How To Hit Pricing & Promotions Out of the Park

Today's e-commerce explosion and changes in shopper behavior have translated into a necessary hyper-focus on improving agility across the pricing, promotion and markdown landscape. While retailers are stressed with uncovering ideal pricing, promotions, and markdowns, a recent webinar highlights how the use of AI can help identify strategic, channel-specific approaches to drive profitable revenue growth. Read on for the webinar transcript, including all of the presentation slides. Tim Denman: Welcome, everyone, to “Pricing, Promotions and Markdown Innovations in the Age of Artificial Intelligence,” which is hosted by RIS News and presented by Hypersonix. I'm Tim Denman, editor-in-chief of RIS, and I thank you for joining us today. For decades, retailers have relied on a number of techniques and solution providers to help them manage and optimize their pricing promotions and markdowns. Despite advances in analytics, complex rules and the availability of competitor intelligence, legacy solutions often miss the mark when it comes to meeting both the retailer and the shopper needs. With shopper behavior rapidly changing, and e-commerce reaching new heights, there is more challenge than ever to uncover ideal pricing, promotions and markdowns. In the era of AI, retailers are exploring a more strategic, channel specific and autonomous approach to pricing and promotions that drives profitable revenue growth. Today we will discuss the current pricing, promotions and markdown landscape, explore the key components to consider when evaluating a future-ready approach in technologies, and share some case studies from the field. With us today to explore the current landscape are two well-known and knowledgeable retail technology veterans, Alex Brannan and Todd Michaud from Hypersonix, a fast-growing venture-backed, Silicon Valley company specializing in AI-powered analytics for retailers and CPGs. Alex is an analytics professional and spent his career helping enterprises leverage the latest innovations in technology and data science. He's the director of solution engineering at Hypersonix, where he enables customers to augment their human intelligence with AI and automation. Todd is the president and chief customer officer at Hypersonix. He has an extensive amount of retail tech experience, holding top positions at Symphony RetailAI, NCR, retailX, Revionics, IDS, IBM. Alex and Todd, thanks for taking some time to join us today. Really looking forward to this great content. Alex Brannan: Thanks so much. Today, we're going to be walking through a quick overview of some of the different evolutions that we've seen in the industry from pricing, promotions, markdowns, and look backwards at where people have been. We're also going to take a look at what we are hearing from different retailers — what are they expecting, what they truly need out of a unified system. We're going to go a step further and talk a little bit about an integrated merchandising strategy and how AI can be incorporated to provide key recommendations across an item's life cycle. We’ll end the session by covering some different predictive and prescriptive illustrations of some representative platforms when you're looking to evaluate a unified price promotion and markdown optimization system. Let's talk a little bit about the landscape of where retailers are today. As we think about where retailers are, Gartner has done some valuable research on the different levels of solutions that exist in the market. At the very early stages you have people who are calculating prices manually. Whether this is just simple back-of-the-napkin math, where you're doing simple margin calculations, or even beginning to put it into Excel — that early stage, unfortunately is still out there today as different systems have lots of limitations and oftentimes it can be difficult to translate those processes. We've seen a more effective solution, that's a little bit mature, where people have gone and said, “Hey, look, we're going to start using Excel to still manually determine the prices, but be able to do so at scale.” Then following this is a very similar stage where you might still be using Excel or another type of system, and it's still somewhat of a manual process, but you start adding in criminal data, such as competitive data, different types of information across your channels. The problem is oftentimes these systems are very at a high level, just providing you a price and not necessarily the impact of what the optimal price would be. Now, where are people going? As we look at other organizations we see that rule-based pricing has been around for some time — Todd will talk a little bit about this — then it's bringing into optimization. As we think about optimization, not all optimizations should be treated equally. We'll talk a little bit about phase four and the stage four of the evolution a little bit. Ultimately, where we see people going is truly optimized pricing across a unified landscape of price, promotion, markdown optimization, that's channel aware. We've had different conversations, and across Todd's career and mine, we've met with a variety of different retailers, brands, and even other providers. What we noticed is that a lot of people fall into this middle bucket. Now you might have different components of a forecast or an optimization, or you might be leveraging specific types of data, but ultimately these things are falling a little bit short. Why is that? Think about why some of the tools