Webinar Transcripts

AI-Driven Category Management & Merchandising

AI-Driven Category Management Webinar Recap.

In this session, industry leaders from Lumi AI, AgFunder, and top retail and supply chain organizations explore how AI is reshaping category management and merchandising. The conversation centers on how data-driven tools are helping teams respond more quickly to market shifts, optimize pricing and assortment, and improve agility across decision-making processes.

Key speakers include Trap Yates (Managing Director & Partner, BCG), Gary Saarenvirta (former CEO, Daisy Intelligence), and Ibrahim Ashqar (CEO & Co-Founder, Lumi AI). Together, they offer real-world perspectives on navigating shrinking margins, rising complexity, and evolving shopper behaviors using AI-powered insights.

Whether you're leading a merchandising function or supporting it with data, this discussion highlights practical strategies for moving beyond static planning to unlock faster, smarter, and more adaptive category management.

Webinar Transcript

Louisa Burwood-Taylor:
Great, might just wait for a few more attendees to join. Okay, I'll give it one more minute. all right. Well, maybe I'll start to introduce the session we're having today. Thank you everyone so much for joining. My name is Louisa Boa Taylor, and I am the head of media and research at Agfunder, and the managing editor of Agfunder News, and I'll be co-hosting this session today about AI driven category management and merchandising along with Ibrahim Ashkar, the co-founder, and CEO of Lumi AI, so we're really looking forward to digging into how AI is transforming the way that retailers and Cpgs make decisions about assortment, pricing and promotions. And if you're anything like me, you're wondering. You know what is AI, and how is it going to change my change my life and change my work? So you know, we really want to share some kind of meaningful insights into how AI is reshaping category management and merchandising. and we have a couple of expert guests with us who can really talk through some real examples, how we can use these advances to make better decisions and create better results, and ultimately hope you'll discover opportunities to apply these ideas in your own organization, and we can also talk a bit about how Lumi AI could help. So we have an hour. We're going to have a short poll quickly. In a few minutes we're going to have some time for Q. And A at the end, but most of it will be a panel discussion with our 2 great guests. Please do feel free to submit any questions in the Q. And A. Function at any time. I'll be keeping an eye on that throughout, and we can. We'll be saving time at the end to answer those so to kick things off. We have a poll, Ibra, if you could if you could launch that let us know how far along is your organization in using AI for category management and merchandising. And while you fill that out. I want to introduce you to our 2 guests. So we have Trap Gates, who is the managing director and a partner in consumer and retail at Boston consulting group he has spent over a decade working with retailers across North America to embed data, driven AI powered capabilities in the merchandising and operations. And he brings a broad perspective on market trends and an executive view, basically on what retail leaders are trying to achieve with AI and what challenges they face. So thank you for joining us chat. And we also have Kari Carrie. Sorry. I just practiced that earlier. I got it wrong already. A deep AI practitioner with 30 years of experience in machine learning for retail, and you are the founder and CEO of Daisy Intelligence, a company known for using AI and pricing and promotions that was acquired last year after delivering over 5% top line sales, gains for clients like Walmart and Car 4. So, as you can see we are in very safe hands, and Ibra. I don't know if you would like to share a little bit about yourself as well.

Ibrahim Ashqar:
Yeah, for sure. Thanks, as always, Louisa, for the for the Intro Ibrahim Ashkar, CEO, and co-founder of Lumi AI. We're an enterprise analytics platform focused on helping brands and retailers find value hidden in their sales. Individual procurement data sets. We're backed by investors from Silicon Valley from New York and overseas. Prior to founding Lumi, I spent my entire career in data analytics. I was a director of Data science at a Supply Chain Technology Fulfillment company called stored, and before that I was in consulting within Deloitte's artificial Intelligence Practice building enterprise, great data products for fortune. 500 type companies. Majority of my work was in retail. And Cpg, so that's kind of been my bread and butter, and really excited to be here. I think we have a really exciting agenda on the docket, so excited to be continuing.

Gary Saarenvirta:
Some minor correction on ex CEO of the Daisy. I'm no longer intelligence after they required, but the former CEO.

Louisa Burwood-Taylor:
Former founder and former CEO. Great. Thank you for that, Ibra. I can't actually see the poll results. I don't know if you can, and you could share share what those look like. There we go. Okay. Great, yeah. I could see them now, fantastic. So hopefully, everyone else can see this on their on their screen. So it looks like nearly half of you are just exploring AI which is is not surprising. 20% of you have pilot projects underway. 20% have implemented it in a few areas. And 13% of you are aficionados widely adopted across the business. Super interesting. Thank you so much for filling that out. So if we jump to the next slide, Ibra. talk a little bit about some of the the. The main pain points in category management. There's a way to kind of set the context. And we're using some research from traps firm here. So as many as you, many of you know, deciding what goes onto the shelf and at what price and what quantity has historically been slow and manual. So you've had long planning cycles, cycles, siloed spreadsheets, and decisions that can often struggle to keep up with fast changing consumer behavior, and even best in class teams can struggle with very scattered data and time consuming analysis. And that can mean missed opportunities and slower reaction times on trends. So, as you can see from some of this, these data from Bcg merchants can spend about 40% of their time on manual tasks. 69% of merchandising leaders believe that automation through AI will unlock more time. Estimated savings from using AI about 10 h per week. So yeah, some really interesting data here from Bcg, about the potential and so I'm going to jump to this 1st question trap. I think I'll come to you first.st If AI is so great, why isn't everyone doing it yet?

