Leveraging contextual data, understanding and planning for fresh (produce and items prepared in store for sale), as well as promotions and everyday items are all reasons why AI is finding its natural home with demand forecasting.
Learn more from Patrick O’Mara, Senior Solutions Consultant, Symphony RetailAI as he talks about our AI based demand forecasting tool with Mike Troy, Editor-in-Chief at Retail Leader.
- RSR Research blog: Next-Gen Demand Forecasting: AI Up Front And Center
- Read more on 3 reasons AI and retail demand forecasting are just meant to be.
- Read Video Transcript
MT: Hi, Mike Troy with Retail Leader, here in Irving, Texas at the Symphony RetailAI Xcelerate event. I’m joined by Patrick O’Mara. You lead all of supply chain replenishment for Symphony.
MT: And what are you focused on here at the event? What solutions are you talking to retailers about?
POM: Sure, well, we’re having conversations throughout the entire supply chain. One of the really exciting things that we’ve been focusing on, is our AI based demand forecasting tool. You know, leveraging third party contextual data, like weather, promotions, price sensitivity, to help increase retailers forecast accuracy. Ultimately drive, you know, the traditional ROI even higher for retailers.
MT: When you say contextual data, is that like unstructured data? Is that the same thing?
POM: Yeah, it’s…
MT: What does contextual data mean?
POM: Sure, contextual data is, you know, any data really, from a third party. So think about like weather, competitive pricing that can be applied, things like traffic. So looking at those kind of patterns and understand how they influence, they build fresh items as well as standard grocery items in a retailer’s operation.
MT: Are there particular channels, retail channels that you’re focused on, you mentioned grocery?
POM: Sure, so really grocery, convenience, any fast moving consumer goods. But Symphony’s got a long history of working with highly perishable items, so grocery, convenience and then even in e-commerce start-ups that are really focusing on that space as well.
MT: So there’s obviously been a ton of change in food retailing, as far like how people buy and you know, pick up orders and have things delivered. How has that affected the demand forecasting replenishment model?
MT: How’s your solution accounting for that?
POM: So it’s made…it’s certainly made a major challenge for retailers to be able to understand where the demand is coming from and where they need the products through their network. And one of the areas that that’s…or that’s an area of major focus for us, so taking demand signals from their root cause, whether that be an e-commerce demand signal or a traditional point of sale line demand signal, being able to factor those and be able to drive that forecast accuracy at the point of pickup for the customer.
MT: Are you able to look into like you know when people search for things online, you know, like what people are looking at? Can you take…do you take that data and factor that into a forecast?
POM: It’s actually a unique challenge we’ve looked at with one of our customers and are looking at embedding that type of information, you know, beyond just that, looking at things like an item’s star rating, an item’s rank on its website, so where it falls on the search results page and if an item is a preferred item that’s pushed to the top. So those are all things that can be included in the forecast to again, drive a little bit more forecast accuracy.
MT: Hmm, so where are we headed with all this stuff? Leveraging AI to get more accurate forecast, improvement replenishment, what…as you look out into the future, a year or two, things are moving pretty fast.
POM: Yeah, so continuing to be flexible, I think, is a watchword for retailers in that space. So being able to understand the different inputs and how those are going to change over the next few years. So as things like e-commerce, your different pickup points, how you need to replenish to specific locations. I see that as a major challenge that needs to be solved, to basically cut down the labour input and make it a more cost efficient process. So really having the AI and machine learning driven forecast to help understand the unique movement of any given item, is a critical factor for driving more margin for our retailer.
MT: Okay, let’s end on this last question. What sort of input, feedback are you getting from retailers on what they like about this best and what they think you could maybe tweak the algorithm?
POM: Sure, so you know, I’d say there’s not a lot of negative feedback as far as how to tweak the algorithm. Really, what we see as more challenges brought to our R&D team. You know, a great example is one retailer had a focus on highly promotional items. They were having a challenge hitting their metrics, so they wanted to focus the forecasting on that space. They were able to actually achieve about an 86% accuracy on promoted items within a one to two week range, you know, which is well above industry standards, so…
MT: 86% they were able to forecast what the sell through on that item would be?
POM: During that promotion.
MT: How would that compare to like say what a normal forecast would be, accuracy?
POM: Sure, so a retailer who’s managing the data extremely accurately, can get regular turn items at about 70%-75% accurate. Promotion items generally fall in the 60%-65% accuracy on a promotional forecast.
MT: Because it’s hard to know what the reaction’s going to be, by store, by geography, based on some of the contextual stuff you mentioned, related to weather…
POM: Based on the weather, the time of year, the season, as well as some things, you know, like cannibalisation and halo effect. So if I put hotdogs on promotion, how’s that going to affect my buns, my mustard, my mayonnaise, hopefully not mayonnaise, but you know, giving you a lot of…giving you a lot more data on items that maybe you’re not having to direct impact on, but have a very indirect impact when you put something on promotion.
MT: It all comes down to improved accuracy, better profitability.
MT: Alright, thanks Patrick.
POM: Absolutely, thanks Mike.