Measuring Assortment Impact With AI: Strategy to Execution

Speaker A: Hello everyone and welcome to the CMA and CIMA webinar series. I'm uh, Mike Wilkening, part of the content team here at ahrq and I will be your host for today's webinar Measuring Assortment Impact with AI Strategy to Execution. Now there's been a lot of member interest in this topic, so we're thrilled to host today's discussion and AI as we know as Revolution Retail and done well it simplifies the complex decision making process for merchants. Now in this interactive discussion we're going to explore how AI elevates strategic assortment and category planning decision making. And we will showcase AI's capacity to transform extensive data sets into actionable lower risk execution plans, ensuring process integrity and transparency and bolstering confidence in every on shelf decision. Now, a little housekeeping before we start today. We won't record today's session and post it in the resource library, but if you have further interest in the topic, our experts contact information will be up at the end of the presentation and us here at the CMA too will be happy to connect you as desired. As well, we will have some time for Q and A at the end, so if you've got some questions, please enter those in the questions function at Wright. Now it's my pleasure today to introduce our uh, panelists. Brian Ruhak is CPG Sales Team Lead at highfree. He's an adept retail consumer goods strategist who sharpened his skills as an insights consultant at IRI where he leveraged point of sales and panel data to inform retail strategy for enterprise CPG manufacturers. Now, his MBA background and IRI experience underscore his role at Hivory where he spearheads AI integration and category management, drives innovation, enhances product placement and delivers actionable insights for optimized retail performance and growth. I'm m also pleased to welcome Cole Decker who's product manager of Curate and User Experience at Hyvery. He's got a lot of experience in the industry tech realm including time at 8451 where he excelled at leveraging data analytics to personalize customer interactions, thereby driving category growth. Today at Highbury, he utilizes this acumen to tailor AI driven assortment solutions, delivering store specific strategies that maximize retail space, productivity and SKU rationalization. His work supports clients in navigating the complexities of Catman with data driven precision and strategic agility. Welcome again to our panelists and I will hand it to them to get started. Brian Cole, the floor is yours.

Speaker B: Excellent. Thank you Michael. And thank you to CMA for having Us here today to speak with all of you guys. I know this topic is a big topic, uh, for everyone. Um, we're all hearing it in all of our roles. So excited to share a little bit about um, what it means to us and what it means to Hivory and then take you on a journey from really strategy to execution. Um, Cole, if we can jump to the agenda real quick, um, I'll take us through kind of how we're going to approach our session today. First we're going to tell you a little bit about Hybree. Um, I'll give you a brief overview of who we are and what we do. Uh, for those of you who don't know us. Um, then we'll dive into the role of retail, um, and AI in retail decision making. M and then quickly move over to how do we do that? How do we take that AI and advanced simulation and make strategic uh, decisions from that and then go from simulation to execution in our final part. And then we'll make sure to leave time for Q and A, um, for the group and feel free to ask any question um, that's pertinent to your role um, or something that seems interesting to what we speak through today. So as promised, I'll do a brief overview of Hybri who Hivory is is. We are an assortment optimization and simulation platform that utilizes AI and ML to drive assortment insights using bottoms up, um, a bottoms up approach, which means from the store specific sales data. Um, we were uh, founded back in 2014, uh, in Sydney, Australia where our headquarters are today, but in a very different space. Um, out of the Coca Cola founders program and out of one of uh, Australia's leading data science companies. We developed an algorithm uh, for assortment in vending machines. Uh, fast forward a few years. We brought that same technology to the retail space, um, and had a lot of success which catapulted our launch into 2020 when we really hit our go to market stride. Um, and now have the opportunity to speak to you guys today on how we are utilizing AI from taking it from a strategy to an execution. Um, Cole, if you advance one slide further, that'd be great. So let me tell you a little bit more about how we do this and then we'll dive into that AI in retail. Um, what makes us different at Hybree is we use a bottoms up AI model that uses store specific sales data to uncover white space opportunities that were previously never attainable due to the complexities of store level assortment or the manpower required to get there. Our AI model uncovers and allows us to rapidly quantify the results that, you know, strategies that might take weeks, um, historically