that might exist in the market might not be optimal, it might be a little bit ineffective. There's some key trends that we come across. The first of which is oftentimes these things can be expensive, expensive to integrate, configure, administer, and pricing as a core part of a business. With all of the various cost changes, internal and external pressures, you need something that's going to be very dynamic and affordable. When we talk about dynamic, pricing systems have generally been managed quarterly or monthly. Even optimizations could take days, weeks, or even hours and by that time you don't have the time to wait in order to change a price necessarily. Even further, the process can be quite complex, meaning all 70% of pricing projects on average fail. Even if it does succeed, you'll have users who might not adopt it. Solutions can often be too complex for users and may need a lot of training — you feel you need a PhD to even operate the thing. You need to understand, can you have trust in the system? On average if you have prices coming out that might be, from a data science standpoint, optimal, but don't meet your business's requirements and they can't be actioned because they're not taking into account your promotional forecast or promotional calendar, your inventory, and all of those important components. It can also be a bit of a pitfall where things can fall short and that's where you really need to think of how an integrated solution could benefit a business, versus just specific tools that might be able to solve point processes within pricing. Now we'll go into the details of different solutions, but as we talk about these four components of being expensive, slow, complex, or without trust, these really perpetuate the feel of gut-feel decision-making. This is quite a problem as when you make a gut-feel decision, it can often result in a bad decision or a late decision if each of those have a cost against them. Really what we see is that you need to be able to make a fast decision with a strong benefit, and that's really how you maximize the value. Todd, can you talk a little bit about the artificial intelligence revolution which is happening, what that means for an enterprise specific to pricing and then some of the maturity models when it comes to pricing, and the different tools out there? Todd Michaud: Thank you so much. I thought that I would start at a pretty basic level, which is why don't we talk about, what is artificial intelligence and why is it a little bit different than tools that we've used in the past for pricing, promotions, or other forms of merchandising? Here's a very simple definition, I think it does a pretty good job of capturing the value of artificial intelligence. Artificial intelligence is the ability of a computer program to autonomously think and learn on its own. It includes a broad set of capabilities that augments what humans would otherwise need to do on their own, such as learning, reasoning, problem solving, and language understanding. There are a lot of words there, but if you think about it, it is about applying technology to the things that we historically had to do manually. Nowhere else, besides these age-old processes in merchandising and marketing, are being disrupted by artificial intelligence with diagnostic, predictive, and prescriptive decisioning capabilities. I want to stress to you that this isn't about the takeover of the machines. This is about actually having an effective blend between artificial intelligence with human intelligence to get better outcomes along the way. Let's put this into a bit more of a practical sense. Let's talk about the evolution of retail pricing, promotions, and markdown offerings in the market. The application of math and data science to pricing began probably in the '90s or early 2000s. There was a series of first-generation pricing and promotion tools and markdown tools that were very siloed. Oftentimes they would do price generation, but the reason why a price recommendation was made wasn't really known to the merchants, because they were largely generating recommendations as black boxes. They didn't do a very good job of managing the rules that we've always had around price or promotion image. They were poorly integrated with core merchandising systems. One of the fallacies with these early generation systems is that they treated all shoppers the same. Our best shoppers may be a little bit different than our secondary, or cursory shoppers, so it didn't make sense to treat everybody the same. There was an inconsistent, limited use of competitive intelligence in the early days. Given the computer that was needed, demand analytics would have been run once per category each quarter and that was deemed sufficient. Now, this whole first-generation gave way to a second generation of solutions out there, where price, promotion, and markdowns were interfaced. Black boxes were eliminated. You could see why price recommendations were being made. We would try to optimize pricing rules, our competitive, our margin rules. Still, these second-generation systems had challenges around being integrated with the core platforms. They would often generate pricing that rewarded less loyal shoppers. There was a little bit more use of competitive data, but it was still sporadic. One thing that got significantly better because of the availability of computers was that we could do weekly demand analytics across all items, all stores and that was much more accurate than the previous generation. These systems are still