Promo Forecasting 2.0: How AI Transforms Promo Planning in Grocery Retail | EuroShop 2026

Thank you so much, first of all, for everybody coming over to listen to the pitch. [music] I'm uh I'm very pleased with the turnout, and hopefully this will be really interesting for you, okay? Promotions are becoming more and more part of our retail businesses. And so, today we're going to talk about how we can use AI to effectively plan for those and execute those. Um as uh as I said, I'm Brad Mitsch from the Vice President of North America for Leaf AI, uh which totally makes sense why I'm here in New York speaking to you, but this is uh this is what it is, so thanks for having me. Uh a little bit about our company before we go into the presentation details. Uh we service retailers all over the world, so we've got clients in 40-plus countries um all over the world, 250-plus projects, and we've been in business for 15 years. We work with people of all shapes and sizes, so retailers in the grocery space, the convenience space, um you know, DIY, pet,

pharma, so on and so forth. But really, you know, all shapes and sizes in terms of retailers, lot of CPG and distribution, as well. We do a lot more than promotions. I know that's the focus for today, but you can see here um we do a lot more. So, first we we focus on the assortment for our retailers, right? What should I sell and where should I sell it, right? We help answer this question using AI. Once we know the answer, we move on to our shelf efficiency uh part of our platform, which tells us how should we visually merchandise this in our stores to make sure that we're delighting our customers, okay? So, what should I sell? How should I merchandise it or visually in store? And then finally, we have inventory optimization, which answers the question, how much do I need to purchase, right? Both at the central warehouse and at the retail location level. How much do I actually need to buy in order to uh you know, make the best decisions for my company, have the right inventory at the right place at the right time? And that's what we're going to focus today because when we think about promotions, right? We think about the

uplift that it may give our business, but we also think about the pain of the mechanical execution that has to occur in order for the promotion to take flight. Um so we're going to talk about the challenges there. We're going to talk about how AI can really really improve our business outcome and process uh around uh maintaining promotions. We're going to talk about using Agentyc AI to boost efficiency in our business and then finally we'll talk about how actual customers are using this in day-to-day life. So why are promotions challenging, right? Um I think first and foremost it's because there's more and more of them every year, right? Retailers that I talk to are constantly telling me that um you know, we're doing more of this. It's becoming a bigger part of our company strategy and our suppliers as well are pushing these promotions on us. So it's like um sort of a necessary evil where we as a business, we have to learn and have to mature in the way that we actually uh perform and execute these

promotions. It consumes a tremendous amount of resource, right? It's not enough to find the right promotion at the right price. We also have to update the shelves, update the price tags. We've got to make sure the replenishment is in line with our promotions. Uh so on and so forth and we can't stop, right? We've got this pressure from our customers. It's driving loyalty. We've got pressure from our suppliers and promo becomes a bit of an addiction. The reason uh why it's so hard to manually manage this without the use of artificial intelligence machine learning is because it's very very complex, right? Price elasticity, uh we don't know what it is for our products, right? We need our system We need our systems to be able to look at our data and tell us uh what the answer is. We take on a lot of out of stock risk when we make a promotion, right? We can't say, "Hey, we're going to promote this item and then be out of stock when our customer walks into our store." At the same time, what we typically see in this case is that people over buy, right? And and and over plan stock for promotion such that we actually have out

of stocks at the end or sorry, over stocks at the end. So, it's really really hard to find this balance, um especially when we've got, you know, multiple promotions across multiple categories, large retail footprints, etc. Um overlapping campaigns, right? If you've dealt with this, I I feel very bad for you, but this is very typical for our clients to have many promotions happening at the same time, often even with the same items. Uh fresh and ultra fresh categories, right? It as if this isn't already hard enough to plan, but now I've got to worry about perishability and promotion. So, those overstocks become waste if I don't do this correctly. Um and finally, complex logistic cycles. Again, because of the replenishment times of these promotions, we have to be very precise in the way that we start and end them. So, all that to say, we can help ourselves with AI, right? AI is here to help manage promotions both on the planning and the execution, right? When we think about the planning, we think about can we find the right

price? Can we find the right uh the rules around the promotion, the right time frame, all of this stuff, right? Um can we accurately predict what the promotion will do for our business both in terms of KPIs and inventory using a forecast? But then we also have to go and execute it, right? We have to make sure that we update the different teams, update the store, update the planograms, the price tags, so on and so forth. And AI can help. Let's talk about the planning of the promotion first when we're thinking about forecasting, okay? Statistical forecasting, uh obviously long history of great performance for supply chains and replenishment around the world, but you know, really when we think about the complexity of promotions, it's probably not enough, right? So, machine learning can help because it can actually look at the you know, the behavior from our past promotions and all of the different factors that go into it and make a more accurate forecast. Which is a big part of the puzzle, right? A really big part of the puzzle. Um it can handle much more complex data

patterns um uh longer time frames um so on and so forth. And so, this is really going to give us more powerful forecast, which will ultimately drive our replenishment. Of course, it comes with the higher costs, not just in terms of money, but also data, right? When we think about using machine learning to forecast our promotions, we know that we need more data in order to have the right outcome. If you're a retailer and you're looking at this and thinking, I don't save this data. I don't know what this data is for my business. I don't keep past promotions or promotional sales discount percentages, things like that. I really don't have the right data here. I would start to think about that for your business, right? Should you start keeping that data for the future? Because using tools uh like machine learning uh forecasting algorithms built for promo intelligence, we can take all of the different factors here and really come up with a much more accurate and predictable business model around each of the promotions that we run. So,