Trap Yates:
Yeah. good afternoon, everyone, or morning or evening, or wherever you're dialing in from. It's a it's a great question, Louisa. The approach we usually apply to the problem of AI implementation is what we call our 1020, 70 framework. And those stand for the percent of the effort that needs to go in to unlock value from AI where 10% is your algorithms, 20% is your data and technology platform and 70% is your people processes, organizational ways of working. Now, you need all 3. You can't have AI without an algorithm. And if you don't have the right data, it's not going to work. But fundamentally, what we find is that if you don't get right the ways of working question, you can't unlock value from AI, and that's where a lot of organizations get stuck. So they'll have a pilot program on AI in a corner. But it's not embedded in how merchant actually works day to day, so it collects, it collects dust on the shelf, and so certainly, where I spend the vast majority of my time actually was 30 seconds late to this call from a client meeting on exactly the subject of AI in in merchandising is helping them think through. Well, how will you actually embed these capabilities that you want for your organization in the actual way that people are working day to day. And doesn't make it sort of side side of desk. So that that's our primary finding on on why this is stuck. Folks have a lot of the tools floating around. The capability to to use more advanced technology is there. But the way to crack it into into your day to day ways of working is is where folks are are struggling the most. This is what I'm seeing.

Ibrahim Ashqar:
Yeah. 1 thing also, I took that was really interesting from from the survey trap that you guys, you guys did. Was kind of around the the estimated time savings. And around that 10 h per week. And that's great, you know. Obviously, time savings is is a fantastic thing. But I would say, it's it's it's not just about time savings. I think there's an aspect of making better decisions. You know this this kind of this future of the merchant of the future, or whatever where kind of they're using some of these more advanced tools. Obviously, barring some of the adoption issues you mentioned about people process technology, but kind of being able to kind of leverage some of these tools to make really more advanced and better decision making on pricing promotions, assortment, and things like that, but curious to get your take on that as well.

Trap Yates:
Yeah, it's I was having this debate with a client last week, right? Which was okay, we want to implement a more agentic automation in their merchandising flow, and that'll take hours out of the out of the week. Fabulous. There's a cost associated to that great. That number is going to pale in comparison every single time to the value of adding one and a half or 2% to the top line of the business. Just is like the physics of retail are such that if you can unlock that incremental growth or another 100 bips of gross margin, it just will automatically be worth way more than your your head office productivity gains because your Sg. And a line fundamentally like that's not where the physics of your your P. And L. Are really sitting. And so I do think you're bang on now, getting the balance right helps because some of those efficiency gains can sort of fund the journey while you prove out the growth opportunity. But certainly we see that. And it sounds like you know, Gary, from your experience as well. You aimed at that, too. Like, if you can hit the big nuggets at the top of the of the P. And L. That is going to be 1020 x. More valuable than the below. The line.

Gary Saarenvirta:
Yeah, we were always going at the top line and and net margin right? So I think any any top line increment you get with AI doesn't really add to the overhead of a business. So it falls straight to the bottom line. So we were seeing, you know, in companies that really adopted was almost a doubling of total company net income, which they never realized, because they usually reinvest that back in price to keep winning share. But yeah, I think one of the big challenges is a lot of these vendor analytics. Tools historically, have coupled workflow into the tool as well. And that's that exacerbates the change challenge that you spoke about trap that you know vendors analytic vendors, you know. trying to come up with a theoretical workflow, and that the change leap is too huge, and I think one of the exciting things in this future of AI is the uncoupling of workflow from access to the algorithmic intelligence. And so either via agents or apis, you can build, you can integrate the algorithmic output into your existing process workflow, because I would say, most retailers in general. their process isn't broken might not be the most efficient, might not, you know, might be slightly better ways to do it, but in fact, they are running the business today, so to force them to change the process, I think that could be should be a secondary consideration. Once you figure out how to integrate this intelligence. And I do agree with you. Trap, in my experience, you know, for every dollar you spend on technology, you could spend $10 on the process compensation job role definition, all of those things. And so I think a lot of companies are afraid to take the leap because those 3 things are a very scary change, you know. It becomes a almost a board level decision to embark on the journey. And most companies, the analytics and AI is a side desk project, you know, either unopened shrink, wrap, or a parallel process beside the core process, which eventually withers and dies so.

Louisa Burwood-Taylor:
Hmm! Well, I mean, maybe we could just take a bit of a step back and talk a bit about you know what are what are the biggest challenges in category management and merchandising today that, you know, make this a really good time to be considering and adopting AI because obviously, you know, there is this theme, you said, there's a big challenge to make that change. But what are some of the biggest challenges. Gary? Perhaps we can. We can start with you for that one.

Gary Saarenvirta:
Yeah, I think it's cost pressure supply chain costs are not going down. If you've been to the grocery store lately, you can see how the prices of everything is skyrocketing, you know, and and I think most geographies are seeing this. And so, you know, increased pressure on cost, I think. you know, is is trying to get. We want to drive down the operational cost of retail so we can stay competitive, right? I think that, I think you know, started with Covid, and you know, rejigging of supply chains, and once prices go up, they're never coming down again. Right? So I think running profitable retail is more and more difficult. Retail has always been a difficult business to generate really large net profits. So I think this increased cost pressure. A lot of the new safety protocols. All of that has stayed in place, I think the rise of e-commerce is, you know, means, you know, some bricks and mortar is losing share. You know the smaller retail players, I mean. I see acquisitions and mergers every few weeks, you know, on my on my text alerts about retail. So I think those are some of the challenges and retail is trying to drive down cost of operations and and Sg. And a costs right? So I think AI is A is a great way to do that, to realize the cost savings and time savings. But, as Trap said more importantly, to get the, you know, sales and and net margin growth, to to be able to stay competitive.