in use today. If you were someone that's already done price optimization, you're probably using a second-generation solution. In the 2020s and the decade ahead, it's about AI-powered unified merchandising. You'll notice a change to some of the jargon that we're using. It's not just about pricing or promotions, it’s about having a unified conflict-resolved merchandising platform, where everything is harmonized. One of the things that is critically important in this new era is visual transparency, where you have validation that the right decisions are being made. Rules need to be operationalized with machine learning capabilities. Today, it's no longer acceptable for systems not to be integrated with your core merchandising systems. Any optimization that's done needs to be customer-centric and the benefits or implications to our best customers need to be known. We need to be a lot more proactive about the use of competitive and market data in the price and promotion determination process. With the availability of cloud computing, the ability to do daily and even real-time on-demand analytics has become much more prominent with today's systems, which provides a greater degree of accuracy and learning along the way. If that's the look ahead, what does that mean practically? What retailers are looking for is unification around merchandising, where all of the capabilities around price, promotion, markdown, planning, and assortment are all integrated on a common platform across the entire product life cycle. It's no longer acceptable to have siloed systems that don't work together. These capabilities are built around a harmonized, AI-powered, demand forecasting environment that is aware of all of the offers and pricing in the system, our assortment, and our space and our inventory decisions. These harmonized demand forecasting environments have to account for seasonality, events, holidays, and even weather. Imagine having one version of the truth from a forecasting perspective, that can empower all of the core applications in a unified fashion, to offer integrated solutions across the product life cycle. The other thing that's so important is that we're no longer just pricing for stores, but also for online and digital channels. Making sure that we have channel-intelligent pricing for curbside and home delivery is so important — we've already touched on the fact that our determination needs to be much more shopper-centric and localized. While we can bring a lot more automation, speed, and timeliness around decisions, the goal is to augment human intelligence with artificial intelligence, where it makes sense to provide some level of hands-off-the-wheel automation. Maybe I don't need to look at 100% of my price change recommendations, maybe 80% of them are business as usual, and the system can essentially process those. Category managers and merchants should spend their time on the 20% that really needs their eyes, that's what I mean by hands-off-the-wheel automation. Of course, this requires a buildup of confidence in the system's ability to produce accurate, timely price or promotion recommendations. The key theme that I want to stress is that it's about unified merchandising. Now, on the previous slide, I talked a little bit about the product life cycle and having a unified environment. Even though this is fundamentally about pricing promotions and markdowns, it starts with having intelligence around the assortments that are offered localized through digital channels and understanding the transferable demand that exists between different items. The fact that the needs that we have in different channels are starkly different — there are 60,000 items sitting at store level, but that doesn't mean that I need to have an online assortment of 60,000 items selling digitally. Understanding channel preferences, what my assortment should be, and having some intelligence to understand when I should add items to the assortment and when I should de-list items to the assortment. In reality, this is where it all starts. With the right assortment optimization pricing, promotions, and markdowns become an important attachment to this fundamental capability. Let's assume we're applying to Teligent assortment optimization around listing or delisting items, that unified merchandising environment. When we think about the product life cycle, everyday pricing or intelligent pricing today, is helping us from the time that we introduce an item through its growth and maturity stage. Generally, artificial intelligence and optimization can be about realizing objectives that we have at the category-level or zone-level around things such as driving revenue, profit, margins, or competitiveness. Intelligent pricing for the 2020s is about understanding in a data-driven way, where can I drive revenue versus where should I be driving margin? Pricing determination can be different whether the shopper's engaging digitally versus in my physical stores. Key capabilities are about being goal focused. In terms of intelligent promotions, promotions have significantly been impacted over the last year — what promotions work, what promotions drive good outcomes. This is where artificial intelligence can help in a couple of different ways. First, evaluate past promotions and campaign effectiveness. Understanding bad behaviors that our promotions may have historically trained our shoppers to do such as cherry picking, pantry loading, or offer stacking. When we talk about predictive or prescriptive capabilities, it's about planning