Analytics Out Loud: Episode 1
Watch on YouTube.
TRANSCRIPT
Beverly Wright (podcast introduction): Hello! I’m Dr. Beverly Wright, and thank you for joining Analytics Out Loud, presented by Center for Applied Business Analytics and the Master of Science in Business Analytics program at University of South Carolina Darla Moore School of Business. Analytics Out Loud focuses on innovation, research and pedagogy related to analytics and data science.
Analytics Out Loud leverages an interview format to pair academic and industry experts. Our topics include advancements in analytics and exploration of solutions intended to support complex decision-making by integrating perspectives from academic research, classroom and corporate.
Thank you for listening to Analytics Out Loud at University of South Carolina.
Beverly Wright (begins the interview): Hello, I'm Dr. Beverly Wright, and welcome to Analytics Out Loud. With us today, we have Dr. Mark Ferguson from University of South Carolina and Dr. Ron Menich from Catalina marketing. Welcome, guys.
Mark Ferguson: Welcome, thank you, Beverly.
Beverly Wright: We're talking today about leveraging data science to optimize promotional display. So, thank you for being here on Analytics Out Loud. So, let's start off with a real basic question about your background. Tell us — we'll start with you, Mark. Why are you so cool, Mark? Tell us about your background.
Mark Ferguson: Well, you may have some people debate the cool part, but I've always tried to incorporate, you know, the use of data and decision making my whole career. Started at IBM back in the early 1990s, as a supply chain manager there, and moved in academics in the later ’90s and focused mostly on supply chain management. Inventory management, moved into revenue management and pricing, which really incorporates, you know, a lot more had at least have more data available for it. And, more recently, into the retail space, working with companies like Oracle, their retail division.
Beverly Wright: Thank you, Mark. And we're so lucky to have you at University of South Carolina. And Dr. Menich, tell us, why are you so cool? Tell us about your background.
Ron Menich: Thanks, Beverly. Yeah, so, much like Mark, I have a background in revenue management and price optimization. I spent 18 years doing that and various perishable asset industries like car rental and airlines, passenger rail and so forth. Then in 2013, or thereabouts, I moved into retail demand forecasting and did that for four years, and then I came to Catalina three years ago, and I now lead a team of nine data scientists who are doing personalization activation and measurement of ads and coupons for groceries and CPD companies.
Beverly Wright: Very nice. I feel like the whole world is, is focused on personalization right now, like making things down to a segment of one. And that's what Catalina seems like it's all about. So, we're very fortunate to have you, Ron.
Ron Menich: Thank you.
Beverly Wright: So, the first question that I have for you guys is -- I’m gonna mostly ask Mark this question is -- tell us about this research and this work that you're doing, this body of knowledge you're building about optimizing promotional display. What does that even mean?
Mark Ferguson: Well, thank you. I mean, it’s a project that we started, I guess about five or six years ago, honestly. We were working, as I mentioned, with Oracle retail and I had a new doctoral student that was looking at the promotion pricing models, which then fairly well-studied in our field. And how we could use kind of these nationwide data sets that companies, you know, like Nielsen and IRI put out to gain insights, you know, for grocery retailers in particular on how they could maybe improve the profitability. And as we're going through it, you know, we kind of came up with was to us an interesting gap, if you will. And the literature and in the practice, you know, solution providers out there is that no one seemed to be providing any guidance on what items to put on promotional display in a grocery store, for example. So, think of end caps or islands or you know by the counter aisles, dump bins, there’s a lot of names for it. They had traditionally you know just been based on whichever CPD company provided the biggest trade promotion and gave them the you know largest incentive to put their product on, on a display or they might have some heuristic they would use like what's the best seller and our soft drink division or beer division or soup division and we'll just put that item on, on display for a couple of weeks, and then we'll rotate it around. But nothing focused on the data, yeah.
Beverly Wright: So it’s relatively subjective when we say best seller, we mean like what seems to be selling well. We weren't evaluating the numbers.
Mark Ferguson: Well, they would, they would know which their top sellers were but they would not necessarily know how much additional lift they would get if they put an item on a promotional display.
Beverly Wright: I see. Okay. Okay, gotcha. And what did your research find? Where their specific response, you know results that we should discuss?