Ibrahim Ashqar:
Is there something about the fact that there's like more data than ever? Now. I'm curious to get your thoughts on that.

Trap Yates:
I might twist it slightly, which is to say, I think that to Gary your point the the existential threat has never been higher. I'll say it that way, especially for retailers that are not Walmart or Amazon. No offense to anyone on the call from any particular organization. and the second and this is where I would tweak your framing is, I think, retailers have been awash in data for quite quite a while. It's it's just sort of a big data problem definitionally. But what what is new is the ability to access, make sense of and do something smart with that data? If you kind of know how to repoint your organization, that opportunity cost has never been lower. Right sort of tools and organizations, frankly, like Lumi, are allowing you to to actually turn that data into something. So it's not just a giant brick sitting on a server somewhere that no one knows how to use except the it guy who, like put it there 10 years ago. It's actually like an asset, and can be treated as such and accessed as such, and so at least as I see it. It's it's yes. The existence of data. In some areas, personalization e-commerce does generate more. But the fundamental notion of having millions of customers going through a till generating millions of basket records like that. That's not new. It's it's the ability to actually do something with that information in a relatively more direct way than hiring an army of bespoke data. Scientists to live in. Your org to me is the is something that has changed quite rapidly in the last couple of years.

Gary Saarenvirta:
Yeah, I think the cost of computing has really been the the driving force behind this AI explosion or seeming explosion the last several years. But I remember 30 years ago I got into retail working with retailers as a tech guy, because I love terabytes of grocery keylog data. So there's been. you know, hundreds of terabytes of grocery T log data and the largest retail customers, and that was excited me because it was a technical challenge that the retailers couldn't overcome to try to make sense of that. So now that the cost of computing and the cost and the proliferation of algorithms is out there. I think you know, the you know, time is great to take advantage of this internal data, and then I'd say 90. In my opinion, 85, 90% of the value comes from your own internal already existing data. And then you can augment that with external facts. But you know, if you're gonna optimize a business. It's data about the business or from the business that's going to have the largest effect, and that has never truly been taken advantage of. But the opportunity really exists today with the continuing plummeting of costs that we are seeing in this compute and data center data management team.

Louisa Burwood-Taylor:
Okay, so clearly, you, you're agreeing that the complexity is growing and the current processes are not necessarily keeping up is there anything else to add around. You know why this is the time to be considering this, or, if not, we can move on to some use cases. Awesome. Great. Okay? So this is always, I think, the exciting part, because often, you know, I feel like when you're speaking about AI being used. It can be very theoretical in many cases, like really understanding tangible examples is the bit that is exciting, at least from my side. So let's talk a bit about you know where you have seen AI making the biggest impact in category management and merchandising today and talk about some compelling use cases. Trap. Perhaps we could start with you. I think you've got, you know, a few great use cases that you've worked on with some of your clients around Jedi.

Trap Yates:
Yeah, we've my, my team and Bcg, broadly have done ad deployments all over. The world with reading, leading retailers all over the place. Personally, my expertise has been in a couple of domains and merge pricing and promotions which I know. Gary, you'll you'll talk quite a bit about localization and space optimization in brick and mortar. Actually, and Jenna is particularly useful there for attributing and naming all your product data. Very exciting. If you're a product data nerd like myself. And then the last is sort of the Gen. AI processes for merging and a merchant. So what does your, what does your future merchant actually look like? How are they going to do their job when they are doing it in conjunction with sort of agents? Helping to run down certain tasks. And so I think actually, if you kind of go through the entire life cycle of merchandising from supplier intake product setup, finding things in the market, pricing that getting it to shelf like you. There's an AI use case at every step of the journey. And those price promo and and assortment optimization ones are the ones that in the last couple of years I've spent most of my time deploying with with my merchandising clients.

Ibrahim Ashqar:
Trap in a previous convo we had. There was one thing that really stuck out or resonated with me, and it was like, I think he. you're talking about Gen. AI is being using, helping classify product recommendations or product descriptions and things like that. And that's like a really good way to, you know. Help inform, maybe features for more traditional Ml, models. That do things. And maybe this is a good point. Kind of distinguish the difference between Gen. AI use cases and traditional like machine learning. AI statistical based approaches. But yeah, that was something that really kind of resonate. Obviously. you know. And and to quote you like, there's all these like engines for for pricing for Promo, for for, you know, shelf space things like that. And then now there's like that Gen. AI, that's enabling newer capabilities. So those those are things that really resonated from from our conversation. And think last week or so.

Trap Yates:
Yeah, happy to elaborate on that, and then and then Gary here from your perspective as well. So the specific instance you're describing. Just to give give folks on the call here, context is as part of a sort of space optimization effort. We needed to create need states of items. So items that are fulfilling the same kind of role in the shopping basket for customers, and the more traditional approach to AI, let's say, was excellent at finding those groups of similar items based on the relative levels of penetration, and and how they were appearing in baskets together. But it was really bad at naming them like, what is this? It sort of found the the mathematical artifact. Let's say that, hey? These 2 products are getting purchased together all the time. There's a relationship here. But is it for a child's birthday party? Someone's buying a candle and a cake? Is it for a barbecue? Is it for back to school. And if you look at it as the user, you can obviously classify it. But doing that across thousands of need states evolving every year across a whole bunch of different banners. It's not a practical solution to that problem. And so we found that in cases like that layering on. Now, this new layer of gen AI capability that can actually go and sort of do some of the research online. Okay, what does this look like, how are people talking about these products? Hey? This looks like, that's what this is. Is it getting it right? 100% of the time? No. But is it taking you from all of these things? Have sort of alpha numerical codes for names to 90% of them have names that make sense. Yes, and and it's a huge leap forward in the interpretability of that output which then enables the change management to flow much easier, because, instead of having a conversation with a category manager on, hey? Your need state 1, 2, 7 x is really productive. It's hey. This birthday need state is one you need to. You need to pay a lot more attention to. So I'm finding it's really helpful for that. Sort of layer between the so deep mathematics of of more traditional AI and the humans that are driving the change like it's really helping to bridge the bridge between the 2.