and forecasting better promotions going forward. What items are you going to promote? At what offer level are you going to promote? There's vendor sponsorship along the way, and what's the right promotional mechanism. Is it going to be an ad or digital circular? Is it going to be a TPR? Will it be something that you're doing specifically for your frequent shopper with personalized deals? When you do promote something, do we understand the cross effects that exist with other items? For example, I put this item on sale and it has a positive or negative influence on other items that I might be selling. If I put hot dogs on sale, am I selling more ketchup, mustard, and relish at higher price points because of the sale of hotdogs? Understanding the cross effects that exist, or what we call the affinity, is a key part of how artificial intelligence can help. Then, very importantly, understanding shopper engagement by segment or cohort, understanding where they're engaging with us, what categories, what promotional offers work, etc. The whole idea is that we have so many more data points available. Artificial intelligence can help to be far more effective at getting yield out of promotions from the right shoppers moving forward. The last part of the product life cycle is around intelligent markdowns. This is all about clearing of inventory, which could be seasonal merchandise assortment changes that you're making within your stores, inventory resets. You may have store closings for buys that maybe you took too heavy a position on, aged packaging, markdown optimization can also help with perishable goods if you're selling perishable goods. The focus of artificial intelligence around markdown optimization is understanding the tradeoff between profit and accelerated depletion of your inventory, and being able to optimize on a localized basis, the right discount trajectory to ultimately out your inventory as it makes sense for that localized environment. Think of markdowns as the end of life environment. One of the things that I'd like to introduce, we talked a lot about channel specificity and shopper segment awareness. The best practice around most pricing is to get personalized with your pricing, deals, and offers for your frequent shoppers or customer segments. One of the things that needs to be brought into the pricing discussion, especially as we see fewer mass promotions being done, vendor partners and retailers are looking for opportunities for better ROIs from promotional investment by doing targeted offers. It is clearly a best practice, in concert with the life cycle pricing, to be able to offer intelligent personalization for shoppers today. Understanding what offers I may have an interest in, specific brands, deals that might be responsive, things that might help grow category participation, and being able to use AI to target it specifically to individual shoppers based upon their preferences on how they'd like to engage with a retailer via email, mobile devices, paper receipts — whatever the case might be. As we talk about pricing in the 2020s, personalized pricing has to be a part of the mix, as well. Alex, is going to show some practical examples of each of the illustrations highlighted, and provide some specific examples where artificial intelligence is changing the way decisions are made. Brannan: Thank you so much, Todd. I want to give you a little bit of a preview into some of the work we're doing at Hypersonix, and some illustrations of these capabilities. One of the things we've developed is a suite of augmented intelligence applications. As you look at solutions, you need to make sure that they are integrated. Todd spoke about the key components of each one of those different types of pricing and the importance of them to be connected not only within a platform, but also from a data science standpoint because a change in one could very much impact the other. At Hypersonix we've done exactly this and we've unified data together in order to provide this. Before we get into specific applications or instances, how do you even know where you should be focusing your time or what's happening within your business? As you look across a unified merchandising system, one best practice is that you shouldn't have to go to multiple places for insights. Rather, you should be able to have a newsfeed of insights generated for you automatically based on your preferences or insights, and specific to our conversation today. I'd actually highlight one of the insights in the middle column, the second one down. Here is an example of an insight that's been generated as new data is coming in in real-time, letting you know that a specific percentage of your stores — 75% of our stores — are having a year-over-year sales decline. It highlights that the stores belong to a specific pricing zone and is calling out that zone. What you might notice is there's a little light bulb next to it. While it's great that artificial intelligence can inform you of key trends, changes, and anomalies within your business, it needs to be explainable. Users need to be able to click into things and clearly understand that it's taking information across the business. In this case, it's alerting us about a competitor that has opened up two stores in a specific radius of our business and it's going to impact our sales. Doing so in a way that anyone can understand, using artificial intelligence innovations (like natural language generation), and in a way that any business user can have a common experience that's consumerized, very