down to a few minutes. Um, we'll touch on AI's ability to do this at scale in the next few slides. We um, are also space and merchandising awareness, um, in our assortment strategy, uh, which allows our results to be truly executable, um, at shelf, um, for both manufacturers and retail teams, which ultimately increases collaboration between both you as a retailer or you as a manufacturer, uh, together to get what's right, um, for both parties. At the end of the day we believe that data has a better idea and that's why we utilize the most granular level of data to drive our insights. Um, and we'll take you through that journey here today. So Cole, if you could advance to the next slide here please. Thank you. So as promised, we're going to talk a little bit about the role of uh, AI in retail, uh, decision making. Now that you know a little bit about highfree, um, Cole, if you could click through. I know there's quite a few follow ups here. So let's explore today, you know, this transformation of how businesses operate across all, all industries utilizing AI, focusing particularly on our space, which is retail. Um, this role of AI and ML is enhancing and expanding our expertise. And it's not just a trend. It's a fundamental shift in our approach to problem solving and decision making. Let's start by looking at the sheer volume and capacity of AI. Unlike humans, we have a fixed number of neurons, um, a machine, its capacity is unlimited virtually. So this enables AI to analyze vast amounts of data, um, that historically have never been able to be processed, um, and that data continues to grow exponentially in volume, velocity, variety. Ah, and these are what we are dealing with today when we're making decisions on what assortment should look like, you know, what products you know, go in the right stores or what, you know, what our planograms look like every single day in category management. Um, so really AI allows us to have that unlimited capacity and make sense of that data and translate it into something that's truly executable. Now AI also allows us to have speed and agility. Decision, um, that once took humans months, uh, to analyze can now be accomplished in just minutes. I mentioned that on the previous slide and Cole will show you how we do that at Hybree today and how you might apply that to your category management strategies. Um, furthermore, AI, it's fostering new creative thought. It's learning from past actions and outcome um, that allows us to predict what the best things are in the future, um, and really impact our business. For previously solutions that were unimaginable, um, AI's role is really just augmenting human expertise and allowing us to open new doors and operational efficiencies, innovation and ultimately creating greater success in any industry, but particularly for us today, retail and Catman. So let's talk a little bit about that further, Cole and dive into how AI is already impacting us, uh, every single day in retail. I really look at, it's, it's impacting our entire industry. I look at really five key areas though that I like to think about it impacting. First is supply chain optimization. This includes everything from forecasting demand to optimizing inventory levels to predicting transportation delays or identifying bottlenecks in the supply chain before they even happen. Um, it allows us to respond quickly and make decisions preemptively, um, which helps the overall supply chain. Secondly is a personalized customer experience. Providing personalized recommendations and offers to individual consumers, um, improving their overall shopping experience. Um, include product recommendations or targeted marketing campaigns or personalized pricings. I, um, know I'm hit with this every single day when I jump on social media. I am a, you know, I get suckered in every single day by buying something that I see on Instagram or Facebook. Uh, it's unbelievable. Um, the other is, you know, automation of mundane tasks really, and we'll talk about this further, but handling routine customer inquiries, data processing, managing inventory, um, those type of activities. Um, it also improves our product development. You know, it helps companies design products that are more likely to be successful in a certain market. Um, it can include everything from the most popular flavors or scents or getting the right ingredients or packaging of a particular product. And then finally, um, I like to talk about this as customer pattern recommendation or recognition. But I also think utilizing AR and VR technologies, retailers are creating interactive displays that allow customers to visualize product in different settings and configuration, which is super powerful for us in the retail space today. If we could jump to the next slide. So let's talk a little bit further about the process and where we are historically. Historically with. Where we are with the amounts of data that we have. Humans and our teams in particular in category management and retail have spent a lot of precious time on mundane tasks of data collection of analyzing that particular data manually and then planning for the future of, um, you know, off of that analysis. Where