Gary Saarenvirta:
Yeah, I would agree, Matthew, one of the big barriers to really implementing algorithmic AI is that master data management and getting the product hierarchy like. So I haven't met a retailer who has a nice. It's nice and analytically useful product, hierarchy, the product hierarchies are built more for organization as an analysis. And so we always, you know, we created like a 5 level hierarchy. There's the skew or Upc at the bottom. Then there's what we call item groups, products like, you know. Let's say, yogurt. You have blueberry, strawberry vanilla in the same brand, same packaging that should all be a product group all at the same price. But then, how do you name it? So we used to come up with these ridiculous name that had dollar 59 Danon yogurt, and then the merchants would go. What the heck is that right? And so I think the Gen. AI helping name. It will smooth that change or change challenge right? I think, because the adoption was, I'm not going to use this master data group unless I understand what it means. And naming them was always a nightmare. So we had all these cryptic names of prices in it, and it was very, very hard to get merchants to adopt that. So I see that as a great role of of Gen. AI, and then having a good product. Hierarchy, then that really supports, you know, help the deep mathematics work even better because you want a nice balanced hierarchy that that makes sense. And I think we always we always struggled with that, and spent a lot of time with every new customer getting that hierarchy right? You know, our focus was always on on selecting, recommending the right items to promote, which sounds like a promotional thing, but it's really total store, because we optimize the halo. Our goal is to say, a customer who buys ground beef if they see ground beef on sale they go. Oh, I'm going to make a pasta dinner, so I'm going to buy pasta tomato sauce, cheese, bread, wine, salad fixings, you know. Nobody eats raw ground beef. Right? So you don't need to promote all the items in the halo. Those you can leave at regular full price, because the ground beef will drive the halo. So the key question was, what items should I promote and what should I not promote? So that's how you get the total store impact. And by promoting items that have a large halo, you know. Contrast, if you promote diapers, diapers has a small halo, you know, there's only one use case for diapers. It's not pretty, but that's all you know, whereas ground beef is in a thousand different recipes. So if you promote items that have a large halo have high penetration. Then you're going to drive more transactions with larger baskets, and if you and if you don't promote the products in the halo, you'll also have a higher margin baskets without even doing any price optimization. So we saw 80% of the benefit from just getting that that effect of what to promote, what? Not to promote identifying the high halo items. because what you find is that the halo ratio, the ratio of items in the halo versus the specific item you're looking at that is constant over years, right? Because the recipe for Pasta Bolognese hasn't changed in 100 years. Whether it's Covid or not, Covid. Whether you promote ground beef or not, the recipe stays the same, and people need to buy the same constituent items. So you can really take advantage of that, because if you get a 50% lift in the item, you'll also get a 50% lift in the halo. And some items like off the top of my head. You know, bananas has typically a 7 to 10 to one halo. So for every dollar of bananas you're buying 7 to $10 of other items, because bananas is part of the weekly shop use case. So we spent a lot of time optimizing that. And we saw 3 to 5% total company sales growth, if we achieved, overcame the change management challenge doing that. And then we got into promotional price optimization, saying, Well, I can. I can. I can influence the halo effect. I can double the sales of my halo item, if I play with the price even more, if it's an elastic product, and that would give me even more bump in the full price. Halo sales, and we got into regular pricing. you know. Hi! What should I price my regular assortment at and forecasting, you know, forecasting promotional and regular item demand. And so our big thing was using reinforcement learning. I think my belief was, you know, one of the challenges with algorithms and machine learning is that there's a different algorithm for every task. So I believe in the value chain of retail used to have one. We invented a mathematical theory where the same theory and the same map applies to item price forecasting assortment space planning. You're using the same underlying mathematics that way. All the decisions are pulling the rope in the same direction as opposed to 2 algorithms that are actually competing against each other, and the benefits offset each other. I felt that was super important. So a number of our patents were there. So yeah, we saw real real results. limited only by the change challenges that we talked about earlier. And I think that's where, ultimately, where my new thinking is around, how to uncouple the change and the and the algorithmic deep math recommendations. And I think that's where I think reinforcement learning, I think, is my belief where the real future of the deep math decision making. I think, Gen. AI and is, you know, around these kind of tech. the user interface having users interact with the deep math, using an interface like Lumi. I think that that's a great as opposed to the traditional point and click through an ugly gui that you need to train somebody to use. But yeah, the benefits it worked in every single retailer I ever went to. I failed every time when it did fail to change management. Right? So.