familiar to them. As we talk about pricing, we can look at information like base pricing and what is the base price strategy. Base pricing is going to look at components including revenue, profit, margin, and units sold. Look at the current values and the suggested values. Being able to automatically generate this information based on artificial intelligence — in compliance with your different constraints that have been applied — and ensure clear, actionable recommendations that a user could follow. In this case, we can see we had 325 items with suggested price changes, which we went through and found 224 that meet our criteria, and then a specific amount that is recommended for your business. It should be something that's familiar — where you can clearly translate it to your hierarchy, and ultimately get to a price review screen for a user to go through and modify, change, do what-if analysis, and leverage artificial intelligence. If I was to change something, how does this impact the specific recommendations? Todd mentioned this earlier, but it needs to be transparent that if I do modify it, I need to know how this meets what artificial intelligence recommended, versus what I've put in. This is something that translates back — after the prices have been changed, being able to understand the full view of what actually happens. You can improve recommendations, but the solution should automatically improve its recommendations through innovations like machine learning and other types of AI techniques. We've covered a lot of different types of pricing. In addition to base pricing is promotional pricing and these things go hand-in-hand. You don't want to change your base price on items during a promotion. Of course, there's common constraints like that, but as we think about promotional strategies, there are certain items and categories that should be promoted. There's also some that might not be best suited for a promotion. It might be better served at an everyday low price. As we look at the different types of promotional techniques, what you need to check is a solution that can decompose promotions and provide clear visibility into performance. Now, this shouldn't be your base sales and your promotional lift of those sales, but it needs to be both from a direct and indirect effect. Direct effects your promotional lift, for example pulling customers forward if they're going to stock up, if you have a category where a consumer can expand consumption. On the inverse side, indirect effects to where you could have key aspects of halo or cannibalization that are occurring within your business. You want to be able to provide a clear insight into how all of these things happen because what might be good for one item, might not be good for a category, department, or the store overall — being able to have that big picture. Now that big picture is helpful, but it has to translate into something that's actionable. Whether you're a merchant, pricing analyst, or an executive looking at what's happening within your business, it needs to translate to real outcomes, clear departments and categories, which are performing, which aren't...even all the way down to the different types of tactics and offers. As we look at promotion solutions and the aspect of an integrated promotion system, it shouldn't just be for someone's weekly ad or flash sales. It needs to consume every single part of a marketing calendar to provide a unified view across a business, so they can clearly understand what's happening. As we further promotion and follow Todd's suit, there's also the end of a life cycle. Markdown in some businesses can be used as a promotion, but you need to be able to consider things like excess inventory. Having a system that's integrated that can connect your inventory systems, pricing, promotion, and understand when that opportune time is to take a markdown and not just provide a markdown recommendation, but clear insight into the financial benefits of doing so. This is a strategic decision that is translated down to individual items. As we look at individual product lines, categories, and items that are going to go through a markdown strategy, it can't just be some elasticity number and a price or elasticity number, and a percent discount. The optimal recommendation in today's era is to provide something that's clearly explainable, a set of goals or strategies. In this case, we have specific items that we have selected to our objective of thought driving sell-through, other items we're looking to liquidate, some we're looking to drive profit while still clearing out by close out dates or sell-through dates, to really make sure we're meeting our objectives. All of these things, what we're going through with these illustrations are constraints. Behind the scenes if your system can't connect between markdown, pricing and promotions, and also can't connect to the core business rules you've put in, you could fall short. If we look at all of these different tactics, we have the opportunity with the expansion of digital and of consumers adopting new technologies, to get to personalized pricing. This could be a personalized promotion, a personalized markdown, a personalized price even for an everyday price. To do so, you need to be able to tailor down to specific customers. Being able to create different cohorts of customers and segments. Then do all of the processes we just talked about — look for customers who spend a specific amount in X-category, have taken this behavior, and are part of my loyalty