we haven't spent a tremendous amount of time is what. Where humans do incredibly well is the strategy and implementation piece. Um, that's due to the amount of time spent on the, you know, aforementioned information there. Um, there are reasons why one is, you know, people constraint. Right. You can't throw all those people at it. As much as we'd like to, we have, you know, headcount constraints on our team. Also time. There's only so much time in the day. I know during relay season a lot of us give up a ton of time to focus on our particular relays or resets. Wouldn't we like to have some of that back? And then two money. Um, obviously we all are constrained by our budgets and bringing tools and new analysis to the team. Um, historically that's been our challenge and why we can only focus so much of our precious time on strategy and implementation. However, let's look to the future and what AI and ML allows us to do. Cole, if we could jump to the next slide. AI and ML in our space allows us to spend more time at what matters and what humans do best, utilizing our critical thinking and intuition to formulate strategies and implement those strategies. At shelf, in the sense of category management, that means less time analyzing large massive data sets, defining the ideal assortment, as well as less time reviewing and drawing planograms. Uh, you gain more time to develop executable strategies that can actually grow your business and evaluating those prior to sending them to shelf. Um, I think it's important, I believe that one of my team members defines AI quite well and how we are using it in category management. We're not creating any new data at all. We're just providing you with more power, firepower to get the best out of the data that you have at your fingertips today. So Cole, if we could jump to the next slide. So for those of you on the call, for the merchants of category managers, let's talk about a few of the benefits that we have, um, and where AI can help you. First and foremost, AI allows you to enhance your decision making project process. I mentioned this before. Um, however, AI allows us to balance our time allocation to focus on human what humans do very, very well, um, and utilizes that predictive modeling to inform our decisions in the future. Cole will take you through how we utilize some of that predictive modeling at hybrid and how we influence assortment through that. The second is optimizing assortment planning. Um, historically we only have time to really dive into one strategy and then take it and run with it for that relay or reset because we are time bound or we're people bound or we're money back right today, AI can now Allow us to evaluate various assortment strategies which allow us to truly define what's best for us as a brand or us as a retailer. Next is improved forecasting accuracy. I believe that humans are fundamentally not great at forecasting and I'm in sales, so I'm going to hand up, uh, not a strong suit for a lot of us. So utilizing predictive modeling to enhance our capability of our forecasting and be able to consider and compute diverse sets of data is incredibly significant. I started my career at a large live plants, uh, manufacturer and forecasting for us was imperative as we had to plan crops 18 months in advance because of yield times. Having ML now would significantly have impacted our production as well as eliminated the need to, you know, overgrow over produce, uh, getting that right amount of, you know, right amount of goods to the store. I'm going to ask a rhetorical question on the next one. I know all of you are on mute and I know this is a tough one, but who on this call truly loves drawing planograms? Maybe a few, but for like the masses, for hundreds, sometimes thousands of planograms during a reset or relay. It's hard, it's time consuming, but we have to do it. However, tools in our space today have significantly reduced the time spent, um, on the draw. And Cole will jump further into how we do that on our platform and what we're doing at Highbreed. But capabilities continue to drive innovation in that space in our industry. Um, last but certainly not least, and a benefit to all of us in Catman and in retail are the ability to have real time insights and respond to those insights. Gone are the days of submitting a report, leaving our desks at 5pm so that we can come back in the morning and open that report and then analyze the data. We're talking about insights in seconds and minutes, um, that historically would have taken us days. Cool. One more slide. Uh, here, um, for me is I want to talk to you guys now about how Hybrid is doing it, how Hybrid is contributing to this AI revolution in retail. I won't spend a tremendous amount of time here because Cole will be showing you how we do this in real time in our platform. However, Hybrid utilizes retail specific assortment, uh, insights and our ML model utilizes the most granular data by store and by UPC to run our engine, which allows us to have never before seen white space opportunities. Uh, we also use interactive store level planograms which take us down to that most granular approach and allows us to see KPIs as we change