Mark Ferguson: Well, you know, the promotional display is actually a really powerful driver of profitability for especially brick and mortar retailer. You know, we found that we looked at the beer category in particular and found about a 27% lift on average for the SKU of beer that they put on promotional display versus the ones you know that were not on promotional display. And that's not --
Beverly Wright: Mm hmm.
Mark Ferguson: That's taking out the effects of any other promotional efforts like putting in a flyer or having a sales promotion, a discount on that, or even seasonality effects. So just a flat off 27% lifts, which is a very, very you know if you work in the grocery retail industry, that’s huge.
Beverly Wright: Huge, yeah. Okay, so Ron, would you tell us like about the importance of this kind of work? Like, if you're trying to -- it seems to me that promotional display’s a big deal. Is it? How big of a deal is this?
Ron Menich: Oh.
Beverly Wright: I know, promotional.
Ron Menich: Promotional displays are, are an absolute huge deal. You know, there's an entire trade promotion business process that goes on between retailers and the vendors. Many times during the year to decide what will be on those displays, what will be in the weekly circular, arranging deals, you know between the category managers at the retailer and the various vendor partners. It's a huge deal.
Beverly Wright: Hmm. Okay. So, this research sems like it would be incredibly valuable, but I would imagine that there are going to be some challenges in trying to implement it. Can you talk about that a little bit?
Ron Menich: Sure. There's always going to be challenges and, Mark, you know, congratulations to you and your, your team on this research. So, you know, maybe I should call this considerations when when when implementing an algorithm like this. You know, first … first thing to note that I’m sure there were promotional displays in the markets of ancient Babylon, you know. There's nothing new in retailing. And as a consequence, you know, many people have been trained over the decades in their careers to do things in a in a certain manner and so any new decision process has to prove itself because it's not it's not greenfield it's coming in to replace an existing process to which many people have been accustomed, you know, whether that be good or bad. It's just that you have to be aware of the the change management implications of of you know, putting in such an algorithm and --
Beverly Wright: Hmm.
Ron Menich: You know, maybe. Go ahead. Did you have a question?
Beverly Wright: No, just that change is hard. I mean, and I see what you're saying there it's not like ‘Whoa, let's do -- let's start a promotion and now let's come up with a way to figure out what goes on promotion.’ This has already been a thing and Babylonian and so now we're trying to figure out what are we going to do instead of this comfortable thing that we've been doing over and over again. So that might be a barrier, for sure.
Ron Menich: Yep. And and when you think about how how are retailers structured. And then what are the, you know, what are the implications about putting an algorithm like this into production at at some retailer so, you know. Retailers are often oriented around category managers who know all the products in their category and they maintain relationships with the myriad vendors of those products and they're constantly trying to arrange deals with, um, with those vendors through the trade promotion process. And, you know, there's kind of a natural competition that comes between these category managers as to what goes on display. And, you know, so as as one deploys a method like this, I would think that some types of analytics that illustrate how fair the solution is and so forth would be useful to convince the category managers, particularly the ones that whose products weren't chosen to be on the endcap display that, you know, in fact this is this is good thing for the company and will lead to a good bottom line results and better results for everyone.
Beverly Wright: So, what I’m hearing, so far. This is a great, Ron, and thank you, this is wonderful. So, even though we understand that optimizing promotional display is like really important, I mean it's a big deal. Especially for brick and mortar. And, secondly, we also understand that huge value could be gained from developing an algorithm to do so, and that sounds like Mark's work with his colleagues has been, you know, something that can be really, really useful. Even with all that! The challenges that if I’m understanding you correctly that summarize our -- Number one, change management! You know, that's hard. Change. Just getting people to reevaluate the way they do things, it's hard. Secondly, is the way the retail is structured can have an impact and make it difficult. Third is there is a domino effect potentially with category managers, especially on the relationship side. And then, fourth, you can't just sort of make it happen. You have to do some convincing and potentially some analytics data science to prove out empirically how integration of this algorithm to help you optimize is going to improve your promotional efforts. Is that capsule -- does that kind of summarizes four main points? Or can we correct them at all.