program. I may begin to engage very intentionally, provide specific pricing for certain products they are loyal to, provide specific discounts, and even rather than marking down inventory for everyone, provide a lesser discount for just a few customers. Being able to optimize markdown spend based on different customers and effectively distribute that to which customers ensure you're putting your money where it counts and also driving loyalty. To do this, it's not practical to go through and set a price and then grab a group of customers and do it all manually. You need an optimization system, and more importantly, you need a system that's going to connect to each component, because you won't only be personalized price — you're going to have regular prices, too. You won't only be personalized promotions, you'll have regular. It's the blended mix of all of these things where artificial intelligence can really help. A system that has a series of applications that's built on a platform across your data sources is a core system that can help drive these things. Tim, I'm going to hand it back over to you. Denman: Thanks, Alex. Thank you also to Todd. We have quite a few questions here and we’ll try to get to as many as we can. We’ll start with this one: is the AI component of the solution proprietary or how do you embed it an outside tool? Michaud: I'll take that one. It is proprietary, much of what you saw we built ourselves. One of the important things about our platform is that it is open. We have the ability to also bring in open source intellectual property, which is a best practice today. Gone are the days where you just depend upon a vendor to do all of the innovation around data science or artificial intelligence. Much of what you saw today, we built, but I do want to stress that we're quite open about incorporating open source and even client's specific capabilities along the way. Brannan: That's a great point, Todd. As we've seen more and more retailers developing a data science expertise or a data practice within their enterprise, you also can't have a solution that doesn't allow them to easily extend and augment it with their own data science. It needs to be something that works for the business as well — that's been a best practice that we've implemented. Denman: How does the system identify and handle the new normal promotions? Obviously there's a difference between the promos and markdowns in the last year and what was before that. How does the system differentiate the two? Brannan: That's a great question. Every demand forecast since the introduction of COVID has been challenged. One of the things that has been instrumental in refining this is being able to leverage machine learning, but also leverage relevant data points that are outside of traditional transaction information. As we look at systems and the types of information to go about recommending a price or promotion, you need to be able to take in everything that's relevant — today that is COVID data. It's COVID data for the past in order to know how that impacted customers, but it’s also being able to ingest the forecasts that are available from external sources, like future-looking COVID information, future-looking weather information, future-looking GDP information that's publicly available, or even things that retailers or other organizations are subscribing to, to be able to generate a highly-accurate demand forecast and ultimately, highly accurate recommendations. Denman: Thanks, Alex. Thank you also to Todd. We have quite a few questions here and we’ll try to get to as many as we can. We’ll start with this one: is the AI component of the solution proprietary or how do you embed it an outside tool? Michaud: I'll take that one. It is proprietary, much of what you saw we built ourselves. One of the important things about our platform is that it is open. We have the ability to also bring in open source intellectual property, which is a best practice today. Gone are the days where you just depend upon a vendor to do all of the innovation around data science or artificial intelligence. Much of what you saw today, we built, but I do want to stress that we're quite open about incorporating open source and even client's specific capabilities along the way. Brannan: That's a great point, Todd. As we've seen more and more retailers developing a data science expertise or a data practice within their enterprise, you also can't have a solution that doesn't allow them to easily extend and augment it with their own data science. It needs to be something that works for the business as well — that's been a best practice that we've implemented. Denman: How does the system identify and handle the new normal promotions? Obviously there's a difference between the promos and markdowns in the last year and what was before that. How does the system differentiate the two? Brannan: That's a great question. Every demand forecast since the introduction of COVID has been challenged. One of the things that has been instrumental in refining this is being able to leverage machine learning, but also leverage relevant data points that are outside of traditional transaction information. As we look at systems and the types of information to go about recommending a price or promotion, you need to be able to take in everything that's relevant — today that is COVID data. It's COVID data for the past in order to know how that impacted customers, but it’s also being able to ingest the forecasts that are available from external sources, like future-looking