak specifically for us. You can use any ERP, we can take the data, pull it in easily and give you a forecast. That also goes to our core values, which is allowing a company to be agile. We don't want to disrupt your business by implementing a new tool. We don't think you should take a year implementing a tool, we want to do it in three months. So you don't have to rip any ERP. Johnston: I have one that came in for Caulipower or Unioncrate. At what level of sales does it make financial sense to invest in technical tools? Either in demand planning forecasting, overall business planning, at what point do you graduate from spreadsheets or glorified Quick Books to more sophisticated tools? I'd be curious to hear anyone's opinion on this. Hosseinzadeh: I think it really depends on where you budget, where you want to be, and how much you can afford to spend. At any moment, if you can implement a tool to allow you to be able to make better decisions, you take that opportunity. As I mentioned, it saves you the energy, time and money in droves coming down the line. I've been part of startups and hyper-growth companies, the spreadsheets can create a time suck to a point where you're counterproductive with your spend, based on correcting some challenges that come with that. If you do have the opportunity to invest at any point in the phase of your growth, you take that opportunity and do it, in my opinion. Johnston: Shastri, do you want to build on that? Mahadeo: I agree. I think this question comes up just because of a lot of the legacy technology providers that makes it so complicated and so expensive for people to get started at any stage. It's never too early to implement a tool. What we've tried to do internally here is, we work with companies that are doing a million in revenue and up over 10 billion dollars in revenue. So it's scaled all across the entire size of companies. If you're starting to see that growth phase, you should implement something sooner than later so that you can identify opportunities where you can capture cash. Maybe you're holding on to too much inventory that you've been using for sales, or maybe you have an opportunity in a specific channel that you're not able to see because you're on a spreadsheet. Early on, brands may think that they don't want to spend the money for a tool, but by spending that money, they can actually unlock more money. Johnston: Shastri, another question for you. Does this replace an MRP or does it integrate? Mahadeo: It'll integrate. Johnston: Then building on that, what's the preferred data set for demand forecasting at the retail level? Mahadeo: It depends. We have over 100 different models that we use internally. Those models get deployed automatically based on the level of data that the customer provides. At a very core level, there's a portfolio of models that we use if the client only has sales data, inventory data, and promotional data. We are also a Nielsen partner so we take in Nielsen point-of-sale data and other data like IRIs, spins, and so on. If you want to forecast down to that individual retail store level, the consumer or the point-of-sale data is very helpful. However, sometimes, it actually has a lot of noise. There are instances where a brand may provide that level of data, but we don't even use it because it's a little bit noisy and actually reduces the accuracy of the forecast. It is important to have that level of store-level data from a third-party provider, but sometimes it's not really necessary Johnston: Simon, this question says we are in a segment of the toy industry that experienced increased consumer demand in Spring/Summer 2020. The big question going into this Spring was whether customers would plan to anniversary this growth or go back to a more predictable demand curve. What tools could a brand use to reconcile the conservative customer with a more optimistic demand outlook and forecast for this Spring and Summer? Ellis: That's a really interesting one because I've had a number of conversations just like that over the last three or four months. One of the big trends that we saw last year was a shift away from experiential purchases more to product purchases. If you have young children, you couldn't take them to Disney World, you probably didn't take them to Florida, and so you bought toys, or you bought a dog. So the conversations that I've had have been in the area of: How should we interpret the demand increases we saw in Spring and Summer of 2020? Should we assume that's the new baseline? Or should we say, “Hang on a second, that was probably unrealistic demand and we should temper that a little bit.” Most of the folks I've been talking to suggest the latter, that 2020 in that regard probably was an anomaly, and it makes sense to temper that back. Again, you've got to be careful because if you forecast that based on 2020 and 2019, you're flexible enough if it impacts you. You will see that greater demand than you thought, that you can be resilient to that. My guess is then maybe some comments around sort of the role that integrated the plan per se. An AI tool, at least helps you to be more nimble. The demand manifests itself in a way that you didn't expect. Hosseinzadeh: Do you mind if I take some part of that question as well? I think what tools can a brand use to reconcile that conservative. Again, the opposite. I think it's what Simon said at some point, that you need a tool to look at what your pattern was pre-COVID or pre-when-these-issues-happened. Give some sort of outlook on that, but the ability, a tool that would essentially allow you to affect that without having to do crazy amounts of modeling, is what you need to do. Again, I'll keep saying that AI is amazing. It can give you great forecasts and optimistic demand outlook (or even lack thereof), but human intelligence is what you all are hearing in the market from your buyers and what you're doing in terms of seeing the effects of promotions. You can then balance out what the forecasting tool is actually giving you. A tool that allows you to do that would be a little bit more helpful to straighten, let them anchor for you. One thing I'd like to add is when the pandemic first began, we all thought what were we pre-COVID, and what are we post-COVID? As the timeline to get back to normal keeps extending out, what we have to understand is that the consumer buying behavior is going to change