things in real time. Three seconds. And then those two things allow us to have effective decision making and collaboration with our retailer. I had a client that states like this is fundamentally changing how we as a manufacturer and our retailers are working together, being able to evaluate different strategies and truly figuring out what's best for us. Um, and then of course, because we are using that predictive modeling in an ML model, we're able to see revenue impacts of changes in real time. Um, so we know what's, what's best or sometimes worse for our categories. Um, which is super important to know. Cole, if you could advance one more. And I lied, I said I had one more slide. But, um, now this is my final slide before I kick you over to Cole and he can take you through the fun stuff. But, um, you're probably thinking that's all great. Um, those are some fancy buzz words, but how does this actually translate to a retailer? And shelf. Um, before I kick it over to Cole, I'll briefly share results we received at a major grocery retailer. Um, we parallel path, ah, their existing assortment process and implemented it at store for the tea category. What the results were, that is, was our bottoms up assortment approach outperformed their current process by about 3% in dollar growth for a total of nearly 8% of sales growth within that category at that retailer. We did that process in about two weeks versus the 12 week process that that current retailer has. Um, that's about 17% of the time. Um, we were quite proud of those results and are looking to expand on that across different categories within the same retailer. But it all shows that what really is the power of AI at shelf and at retail. Um, Cole, I'm going to kick it over to you to show the group how they can do this, um, in real time through a traditional business case.

Speaker C: Sure. Thanks, Brian. Um, hi everybody. As I was introduced, my name is Cole. I've been with Hybree for coming up on two years now. I just recently transitioned roles from product manager to sales engineer. So I spent the last year and a half, uh, working with the engineering team to build the product that I'm going to show you today. And now it's my new job to sell the product. Um, that said though, uh, this is not a sales pitch. Uh, I just want to talk to you about AI because I get really nerdy about this stuff. So I'm going to take you through an example. Um, for this example though, I think it's probably important for us to level set on how this works tactically. How do I actually use AI as a merchant, as a category manager, as a representative of a supplier who's in sales and attempting to get additional distribution in for my items. How does this work? So when we start working with a client, the first thing we do is we identify typically a category and retailer combination to start with. Right. We don't want to sign a big uh, contract without proving out the secret sauce. So we do that first. And what we'll typically do is we'll identify that category, retailer combo and then we'll work with you to gather all the information that you have on that category and that retailer. So this could be um, stored level, detail, um, which items are in which stores, assortment information. Right. So looking at all the items that are currently assorted, um, and all of their attributes, um, we will also uh, ingest a CDT or need states if you have one, um, to help us identify uh, the way that we need to train our demand transfer model that looks at dollar shifting between items, um, more to come on that and we will also uh, ingest the movement information for those products. Um, we are doing a pilot uh, with one of our clients right now, looking at uh, some third party, uh, kind of rest of market data, um, to do a project. But typically our sweet spot is working right in where we've got um, all the revenue information. Now one common thing that you'll see through this presentation is that the outputs are as good as the inputs. And that's not just this AI, that is everything. Right? If you go to ChatGPT and you give it bad inputs, you can't necessarily expect a super high fidelity output. And the same is true here. So it's our job as stewards of the science to make sure that we're getting relevant results based on the information that we're feeding in. So when we start, uh, we collect all this information as well as planograms, PSA files if you have them. If you don't have them, uh, we can do ah, an assortment only project and um, I'll kind of point out the differences there. Um, but we'll ingest all this information. And what do we do with it then? Well we feed it to a model that uh, will help us come up with predictions. So last setup slide I have before we really get into it, how does Curate work? So Curate is a product that we have at Hybree and the prediction sciences that Curate is built on are actually patented uh, by our research founders. Um, so what we first do is we analyze your historical performance so looking at all the information that you've provided, details on stores, details on items, information on where items are currently selling, information on where items are available to sell.

Speaker A: Right.