Ibrahim Ashqar:
Yeah, that's that's super interesting, Gary, and and you know, I think you're talking about the interfaces being, or Gen. AI being a great interface as well. So I know Trap was mentioning Jenny. Kind of helping in the classification, labeling aspect. This is another one. What's like the the user interface. And here's actually an example that we can actually pull. Let me just pull this into into the screen here where it is like, actually, can kind of show you an example that we've done with one of our clients. I'll just make this full screen. And yeah, you know, it's basically a case study of of, you know, the Lumi interface and kind of an experience from one of our clients, a big retailer, actually. And the the idea was like one of the key stakeholders there in the customer and merchandising team, was curious to know, like what would happen if we reduced prices slightly across our assortment right? And and this was something we kind of put lumi to the test in, so to speak. So this was a great use case so initially, you know this is the 1st question. Lumi obviously is a chat interface for your analytics. It's on top of your your data, your reporting layer. If you have it connected to one of these engines. It'd be fantastic and kind of surfacing even more granular insights. But, you know, start off with like Hey lumi. I'm looking to optimize pricing by analyzing elasticity. Can you kind of recommend an approach, so to speak. And this was super interesting. Lumi kind of gives us a recommendation on, on how he wants to do the analysis. We went kind of back and forth on it. A couple of times we had a conversation. But in the end it kind of got us something super interesting, right? We're able to get down to a list of 38 items. And obviously, you know, there's like hundreds of thousands of items that that you know it's analyzing here. But it really found the subset of items where, like a small change in price, a 5% reduction, if not more, had a profound impact on the quantity sold in this case, you know, like you know, in some cases over 5% reduction had over a hundred percent increase in quantity sold, which is super interesting. And let me kind of summarize the insights there. So, instead of kind of manually, you know, spending a few days running this model yourself, let me kind of handled all of that process, so to speak, generated the code on the fly. You can kind of see that over here and kind of generated some interesting insights. Now, of course we're not. We're not kind of claiming that price alone is what caused the the volume change. You know, it's it's a correlation, not causation. There's obviously other factors into play here. Seasonality promotions may have influenced demand, but what we're trying to showcase is kind of that directional insights. Skews were like. There were some interesting price changes consistently aligning with volume changes and things like that. And this helped kind of the stakeholders there. In this case, the category managers, you know, an informed short list of skews to potentially dig deeper into or to test, you know, like, Hey, what do we need to discount more aggressively? Which ones to hold firm on? Where do we need to shift inventory based on expected lift, so on and so forth. So this was just an interesting use case. I thought it'd be interesting for the audience to see on, on how Lumi has been kind of directly applied in more category management type use cases. But I'll kind of go back to the to the screen here, I know I've been talking for some time.

Gary Saarenvirta:
And I think I I think if you know, if you have the mathematical repository created by, you know, reinforcement learning, or some or other algorithm than having Lumi or an interface sit on top of that to allow the merchants to query the intelligence that's created by algorithms. I think that's valuable because I think merchants and category managers aren't analysts and looking directly at raw, analytical information has always been a challenge. You know, how do you present information so that it can be used by the users. As I said earlier, the traditional point and click workflow is a very painful way that merchants don't want to spend the time to learn, but putting a natural language interface, letting them query that deep math output, and making it much, easily, much more easier understood and presentable. I think that's you know. Some people will like the old fashioned Gui, but I think a lot of people will will, you know, default to this kind of natural language interface. And I think that's a real. I think Lumi is a great technology for that. Not to give you a plugg. But I mean, I know I I see that as a use case for some of my clients that I'm working with today where we create the deep math, and then Lumi interfaces to the deep math output. You know.

Trap Yates:
And I would just extend the logic like one half step further from what? From what I'm seeing, which is, I agree with all of that. I think the other thing that that the merchandising teams are gonna have to get their heads around as they move into this future is the traditional decision structure around how to promote related items in the basket as you're describing, Gary is is historically fragmented along those divisional and category lines. So your beef category manager reports into a totally different arm. Then your pasta category manager, one's probably into the fresh art team, one's probably into the grocery team. They only converge at like a chief merchant or something like that. And so, as these insights get surfaced that say, Hey, I'm I'm pinging against the deep math, and I'm learning about these halo relationships. If you are viewing that information through a report that is designed for your category, Silo, you're actually missing the entire point of the of the exercise, and then the decisions that are flowing from that are going to remain in their silos. And so I think, as we put these interfaces on top of that, that deep math Lumi and others as much value as sort of getting out of that Gui trap, I think, is in helping organizations move faster to that less siloed like decisioning structure, because otherwise you're just continuing to maintain these islands of analytics that are making unrelated decisions. And you're losing the point of the annoying project.

Gary Saarenvirta:
Yeah, I would agree. I mean, absolutely, though I'd say, you know, category management is turned sideways. Right? Because if you're running a category, not all categories will grow because some categories compete against each other. And so this some of the smart retailers that I've worked with started to create category portfolios where they grouped categories that kind of were complementary, or they set up or they gave category managers a negative target, knowing that I can't grow that category, because it'll affect a different category that I want to grow. So I think when the halo effect turns the category management paradigm sideways, and that you're looking, you know, one category has an impact on a completely unrelated. And so that's part of the change challenge is compensation. So now you're traditionally compensating category managers on their category alone, their gross margin, their revenue alone, whereas that's not necessarily optimal for the total store impact. And that was one of the things I we definitely ran into, which was, you know, I say, the smartest retailers not to not to denigrate anybody. But you know the head merchants say. you know. Hey, Mr. Beef Guy, if you give up some ad space, take a margin hit for the good of the company, and we'll show you, and then, and we'll back your bonus up, even even if you don't hit the numbers we originally gave you, because it's better for the company to do that. And then, you know the you know, the beef, guy would say, sure, just prove it to me that that actually does happen. And then I'm happy to do that. So it takes some real forward thinking around compensation. And and and I think it's category management 2.0. It's like getting away from that siloed mentality, because that that is definitely missing the huge opportunity of the halo ring.