permanently. Looking back at history, then trying to look forward and use history to be able to make those decisions, may not be all that effective. We need that AI tool to help us understand the consumer buying behavior so that we can, as we're talking about in this webinar, demand plan in the new normal. There's a couple different factors to take into consideration as we look at how to fill that gap. Johnston: Shastri, we have one about utilizing AI, to what degree of accuracy increase should we expect or anticipate? I wonder if this relates to your previous point about the human touch. I was wondering if you could take this one. Mahadeo: It goes to the earlier point of what level are you forecasting for? When we work with brands, we sometimes see brands that are just forecasting at the customer-level. They're forecasting for Walmart as a whole, or Whole Foods as a whole. Then obviously, their accuracy is going to be a lot better than their forecasting to a ship-to level or a ship-from-information. With our clients, we've seen going to ship-to level accuracies upwards of 75-80% when you're going through the SKU at a ship-to level. I'd like to note in some instances, we have seen even 91% accuracy when we're going to SKU-, to ship-to level. Comfortably, I would say in the 70s, upwards to 80s at the SKU to ship-to. That's what we hope to expect. Most brands will not see that on their own without leveraging some sort of technology or AI to do that. Accuracy can always be a lot higher when you're forecasting at the total company level because you're looking at a lot less data points. Most companies are doing it that way. The level of accuracy can be dependent on the granularity of industry. What AI allows you to do is go to that most granular level. For example, we forecast demand at the SKU ship-to and ship-from level, and then build it up. So we're building a true bottom-up forecast. We're trying to get the highest level of accuracy, 80% plus at the SKU- and ship-to level. Then we're consolidating that up to the customer, then we're rolling that up to the brand, and they're rolling that up to the total company. That's the kind of level of accuracy and granularity that we try to achieve using AI at Unioncrate. Ellis: Shastri has made the point a couple of times that the technology has to be there to support people, not replace people. That's really important because AI could work really well, but it has to learn from the data. So the first couple of times that it does something, it may not do it particularly well because it doesn't have enough learned from the data. That is where the people come in. It's also very important to note, because there's been lots and lots of studies over the years that suggest if you mess around with the forecast too much, you end up with worse or not better performance. The analogy I always use is the guy in traffic who's constantly changing lanes, and then he ends up behind you, not in front of you. It's important to let the forecast do what it does. Then use the technology to say, "Okay. Now how do I affect those other levers?" How do I think about the interim? How do I think about how the factory would have to be? Mahadeo: 100%. I know we're running out of time. Just one other thing because I feel very passionate about that. We see clients now that they're so used to this legacy technology, and there's some competitors in the market that do this, that make you take a modeling class to use their software. Why am I taking a modeling class? What the industry expects based on the legacy software is, you have to go in and see what demand factor is going to affect your brand. I'll build your forecast using certain demand factors. They play around with the demand factors to build a forecast, but why? Let the AI do what it has to do, then like you said, affect it by other things. For example, if an AI predicts you're going to sell 1,000 units — let’s say it's a Walmart. What the AI won't know is that you then had a phone call with Walmart, and they said you're going from 100 doors to 200 doors. We're not going to know that unless you start recording people's conversations for the AI to learn, and then we are very far into that. We're already factored into, we're predicting 1,000 units because of what you did previously and the velocity in the stores, the promotions you're doing within Walmart, so we predicted 1,000. But then we threw a rock in this by calling the buyer and saying, "Hey, it moves from 1,000, from 100 to 200." So all you would need to do, and what we want clients to do, is just go in and say, "Oh, I'm going to call this higher because I had a call with the buyer." Now the AI will then learn that okay, we have a distribution gain. It's going to increase for that. That is a much better and more effective process than someone taking a modeling class and going in and saying, "Oh, I think the promotion will affect it by 10%, or I think this velocity has affected by 40%." You're essentially just guessing. Let the AI do its job. You affect it by turning one or two levers, and then you would get it right. Johnston: We have time for one more. Shastri, I'm just going to ask you if you have any best practices that you could leave our audience with. For those who are looking to get started or really make progress on their journey to improve their demand planning. Mahadeo: I think the No. 1 thing is before you even start using the tool or technology — again, be very familiar with your data, make sure it's organized and clean. I know that seems like a very impossible task and that's when people say ... that's why we exist, we help with that. But I think very, very small things like making sure you track SKU transitions, or making sure you track what customers, or to what warehouses. Those very little things can actually help eliminate a lot of errors in your Excel sheet forecast. Obviously, a step above that is implementing a tool that can identify the errors and why they're being caused. One thing I would say, don't underestimate how important it is to have clean, or some level of, structured data internally. Johnston: Okay, great. Well, thank you. I'm afraid that's all the time we have for today. I'd like to thank all of our speakers for giving us their time and subject matter expertise today. Finally, thank you for giving us your time today. We hope you found it worthwhile. Have a great rest of your day.

Related Topics