COVID information, future-looking weather information, future-looking GDP information that's publicly available, or even things that retailers or other organizations are subscribing to, to be able to generate a highly-accurate demand forecast and ultimately, highly accurate recommendations. Michaud: When we see deviations in customer behavior from what our predictions might've been, being able to pick up on that deviation at the earliest moment and react to it is part of the value that these scalable AI platforms bring to the table. One illustration of the importance of this is that if we were to go back a year, think about how much more of our business was transacted exclusively in our stores and how many retailers have had to get really good, really quickly on digital channels, such as curbside and home delivery. The reality is that a lot of the forecasting systems out there have a store bias in their predictive forecasting and artificial intelligence, particularly one that's designed to be channel capable. It gives you the ability to understand that we're seeing a lot more volume through curbside pickup and maybe it's at the expense of a traditional annual store shop experience. Today, just one new reality perspective that I would share is that your forecasting has to be channel aware, channel specific. If it's not, then it's not ready for the post COVID era to begin with. Denman: Todd, I'd imagine so, but there's a solution for all retail segments — we've got a question here on whether you guys work with apparel retailers. Michaud: Yeah, we do. As you look at our customer base, we have customers across an array of retail industry sectors. Most of our customers are fast-moving consumer goods retailers, for sure. But we actually do have a couple of apparel retailers, consumer electronics, automotive, and so as the company continues to grow, it's our intention to deliver industry-specific capabilities for the needs of an apparel retailer versus someone that might be a grocer or convenience retailer, as an example. Denman: In terms of software integration. I don't know if we need to name names, but there is more than one question here whether you integrate with this software or that software? Can you talk about your biggest partners that you integrate with? Michaud: Let me take the question a little bit more holistically. One of the fundamental things that we do differently with Hypersonix is help our customers bring a wide variety of enterprise data and external data to this unified platform. We have done 30 different point-of-sale systems and virtually every loyalty or marketing system out there. We've connected to a number of the primary e-commerce solutions, including some of the third-party marketplaces. If you think about any retail enterprise, they have such a complex landscape of different things that are running within the business. One of the core value propositions that the company uses to deliver this unified experience around merchandising and marketing, is to unify the data. We use some best practices around auto-machine learning to adjust the data with a much higher degree of automation. We've got some IP around auto-unification, which helps us to have a 360-degree view of our shopper, our product assortment, our channels, our store operations, etc. That whole landscape of enterprise data blended with external data, that external data could be competitive intelligence, market or panel data. Alex already touched on COVID-19 data as being samples of external data that could enrich the decision process around merchandising and marketing. We largely connect to anything and we don't really see ourselves as an adversary to any of the core platforms out there. Largely, we sit on top of those platforms, ingest data from them, analyze insights, and then publish insights back to whatever that system of record is. Whether it's for pricing, promotions, inventory, or other capabilities that we have. We think of ourselves as a system of intelligence that sits on top of the systems of record that retailers have already invested in so far. We have connected to hundreds of different types of systems, and the average retail client may have 10-12 different systems that we're connecting to, to provide this unified merchandising capability. Denman: Alex, the system that you showed at the end, you had some screenshots of a demo. How does this system actually handle cost changes as a consideration for valuing price changes? Brannan: That's a great point. As we think about the types of data we take in, cost change is just one of the many variables that are relevant. If we think about the external sources that Todd was just talking about — cost change, historical transaction information — we think about different types of information around things that are related to two products being related to one another. These are all things that are inherent in how we've implemented and created this platform to where or when things like a cost change happens. Oftentimes your instinct is, “I need to raise my price if my cost goes up.” That very well may be the right decision, but the way we built it is to look at, yes, your cost went up, but then what is that optimal price to drive your revenue, profit, units, a halo impact into the store. It is just another one of those many components that we leverage to derive an optimal price. Michaud: Even more specifically, just because the cost goes up on one given item, doesn't mean that same item should experience a price change. I get something that's more