Speaker C: If you can't assort a certain item into a certain store, uh, we may not want to have that as an option for you. And then what we do is we essentially create a giant table in the cloud. This is the easiest way to think about it. And where we have our, in our columns are our stores and in our rows are all of our products. So we've got a combination for each store and items, and where we have observed behavior, so where items are currently selling in stores, we then go fill in all of those cells and we're left with what will resemble a Sudoku board. Right? You get it. There's a few things filled in. There's a lot of things that still need some work. What we'll then do is we will look at all of the interactions of, um, products and stores that exist, and then we'll look at similarity of products to other products and stores to other stores. And, and we're essentially able to triangulate a prediction for how any given item will perform in any other store. So once we have these predictions for every item store combination, we can then use this to identify the best assortment at each store. And what we'll do is we'll then identify that assortment. We'll filter out pods or items that are unassorted in any given store. We'll apply that demand transfer model to adjust our revenue predictions. And then we will use that information, uh, to summarize at various levels of aggregation. So let's get into this example. We are looking at the, um, carbonated soft drinks category, um, in a major retailer. Side note, this is demo data. So this is not live customer data. Um, it has been mocked up for demonstration purposes. So the first thing that we'll do is we have collected all that information, we've uploaded it into the tool, and we essentially have all of our predictions. The very first thing we'll do is we will run a baseline analysis. And what this does is says based on all the current points of distribution I have, if I didn't change any of those, what are my current projections for how this assortment would perform this year? So we ingest up, uh, to 104 weeks worth of data on the front end, and then we'll use that information to create predictions or projections for this year. This becomes our baseline. If we don't change anything, these are the results we expect to see. And I've just cherry picked a couple of metrics that we'll use throughout this example. Um, we're looking at category revenue per week and category points of distribution. And uh, I'm pretending that I'm a category manager at major retailer and I'm really interested in private label also because in the back of my mind I've got the private label team who's very interested in what my strategy is going to be and how their items will play into it. You could do this with a subcategory, you could do this with brands, series of brands, uh, kind of any, any way you want to split it, you can, but we're just using these, uh, for this example. So right now I can see that I'll do about $1.07 million in revenue for this category per week, uh, if I don't change anything. And about 66,000 of that is private label. And in this total category I've got 10,600 points of distribution with private label being 1,000 of them. Call it. So we can already kind of see that private label, the relationship there is a bit interesting, right? Private labels got a lot more points of distribution, um, compared to the rate of uh, the sales that it's capturing. So that's just something for me to take note of. We might be over assorted in our private label and I might want to use that art to apply to the science here to find a different result. The next thing we'll do is we'll see what's the total potential. So if we let the model just throw every item, every point of distribution on the floor and reassort it in every single store, as a store, uh, specific, what would this actually recommend and how would that assortment perform in aggregate? We can see that if we did that, we would actually see about a 7% increase in category revenue per week over what we expected for our baseline. If we don't change anything, this is the total upside right here. The total upside for this category at this retailer is about 7%. And again, this is. If I go to implement everything that the model recommends that I do, my private label revenue per week is actually up 22.9%. So we're really showing some strong growth there in private label, uh, points of distribution. You can see though there's a mismatch here. I have to increase 34% on my points of distribution in order to achieve a 7% growth in sales. This is not feasible, right? As a retailer, I'm not going to chase down the long tail of 34%, uh, increase in points of distribution just to get that 7%. So I'm going to put on my business cap and say, all right, science, I know a little bit more context than you do in this category. And what I'm actually being told is that I need to reduce the items that I have on the shelf because, uh, my retailer is really focused on making sure that we are always in stock. And the way that they want to go about doing that is reducing the points of distribution. There's no way that I'm going to grow my points of distribution by 35%. If that's the case now, we can start layering in different strategies. I can tell the optimization engine that I want to reduce my points of distribution, uh, for this total project by 10%. And I