obviously type of promotion the price, very critical, right? Critically, we got to understand the price and the elasticity of the price, the impact that it has on sales. Um sales before, during, and after the the promotion, right? What does it do to other items? What does it cannibalize? All of this information and data is really critical to create and execute an accurate promotion, but we need to start keeping it today. Cuz if we go to a client today and we they say, "We don't have it." We can start keeping it and over time we can build a more accurate machine learning algorithm, but um you know, we're going to always going to have earlier success when we have the data. So, think about that for your business. Um, machine learning will do the job of understanding what's really important and and forecasting accurately promotions that you have in your business. Typically with statistical forecasting, we have to um understand and know what the factors and how they're going to influence our model, but machine learning can look at the the model, the history of your

performance of promotions, and actually uh you know, itself improve the accuracy of the forecast by finding the most important factors for your own business. Um, it can do incredible things, right? When we're looking at a at promotions, one of the big challenges that I see with retailers is that we're promoting across hundreds or thousands of items, hundreds of stores, many suppliers, many categories, and it's simply it's too granular to look at the item level and find problems with data. But you again, using a machine learning algorithm, the the actual uh algorithm itself can find where we have gaps in the data, right, that need to be filled. Where we have seasonality, um you know, where where we have holidays and events and things like this and factor that into our promotions going forward. So, it's really really incredible and takes out the manual pain of analyzing these in a massive spreadsheet or a statistical model. We measure forecast accuracy, right? The

first thing we do after we have a forecast and we run a promotion and we use machine learning, we say, "Well, how did we do, right?" So, we go back and look at the accuracy of the of the forecast. And uh there's a number of different ways to look at it. Obviously, we're going to look over the relevant time frame, the relevant aggregation level, whether it's at the company level, the store level, etc. for those items that were on promotion. But mainly we want to look at the um the calculation error, like how far how far were we off? And there's a number of ways to do that. We prefer to look at it using WMAP A, which is the weighted average. And this allows us to put the forecast accuracy in terms of cost, right? Not just were we accurate or not on the unit level, but on a cost basis, on a margin basis, how much how how far off were we in terms of our forecast? And then we can use that information to improve the forecast over time. Um but it's not enough to just look at the forecast accuracy and say whether we were successful or not, right? We should

measure the success of promotions with the KPIs and the outcomes of the promotion, right? Um so we need to look at the sales versus the actual, right? Again, going back to that accuracy. Overstocks. Very, very critical to measure overstocks as part of our success factors for promotions. If we plan a promotion and we had a huge uptick in sale and we have all this inventory to cover that, that's really great, but what's the impact of having a lot of inventory left over after the promotion, right? And if our system isn't handling that for us, it's a manual process, we know that we are leaving something on the table in relation to uh you know, the KPIs and the results of our our promotional planning. Inventory turnover, availability and out of stock, right? We simply cannot have a promotion that has an out of stock or we lose the loyalty and um uh of our of our incoming customer that comes to our stores to shop for these things. So,

it's one thing to accurately, you know, forecast a promotion, right? And order the right amount of inventory and have the replenishment, but if it's if if we're completely 100% accurate in terms of the forecast, it still doesn't mean it's a a success. We've got to go look at these other things and make sure that we executed the the promotion in terms of these other metrics, right? And if we can do all of those things, now we start to get much more out of our promotions. If we do not leverage AI or machine learning, right? And we're doing all this manually, or we're communicating manually, we're doing this in spreadsheets to try and manage these intense promotions that we're running, we're probably leaving a lot on the table in terms of the outcomes, right? So, if they're a necessary evil, and we have to do them, let's do them in the most automated fashion possible using machine learning and AI. Um we can use Agentyc AI to help us, right? Not just to execute the promotion and make sure that we do the right things to

have it on time and right in stock, all that stuff, but also to analyze the results. Um so, in our case, we have a personal assistant that allows us to query the system after a promotion to say, "How were the results? Um you know, what went into the pro- the promotional forecast? Uh what can I do better next time?" etc. We often see that promotions are misplanned or misguided, right? We have a lot of items on promotion that don't need to be on promotion, a lot of items that um we need to be on promotion that aren't on promotion enough, etc. And so, the personal assistant can all help us steer us in that direction so that the next time we have to uh to execute on this, we can improve and do a better job in terms of the parameters that we set across our business. Um a few lessons learned here related to the promotions. Yeah, it's impossible to to have a clear forecast or an accurate forecast without knowing the main aim, but it all comes down to data, right? I

showed you this data uh slide earlier, which shows you all the different factors that machine learning can factor into promotional planning. If we're not tracking this data today, we can't have accurate forecast tomorrow. So, very, very much encourage and encourage you to start tracking data and interviewing that. Um price elasticity is very key and we often don't know what the elasticity is. So, uh in our system we can use uh price scenario or promotion scenarios to look at the projected outcomes of different promotions at different price levels. Right? And uh and this will help us prepare and choose promotions according to our goals as a business. Uh very very important. In terms of process, we have to focus on the financial outcomes, not just the forecast accuracy. We talked about that. Um we have to take every promotion and look at the outcome and decide where we need to change in the future, right? Again, we can plan and execute, but we have to

go back and actually improve. So, these are just some of the lessons learned over, you know, implementing this with clients around the world. We talked about precisely finishing the promotion and having the right stock at the end is just important as having the right stock at the beginning, right? Avoiding those overstocks. Um and and really getting our our our item level replenishment back in line to where it was before we began the promotion. Thanks so much uh again for listening to my presentation. Leafio is located in Hall 5 and booth F38. It's like a straight shot that way. Uh I would say you can't miss it, but you can totally miss it. There's like a million other booths, but if you keep [music] walking this way and ask around you'll find Leafio AI. Please come and visit us and we'd love to talk more about your business and uh promotional planning as it relates [music] to you.