shopper centric by passing the cost change to other items versus making adjustments on pricing for other items versus the item that experienced the cost change. Denman: What are some best practices for competitive and market data? How much weight should they be putting into competitor data and how frequently should it be collected? Michaud: I'll take the first stab at this. First, as we work with retailers, sometimes we find that they're using too much competitive data for price determination. They're reacting to competitors pricing on items that they don't get credit for, from their shoppers. Understanding sensitivity at the item, at the location level, is really paramount to understanding where you should be chasing or following competitors. Just because you can get all of Walmart's prices or whoever your competitor happens to be, doesn't mean that you should necessarily react to every price change or competitive price that you get. A key part of what we try to do is understand sensitivity at the item level by location and help our customers understand where they should respond to competitive pressures and where they should not. Will the shopper not give them credit? This goes into what are the key value items, key value categories, where is it useful to be aware of what the competition is doing versus where should I actually trigger a price change because my shoppers expect me to be in the same ballpark? In some segments, I will point out, with particularly high ticket items, competitive intelligence is dynamic at the time of an online transaction. That's quite a common use of competitive intelligence. If I'm buying something online and traditional consumer goods sectors, not quite that dynamic, usually getting updated competitive pricing once a week is plenty sufficient. Denman: At one point you guys were talking about “hands-off-the-wheel automation.” What does this look like for a pricing analyst or category manager? Michaud: I think for a pricing analyst one of the challenges with previous generation pricing solutions is that it would generate a couple thousand items and then, literally, line item by line item, the category manager or pricing analyst would have to go through and say, do I want that? Do I want to make a change? It was too laborious. The basic idea around hands-off-the-wheel is to say, what's the 80-20 rule? Are 80% of the items that are performing changes completely inspected, but don't need the eyes of a pricing analyst on it? Can I free up their time to focus on the 20% of items that we really need human intelligence on? That's the concept. Of course, this comes with time as pricing analysts and category managers get more and more comfortable with the system's ability to produce pricing that is accurate and trustworthy. As their confidence increases, the willingness to depend more on the automation for some of the basic blocking and tackling increases. In the end, this is about giving merchants, category managers, and pricing analysts more of their time to focus on needle-mover items or more important things that they should be looking at. Denman: Great. We have time for one more question. When doing price optimization, some systems require price change history to know how customer data is influenced by the price, but how do you deal with items that don't really have a price change history yet? Brannan: That's a great one. It's going to depend on that scenario. Let's cover a few of them. The first type of scenario that's often run into in that instance is that you have an item and then you have an item where it's really the same item, but it might be a different number. In that case, you create a linkage. That's easy to do in a core system, but it also can be done in Hypersonix. The other thing you should consider is you might have an item that comes in and it's very similar, but it might be a variant. In that case, you can still denote and say, this is a similar item as this one, but the important thing is to have artificial intelligence come in, be able to understand, and create a forecast for that item. As new data comes in, your forecast shouldn't have to be reliant on months and months of data for this new item to get its own entity identity. Rather with AI, you can have an item begin to create its own identity with unique behavior, unique elasticity, just days and weeks after getting initial transaction history. The other aspect is there are cases where you might get a brand new item, or an item that's new to you, where you might take a competitor's price and have this also be an influential aspect as a nice starting point. I think there's also no shortage of options as we think about the different techniques to implement this. Todd, can you think of one or two I might have missed. Michaud: The one other one that we see within our customer base on a brand new item, or perhaps an item with no price change history that's been in the assortment for a while, was that we can put it up for a test-and-control environment. This essentially deploys an array of price changes down to store levels and different price points for that item to see what works best with our shoppers. Then we inject a learning process on something that we don't have any prior history on. That's the other practice that we can do with this platform, that of an item without prior price change history. Denman: Great. Well, thank you gentlemen. As I mentioned, we are out of time for today. Thank you both for joining us, and thank you everyone for listening.

Related Articles

Related Topics