can do it in a couple of different ways. The first way I can have it reduce my points of distribution by 10% at each store. Or I could say, you know what, across the board, let's look at a 10% reduction in points of distribution. And when I simulate these strategies, which by the way, take a handful of minutes to simulate, I can then see that I actually, it's a better result for me if I look at implementing this as a 10% reduction across all stores rather than taking the bottom 10% points of distribution at each store. Uh, it's not a ton of difference, but it's $100,000 a week. So we're going to take this option. So I'm essentially going to save this strategy and I'm now going to add on another business strategy that I have. Remember, back to the private label example, this category carbonated soft drink, is heavily brand driven. It is not really in my strategy to increase private label points of distribution, which if you see here, Even with a 10% reduction in total points of distribution, our engine is actually recommending that I grow their points of distribution private label by 25%. That's not something that I've negotiated with the private label team. I'd like to hold their points of distribution flat. So what I'm going to do is I'm going to add in then another strategy to then say, okay, I want my private label points of distribution to remain flat and I've got a choice to make here. I can either say it has to be the same items that are currently on shelf at those stores, or I can say, you know what engine, you can choose whatever items that are private label you want to put into certain stores based on how you expect them to perform. But it has to be the same number of points of distribution, right? So what we do here is we essentially keep the points of distribution flat, but we allow churn within private label. So if you can find a better alternative for those private label items here, then I'm going to let you do it. So I have decided, I've run the scenario again a few more minutes and I have decided that uh, this is the choice I'm going to make. I'm going to keep those pods flat, but I'm going to enable the engine to make better soyman choices. This actually in turn increases my private label revenue compared to what my base would have been. The next steps here are to review the impact at different levels of aggregation. Review the store specific assortments and pogs and manage those assortments or I'm sorry, and massage those assortments and pogs in the interactive editor. So at this point in time, I'm going to swipe over here and we're going to go on a little demo. Here's my project, my uh, CSD project. And that last strategy I was looking at was actually this one right here. Here's my strategy. I've got uh, configurable KPIs so I can choose which KPIs are important to me, which ones I want to see if I wanted to view this again. We can view it by, um, we can view it by individual manufacturer, we can view it by brand. There are different aggregations that we have here on this piece. And for the individual stores we can also start to see what are the individual store assortments that were included. Remember this project is space aware, so the recommendations that we're giving actually do fit onto the shelf. That 34% increase in points of distribution, those actually fit on the shelf because in this instance there's actually a lot of dual placement of items for different shelves within this category within the same store. So here I can see all of my items. But I might be to this point where I'm like, you know what, I really like the assortment that I've identified. It gets me to the results that I'm looking for. But I really have to work out changing, moving these items around. Right? The planograms are not where I need them to be and I've got options I can export into different space management software of choice and start moving things around. But then you start making some assortment changes and then you no longer know exactly how your items are going to perform. So instead of doing that, what we would actually do is we would open These results up in an interactive scenario, which is what I'm showing you here. So in this scenario we can actually see all of our products again with all of their global KPIs which are configurable. Um, we can see any given product, uh, its product details across the enterprise and then we can also see which stores it's been added or removed from. Right. So points of distribution from the private label example may be flat, but we anticipate quite a bit of churn because we're actually making different choices for those stores, um, where those private label label items are going to go. Conversely, if I wanted to look at a specific store, I can actually view the planogram, uh, in this format and I can see across the top all of my key performance indicators. Again configurable. These are now store level KPIs. On the right hand side we've got a product list that will show me all of my assorted products in the store and their performance as well as any unassorted items that are available in the store. We uploaded an availability file which helps us to know which items are available in which stores. Based on that, we've actually filtered this list. I can see here, uh, based on the incremental revenue. If I'm looking for an item to bring in to fill a certain space, I can actually use this