Louisa Burwood-Taylor:
So great. You guys have kind of naturally taken us on to talk about some of these barriers and and how to overcome them. I mean, what? What other barriers are you seeing? You know you've mentioned change management several times you talked a bit about potential for new training, for merchandises, and so on. Are there? Are there any other particular challenges and barriers to adoption and and ways you think they can be? They can be overcome.

Gary Saarenvirta:
I think it's the mindset of merchants right? I think you know. I always ask retail executives, let's say, you know, are your merchants 10 years from now gonna do the same job that they're doing today. And you know, the 25 year old veterans that you have, that have done it the way to got the business to where it is fantastic, successful, awesome. But is that gonna continue? And a decade from now, are you? Gonna is your business gonna look like it does today. And if that is the case, then how are you training the new people coming in? Are you training the same way you trained the guys 20 years ago who came in. And when are you going to get on the AI roadmap? Because if you don't start, you won't ever see that future come, and I would think Amazon is a great target for bricks and mortar resources. Amazon does not have category. Managers and merchants. Right vendors list their products, algorithms automatically score them, and they get delisted. And they have people that sell marketing programs, not to say that that bricks and mortar will go a hundred percent in that direction. But that that is kind of a target that you don't necessarily need, that that category management merchant role as it was historically defined. I don't know what it exactly looks like in the future, but I think envisioning that role of merchant. And then how do I train those people? I think the younger generation, I think, are more analytically and technologically inclined, and they want the help. You know, our best customers were the ones who were analytically inclined, and they saw value in the algorithmic output, and they they struggled through the change and the interfaces we gave them, whereas the you know, 25 year old Diehard experts were a little more difficult to crack that. Not that there's anything wrong with them, you know those people made businesses what they were today. But I see that role and training and envisioning the future. What is our business going to look like? I think that needs to be painted out. And I think that is the comfort level that boards and executive management teams need to really embark on the journey, and I think we haven't talked enough about that as algorithmic vendors who sell this technology. I think that's a big oversight. I think the company like Bcg. Or somebody like that, who gets that template. change, template or roadmap template, correct, will run the market. I think that's the race. Who can come up with that AI implementation, template, and then run the market in all industries, not just retail.

Louisa Burwood-Taylor:
I mean, it's interesting. Here on this slide here we could see the difference between the the merchandisers, and the and the leaders. But you know, even Deloitte report report found that fewer than 30% of companies have their Ceos directly support the AI agenda. So even the leaders are, you know, pretty behind the curve on on, you know. Ensure, you know, promoting AI in their in their organizations. trap! Do you have any follow up on. You know some of these barriers.

Trap Yates:
I might add 2 we've talked a lot about the 70% barriers. Let's say people process kpis, etc. I would say I. And this may be specifically in the Canadian market, but I I do observe a little bit of so like tech paralysis on the 10 and the 20, and especially as you folks are encountering this Gen. AI. Or agentic future, I think it feels like a very destabilized ecosystem in terms of well, this is not just set up my sap erp and run it for 20 years like this is a rapidly evolving technology landscape. And how do I, as an executive place my bets on on what kinds of solutions or technologies or infrastructure are going to be relevant 5 years from now. Given that every 3 months there's some new Gen. AI release. That's sort of fundamentally changing how this thing works. And so I think helping organizations break through some of that paralysis and sort of how do you build yourself? A system that is flexible for the future, but also is unlocking these benefits, like, I see that as the primary issue in the 10 and the 20 sort of technology side of the of the equation. The other thing I would add to exactly to where you're going in this slide. Lisa, and this might just be my my personal issue, which is for entirely possible. I find that part of the disconnect that I see between merchants and leaders, especially on the left hand side. Because everyone basically agrees on the right. Like, we have a lot of data. We spend a lot of time doing manual stuff and upskilling is a bit, miss. Right? 30% say, it's effective. This isn't working. Okay. So we all agree on that. But then there's the big disconnect on. Should this be great in the future, and is it going to be useful? And the the problem that I see there is the leadership of organizations is one sentence deep on AI in a lot of places like they have the one sentence answer as to Why is AI, good. AI does X and will be good for y thing. Okay? The problem is that answer doesn't survive 4 sentences deep. Well, how is that actually going to work. And what am I going to say to my suppliers? And how is that going to change? How I list an item like Oh, I don't know. And that's where the dark green merchants are at is they're like, I'm not at galaxy. Brain agents are gonna run category. They're at. I actually have to get stuff to shelf this week, because if I don't, we can't sell anything. And so what we spend a lot of time with on our executive clients in particular is trying to help educate them through those sentences between level one and level 4, like, what are you actually saying when you say you want AI in your organization? What, practically speaking, do you want people to do? Was it the actual target outcome here? And I think once you've sort of. I find, as you force people into that, and then you let them play around in it in a bit like, Hey, try one of these technologies. You get to the level of specificity, you need to actually drive the change versus, I think, what exists as a big disconnect in a lot of organizations between. I read the white papers, and I'm doing all the work. And I think that gap is wider now than it was 10 years ago. because the technology is starting to diverge at an increasing clip. So that's like an education gap. Let's say, in addition to some of the the process pieces. That we've talked about.