Ron Menich: Yeah, I think that's good, and you know, there are, I think further considerations that align with that as one thinks about implementing maybe a software package that would support this type of algorithm. One of the the aspects you know I’ve read a bit about Mark’s algorithm and it is a hierarchical one that that first makes a decision at the category level and then at the subcategory level and then at the product level, if I’m quoting accurately, and and so that kind of naturally leads to maybe some thinking about how those decisions are made, and maybe when. Perhaps one makes a decision as to which category goes on display before one makes a decision as to what subcategories go on display and and you know the the the the -- If beer is chosen to be on the display for this particular week, then once that decision is made what subcategories go on display is of no value or only tangential value to the paper goods category manager, and so you know when you think about creating displays or reports there, you know, once you've gone gone past the category decision then then you're dealing with a different set of stakeholders. And maybe you have software that has login privileges that enable only, you know, to look within a category or some such thing. So, you know. Nothing insurmountable here, but as I hear the algorithm, I think, well, how do I apply it? How do I create software? How do I, how do I fit it in over time to the decision processes that need to happen. All of that.
Beverly Wright: Yeah, no, that's a good point. Thank you. And Mark, any thoughts on that before we move to the next question that has to do with the application? Because surely there are some areas that could absorb and take this in more easily than other areas. But, any thoughts or, or commentary that you wanted to add about some of the challenges?
Mark Ferguson: Yeah, I mean, what Ron mentioned are certainly challenges. They're not ones that you know we academics, we kind of blithefully kind of ignore a lot of times. Right? You know, you've got to convince someone to use this. But, over the years I found working with you know kind of the in-betweens between the practitioners that are actually, you know, doing doing this in their day-to-day jobs and the academics that are working on these models are the either the software providers with a solution providers that are coding, you know, some of this some of the analytics up into software and then going out and doing the sales and the change management, you know and all with the organizations is a sweet spot, if you will, for getting more of our science, actually, you know, into practice. And, you know, I’m very thankful that I’m on my end, because I’ve been on the other end as well. I know how much, yeah.
Beverly Wright: I’m sure it is. And, so what, what do you think about either either of you all, if you want to comment. But Mark, maybe what was intended, and Ron, what do you think would be best. But surely this doesn't fit easily with every type of retailer. So are there opportunities you see that are better for being able to take this sort of solution and integrate it, whereas other retailers, maybe there's a set or something of certain types or attributes about those retailers versus ones where there are higher hanging fruit.
Ron Menich: And, and so this hopefully it goes to your question, but there's, you know, first off the retailers who have endcap displays or who were, shall I say, high-low retailers who have items that are constantly on promotion and you can think about who those are, you know, versus, shall we say, everyday low price retailers who are not as focused on that. And, and so that's a …
Beverly Wright: Got it.
Ron Menich: That's a first division line. And then I think about who's implementing the software. In in retail there's been a lot of in housing recently, and so perhaps the biggest retailers might want who build this themselves and have you know have it be have have some customizations that are very specific to their business processes. And then the other model is, you know, an external third-party vendor who creates software and and maybe maybe then that that vendor goes after second-tier retailers who who do not want to build it themselves. Right?
Beverly Wright: I see.
Ron Menich: And I think, you know, what like what are we talking about here. Are we talking about trying to deploy this too many retailers, or one particular retailer. Those are some things that come to mind.
Beverly Wright: No, I love it. So first, table stakes. You’ve got to be a retailer that does promotion. You can't be some, you know, everyday low price or everyday high prices and – not that anybody admits that. But there are some that they they sort of live by that, maybe they're higher end. But you have to be a retailer that believes in promotions, and then actively does them. So that's kind of table stakes. And then the second piece being the consideration of is this something that you want to take as it is kind of and use it, or do you want to use internal teams to kind of make it your own. Or just take the idea of it and start from fresh or even use an outside vendor to take a different perspective and create a solution in, you know, in the form of an API or or some sort of software tool. So it sounds like those are some really good options and thoughts about who might be able to apply this in a better way. Mark, do you have any follow-up thoughts to that?
Mark Ferguson: Yeah, you know one thing we tried to do when doing this research is, typically, when you talk about data analytics, the larger chains have an advantage over the smaller chains. Right? There's just more data and more stores. They have more SKUs they sell. So we designed our methodology around the data set that's really available to any grocery retailer, which is the IRI, you know, data set, which is a sampling of stores all around the country that other retailers voluntarily, you know, share their point of sales data into. And you really need, yeah, you need a large kind of representative data set like this for our methodology to work. So wherever you're, you know, a Krogers or you're really large grocery retailer, Walmart or just a small regional retailer, you can get access to this data set and apply the methodology.