KPI which is the same KPI that our optimization engine used to make an assortment decision. I can actually bring this in, I can place it right here, I can place it right on the planogram and uh, it will then update all my KPIs. My uh, KPIs will update in real time. Then you can see that my revenue per week metrics are now recalculating. This is because we're calling the demand transfer model. And it's now looking at that new assortment with that big red can pack on it. And it's saying now that that is on the assortment. How do I anticipate these items to be incremental moving forward? Right, so the thing is, not only are we you know, massaging our planograms, massaging our assortments either for a single sort or for a cluster, um, here. But we're also being able to track the key KPIs and see those changing and we're also able to then have an updated in real time look at expected incremental KPIs for when we do need to make a swap decision. Now we also have the ability to make bulk changes. You can add a product to a series of stores where it performs well in, or remove a product from a series of stores where it does not perform well in. If you get to this point in the execution phase, um, we also have a couple more features that are coming out here between, uh, now and July. Right. So now the end of June. Um, these are all things that are in the works with our engineering team, but I'll speak to them briefly. So the first is assortment strategy, which is essentially thinking about those points of distribution changes that we were discussing on the slides. And it is really visualized in a slider type of, Type of interface where I can choose a brand, Coca Cola, for example, drop the slider to a certain point and say, I need it based on contractual obligations. I need them to have this many points of distribution. We can lock that in and we can let all of the other sliders figure out where they need to go to round out the assortment. That's most incremental. Um, the other thing that we've got coming out, it's a little bit shorter term here in the next couple of weeks is Recommender. Um, this is a little pane that will live right up here at the top of your product list, both in the single store and in group store, bulk store. Um, this will show you at any given time what is the best item to add to this assortment, or what is the best item to remove from this assortment based on the level of incrementality. So the example, let's say there's a late ad, we got this new innovation item. We expect it to perform well, but there's an agreement that's signed that says it needs to go into 50 stores. How do I know which 50 stores to put that item into? And then once I make those store selections, how do I know which items to take out of those 50 stores that are going to be the least impactful to my overall revenue in those stores? Right. So what we can do is actually use recommender to say which are the 50 stores in this project where this item is anticipated to be most incremental. And then based on the assortments of those individual 50 stores, what is the item that we should pull out in every single one of those stores to make space for this item? Uh, and then the other thing that's coming is smart placement. So smart placement really helps us kind of round out the POG automation piece. So here you can see we're looking at an individual store or a cluster planogram and making massages at that point in time. If I were to add an item to the store, it would actually show up here on the floating shelf at the top of the screen and then I could place it down into the planogram. But we're actually piloting science right now that uses nearest neighbors methodology to say what is any given item next to in the groups of stores that it's currently assorted in. And based on that, recommend a placement for that item in this planogram. And this is really kind of the beginning to the end of us being able to kind of wrap a bow and uh, put the 2 label on, uh, the planogram automation piece. So, thinking back to AI and why AI is so important, it's really important because it is essentially revolutionizing the way that we will actually work here. In this example, the final slide we have before we open it up for questions. The idea is that you, as a, as a buyer, as someone in sales, you're always thinking about what is the strategy that I need to employ and what are actually the results that I need to get to. And what we can do is enable you to focus on that and let science figure out the path to get there. Right. As long as you have a good idea of where you want to go and what your strategies are, that's the art piece of this. Science can kind of fill in the gaps for you along the way. So when we talk about strategy to execution, we really talk about the journey from at a high level, identifying what is it that we want to do with this category, what are the things we want to implement here, and then letting science take care of the nitty gritty store item, uh, assortments, and then making any massages that you need to in the app where you can see the KPIs, uh, changing in real time. And then you can export those to PSA files and then port them right back into the space management, uh, software of your choice. Um, with that, I'm going to turn it back over to Brian Ruhawk. He's going to go over some key takeaways and then we will address some questions.