Gary Saarenvirta:
Yeah, I think I think I agree. Hundreds or a senior leadership team led initiative is AI, right? Because there's such dramatic implications for the organization that you know, you have to have a fully educated leaders to really drive the. you know, drive this through an organization. You can't do change management from the bottom up. It's got to start at the top, and then you bring the bottom along. And there's the. I definitely agree that the Education gap is getting wider. You know, this is not like implementing pos or self checkouts like those are easier understood. You know how those things work is complicated, but the use case is easily understood. But when you get to Gen. AI or deep mathematics, it gets so complex that it's like, you know, you really need to bring the leaders along, and then they need to decide that. Yes, we want to do this because the organizational change implications are so so large, you know.

Ibrahim Ashqar:
Yeah, yeah, I mean, this was such an interesting slide. It it just kind of put things in perspective around, you know, not the tech piece. I think the tech from my perspective, it's like, maybe the tech geni technology right now is like, maybe more advanced than people think. And it's just more so like the people. Process adopt an adoption aspect of it. A lot of the time that we spend at Lumia when we work with clients post implementation. Just kind of helping them learn how to use the system, how to ask questions, how to prompt things like that. So you know, I thought that was interesting, you know, especially, you know, if you look at the biggest gap here, 25 basis points between innovation bears are minimal. The disconnect between kind of leadership and what the merchants actually say. And we kind of kind of see that firsthand right? Like all right, we're definitely seeing some friction points with with users on, okay, well, how do we actually use this technology? And and you know, how can we get it ingrained in our workflow? So yeah, I mean spending more time on on, how do you increase? Adoption is something we do a lot. And I think it's just ripe for consulting, actually for for consulting firms to kind of jump in and and really help with that education piece.

Louisa Burwood-Taylor:
Yeah. Well, this would be great to to jump into. You know the best way to to pilot AI and scale it. And perhaps you can also talk about some of your examples, but you know what should teams be mindful of on that proof of concept to broad adoption, pilot, step, pilot, phase. trap. Do you want to stop.

Trap Yates:
Sure. I I would say I found Let's say, let's say like 3 3 things come to mind. One is you have to be equipped to manage through. And usually this comes from sort of an executive mandate. The disillusionment of 1st use sometimes what we call it, which I think is what is driving that big gap there in the left right, which is, someone gets a tool. It's not quite working the way they thought it was like, this is a bit useless. They put it on the shelf. That's the most dangerous moment in this change curve, and you need to be equipped to to power it through. The the second. I would say, is never underestimate the value of a good pilot in market. So I'd say most of the big deployments. We did. We took a handful of stores. We took a handful of categories and we made change champions out of those category managers, promo directors, whatever they go into market using the new tooling, and then they come back to their peers and say, Well, my results are up. X. And then now you have a pull for demand rather than a push, and I would say in many of our initial AI implementation, some years ago we shortchanged this pilot stage. It is as useful, if not more useful. as a change management exercise than any actual algorithmic learnings. You get along the way those are good. But it's actually a change exercise that you're that you're implementing And then the 3rd piece is the Kpis, which I won't belabor, because because Gary talked about it. But one of the longest stages of of one of the implementation programs I did some years ago was rewiring the category targets just took forever the 1st time you go to a bunch of Cms and say you are in the negative growth categories like, well, am I on the B team like, what does this mean? Is this like a signal of my career? It's like, No, like, actually, in some ways you have a harder job, which is a managed decline of part of the business while not falling off a cliff, but but that whole exercise can't be understated as well.

Ibrahim Ashqar:
Yeah, that's that's interesting from from our perspective kind of when we when we work with with clients, the the best way, we always say, kind of start small prove the value it's kind of. We have like a pilot program. so to speak. This is this, is it where it's like all about, start small. Let's help you unearth the value and kind of scale across. It's supposed to be like a a low risk, low effort, way to to kind of get started. The the idea is for, like, I think, trap, you said, it's like less. So like algorithmic enhancements, or in our case, like finding ways to just improve the the accuracy. It's more so as a way to just see, can we ingrain this tool in people's workflows? Is it actually driving some sort of valuable change, like things that we end up doing kind of a thing is like, obviously post the implementation which we which we, you know, kind of take on or work with our clients to do. spend a lot of time on the support and user training aspect of it. And it's like part of the pilot. Evaluation is we send out like a survey an Nps survey like, hey? Has this impacted your workflow? Positively or not, you know. And and like, we kind of collect that qualitative feedback from from the users to to then kind of funnel to the leadership team on like, okay? Well, all right. We we know Lumi's performed, you know. Let's put a number here. 92% accurate on the evaluation when we did commissioning great, you know that's always valuable. But here's what you're here's what the end users. The people who are using technology actually saying about this tool and how to adopt it, and things like that. And so to us, that's been the most successful way, and and kind of Louisa, to answer your question. Like most successful way to scale kind of some of these initiatives across the organization. Start small. Have, like a clear use case in mind, so to speak, pick a few users put that in their hands and ask them for kind of feedback, and then make a decision on it. From there.

Louisa Burwood-Taylor:
Right. Well, we we were nearing end here and question time. So do participants do start adding questions to the Q. And a. If you'd like our great guests to to answer those before we get there. I have a few more questions. I mean 1 1 thing you know. Can you share, you know, a key lesson learned, or even a mistake made that you have learned from, you know that can be a piece of advice to to audience members who are considering implementing AI.