Speaker B: Thank you so much, Paul, for that journey and how we go from strategy to execution. As Cole mentioned, I'll leave with a few takeaways and hopefully there's something here in the last 41 minutes that you'll be able to take back to your respective desk or your respective team and help to either optimize the process, create more efficiency, or ultimately grow sales. Um, what I believe is really fundamentally a strategy is only as good as its execution plan. Um, and the way which you translate that strategy to the shelf matters. So through utilizing AI and ML to eliminate or reduce the time spent on those mundane tasks so that you can focus on that very, very important piece of strategy, um, is so, so powerful no matter how you look at it or what tools or platforms you're using. Um, second, and I think Cole took you through this journey on how we built that strategy. Right? He put that arm to the science. Only something a buyer would know when you know, uh, it looked to be, you know, private label over skewed in that category and making sure we take a look at that constraint. What I mean by this is prioritizing those constraints or your ultimate end goal is just as important as setting them right. It's a combination of the two. Um, but you need to make sure that you're prioritizing those right goals or constraints ahead of time so it can impact the future. And then finally the last takeaway we'll give you is bringing that predictive modeling downstream and into the executionary process. Gives you that scientific view of what historically or typically has been regarded as art, um, and ultimately drives that performance accountability and gives us confidence in what we are delivering where we say yeah, that feels good. But also we now have the science, um, and the engine to back it up. Um, and also doing so very quickly, um, thanks to the ML models. With that, ah, we're going to pause and open it up for the last 10 or so 15 so minutes for a Q and A session, um, for the group.

Speaker A: Cole, Brian, this was, this was awesome. A, ah, ton of questions have come in um, and so want to be respectful, everybody's time. We might not be able to get to all of them but want to get through a couple at least. And there was one around how the system works with panel data, more thought leadership focused data. And we know there's a lot of data. Our members have to sort of wade through it, sort of making the right decisions on recommendations itself and those things. Can you speak a little bit to how the system sort of intakes that and how it can work with that?

Speaker B: Absolutely. Um, Cole, I'll tackle and you fill in the, the blind spots and hopefully I can give some examples to this one. This is a super common question uh, Mike, that we often get um, when we're in process but we actually don't currently ingest any panel data. I um, mentioned this before because we are a bottoms up assortment methodology. We believe that the best predictor of future uh, behavior is the past purchases. So what we're able to do by actually seeing buying patterns and what's purchased at a store, we're inherently uncovering um, panel demographic data, um, specific household data through actual purchases that are ending up being better predictors of what's going to happen in the future. For example, um, Cole and I, uh, fall into, you know, probably we're close in the same age, we both live in Ohio, we both spent time in Columbus, we both have golden retrievers. Historically, panel data, we might fall into that same demographic. However, Kohl's purchasing might be completely different from me, um, due to I have a three year old, so a lot of my purchases are at 3 year old buying kids food, whereas Kohl's are not. Um, so we use that past behavior and actually what is being purchased at the store to drive um, the future protector of success.

Speaker A: Certainly good insights there and a good description of again the process, the bottoms up process. So we're actually at time because we've got 45 minutes here and uh, we'll be sort of wrapping up here shortly. But uh, again, wanted to commend you for the sort of outreach, uh, and sort of overview of this tool Recommender. Sounds really, really interesting. Um, again this idea of the incrementality question and uh, it seems like there's a good deal of automation and insight that can kind of make those decisions easier for people and I know that those can be tricky. So, uh, excited to kind of see what's coming here with the tool. Um, and thank you so much Brian and Cole for your expertise and for walking us through it to our attendees. Uh, again, uh, if we didn't get to your questions, and I get a lot of them and I want to make sure that we can get, give them all the proper amount of attention. We'll make sure that our team uh, gets those out and makes sure that our panelists can uh, can address them and uh, can kind of give you some of the information you need uh, to proceed forward to kind of uh, get those questions answered. Attendees, thank you so much for coming. Uh, we'll be back next week. We've got a busy, busy couple webinar weeks coming up. The presentation company with Data Driven Storytelling next week. So that's always a huge attended one, so be on the lookout for that. And again, Cole, Brian, friends at Hivory, thanks so much uh, for bringing this forth today and uh, talking about a topic again that there's so much heat and interest around. Everyone, have a great remainder of your day and thank you so much for your support of the cma. Bye bye.

Speaker B: Thanks Mike. Thanks, everybody. Goodbye. Thank you.