Gary Saarenvirta:
For us. It's always a trust journey, and you really have to build trust in the merchants and category managers that it's not a black box. They want transparency. And so we always structured our pilots and our implementation to be about building trust. So it's first, st you know, what would AI have done differently than you did last year. So you look at the last year you build the paper business case and you show them that AI very granularly. Here's the different items. Here's the different price points. Look, it's not crazy. It's not a 2 headed hydra. It's just doing things that make sense. And then you definitely have to do the in market thing because the paper business case is meaningless. Those traps and the in market pilot, once you have some trust and they go. Okay, it's not crazy. Then you say you know you let the AI evaluate the merchant created plans before they execute. Give them some guidance to say, Hey, here's some changes you could make. and then the merchants can make some small changes. See the financial benefit before you go to the 3rd step in the journey, which is, let AI recommend what you should do, and if you start on that 3rd step, that's like a holy war of change which we made that mistake in the early days, and, you know, drove a lot of failures by trying to go too far too fast. So you really need to, you know, build that trust because you will only win if you have the trust of the people who have to use the technology every day. So I think that's the way we view the pilots. And yeah, I agree, trap in market results are where you create the poll because you can't push this stuff in. It has to be pulled.

Louisa Burwood-Taylor:
Well, we have a great question from the audience. We also have a poll that we'd love for you guys to fill in. If you want to launch that now. But I will go ahead and ask this question that we have from the audience. If change management must be top down. What foundational AI fluency do the leaders need to actually lead transformation and not just sponsor it? Trap. I think this is probably a good one for you, as you had that great comment around being more than one line knowledge of AI.

Trap Yates:
I do love this question. So I think that executives need fluency on. I know I keep going back to this framework, but I have found it's the most effective, like the critical pieces of each of that 10, the 20, and the 70. And so if the 10 is the algorithms. I don't need them to be able to explain how a neural networks like that's not necessarily gonna happen. What I do need them to be able to explain is what is the clear objective function you're going after. And why is AI a better solution than the alternative which you can do without having to understand all the deep math great. And then finally. like, How how is this thing? Not a black box? Because I agree, Gary, with your point. A 100% cool, I think, in the 20 on the tech. They I probably want more from executives than what is pro plausible. But I want them to be able to explain the critical components of the stack and never say the words. We need the data like, just never speak that you have to understand? What does it actually mean to have business ready data? So you need a, you need a definition of business rate data that withstands like one level of questioning as to what are you talking about? And then on on the 70? I know it's a little bit less this AI fluency question, but I do think you need to understand what are the unique change barriers that come with AI as opposed to other kinds of process change, and I think most of them relate to what what Gary's describing on the kinds of answers you're getting are different than I did a calculation in excel. And now I have an answer. You need to be able to work through these issues of trust, of transparency, of responsible AI of ethics that maybe don't come in with some more traditional models. So maybe more long winded answer than what you wanted. Audience member. But that would be my like if you if you're taking off those 5 or 6 boxes. I think you're ready to have something coherent to say about an AI journey. In my view.

Gary Saarenvirta:
I think it's very interesting that you know. If you're creating a list of items or a list of prices, that's what your deep map creates, you know. Here's the items you should promote. Here's what you should not promote. Here's the prices. Here's the forecast order quantities, and but you already have a list created with with simpler math. Right? And so the merchants go. Well, I have a list. It works because I've been running my business for 20 years. So it's like, why is your list of items better than my list of items? You know? I think, that you know, after you spend all this energy and change at the end of the day, you know. So I think one of the challenges is really not a business problem, that is, it's you're solving it today. It's hard selling. I can do it better, right? And I think that's that's where the executives need to understand that motivation and really be willing to say is my goal to minimize labor, redeploy labor? Am I getting rid of people, or am I redeploying those people to something else? I think they need to be prepared to have an answer for that, because I think one of the big fears with merchants and category managers, and we saw it is like, Am I gonna lose my job if we implement this, you know. So I think the executive team has a has a say in that I think mid market retailers generally they want to redeploy the labor because there's so many other things that people could be doing. But you know, some of the larger retailers maybe want to keep driving, you know. Operational costs down. So you know, I think, having a stance on that, I think. It's also super important for executives right up front and need to communicate that to the organization.

Louisa Burwood-Taylor:
Interesting, Ibra. Could you show us the results of the poll just before we finish.

Louisa Burwood-Taylor:
Oh, great, yeah. Here we go. So most of you. Seeing a lot of opportunity, a lot across a lot of the different categories, automating manual analysis and reporting uncovering hidden sales and inventory trends. Improving product assortment was 71% alongside, optimizing, pricing and promotion. So those were the 2 most popular and enabling non-technical teams to access insights, which is obviously something that Lumi AI can do. Well, thank you all so much for coming and joining this webinar. If you want to continue the conversation with Lumi, or have a demo, then do reach out to them. and they will do some follow up email from this webinar. And lastly, a really big thank you to trap and Gary for joining us today. You know, it's clear that there is a lot of potential here for AI to make some changes, but that the people involved are really, really key, whether that's from the leaders through to the to merchandisers. So it's a really exciting category. And it'd be great to see how this this plays out in the coming weeks and months. And yeah, thank you all so much for joining us.

Social Media
Anthony Scalzitti

Anthony Scalzitti is a Value Engineer at Lumi AI, specializing in implementations and customer success. With a background in computer science and experience in AI strategy and analytics, he ensures Lumi’s solutions deliver real impact for customers.

Lumi AI Connection Graphic for Analytics 101 blog page sidebar

Illuminate Your Path to Discovery with Lumi

Explore Pilot Program

The New Standard for Enterprise Analytics

Make Better, Faster Decisions.

Book A Demo