Beverly Wright: Okay, the larger is usually better just because more SKUs, more data, more ability to analyze. Okay. I also wonder if I could add to your attributes myself. Which are, I was thinking it needs to be fairly open-ended. And I know that sounds simple, and maybe it is, but it's also kind of rare. It's hard to find companies that are open minded about the way they do things, especially if they've they've been implementing the golden gut and it's worked well. Sometimes success can be an inhibitor to trying out different types of data science solutions. And then the second thing is potentially to have multiple locations, so not only big but also multiple locations, so that they could potentially develop the design of experiment. You know, Ron, you were talking about you should almost have to run analytics to prove that your analytics are good or you may have to to convince people to integrate and implement. So if you have multiple locations, a lot of times that gives you more freedom to develop a DOE sort of environment and really show empirically that the solution is beneficial. Would you guys agree with those or do you have anything you'd modify?
Ron Menich: Yeah. You know, to your last point, it's kind of a localization concept that you just talked about. And and you know, retailers have a long history of doing things manually or or then, you know, translating that into a thousand spreadsheets. And and as one tries to do some of these decision processes, assortment planning, display, etc., at a more local level, you know, either you have to do that with lots of people and spreadsheets or you have to have an automation mechanism surrounding that and and I think, you know, this piece of work is kind of oriented around the automation mechanism to help support a localization of displays.
Beverly Wright: Yes. Okay. Cool, cool, cool. And I know we're crunching up on time. This has been really useful. Very, very thankful, because the retail environment is such a big deal. And the use of data science to optimize within it is also a big deal, as you guys know. So thank you, thank you, thank you for this content. And my final question for you guys is for our listeners that maybe they are starting a new job or maybe they already work in a job or maybe they're a top leader and they're trying to figure this question out. What one piece of advice would you give someone who's trying to leverage data to optimize promotional display?
Mark Ferguson: I could take the first crack at that. If you're trying to do this within like your own store in your single location or even if even within your own chain and just using your own data, you're probably just going to be frustrated. Because what you're going to find is that only a very small subset of your SKUs and each category have ever been put on display and that's the only SKUs you’re going to have data on.
Beverly Wright: Wow.
Mark Ferguson: So, you know, just to use an example from the beer category, right? A store will put, you know, Bud Light on promotional display multiple times a year. So there’ll be plenty of data on how much lift you get when you put Bud Light on display. But there may be another brand, I’ll just make up one, like Happy Hops. You know, it's a craft brew that you've never put on promotional display.
Beverly Wright: Right.
Mark Ferguson: Unless you have that in your data, you're not going to be able to estimate how much lift you would get by putting this maybe lower sales quantity brand. But that might actually be your more profitable choice. And this is where using the the nationwide data set, you know, the much, much larger data set, because there's probably some store somewhere that's put Happy Hops on promotional display.
Beverly Wright: Right, right. Okay. So your number one piece of advice is think about additional data sets. Don't try to rely completely on yours because you may feel a bit like you're using only anomalies. It’s that scarce, or it could be. That’s great advice. And normally something that someone would have to suffer many months to figure out. Ron, what what final piece of advice would you give our listeners who are trying to leverage data for optimizing promotional display?
Ron Menich: I, you know, if they're new new entrants into the market, new data scientists, new analysts or whatnot, I would I would just counsel that if you have a chance to do anything with promotional data, take it, because this is just such an incredibly rich field.
Beverly Wright: It is.
Ron Menich: I’ve spent the last seven years of my career dealing with some aspect of this. Either, you know, personalizing ads for it or or forecasting demand. And and and Mark has this algorithm on what to put on display, and there's whole sets of separate supply chain issues on on how to stock up those endcap displays. Like, this is an incredibly wonderful area for practice and probably research as well.
Beverly Wright: Yep. So if you get an opportunity, take it. Do something, try to do anything, even if it's something little. Very nice. Thank you guys again.
Mark Ferguson: Can I put something quickly to that one? Beverly?
Beverly Wright: Yeah. sure.
Mark Ferguson: As someone who's from the supply chain side earlier in my career to more, I guess, the revenue pricing side later in my career, you make supply chain decisions, you may not see the results for months or years. If you make a pricing decision or promotion decision, you often see the result immediately.
Beverly Wright: Yep.
Mark Ferguson: That's that is, I’ll second Ron’s, you know, if you get an opportunity, this is a very dynamic area where you can learn very quickly.
Beverly Wright: Yes. No, that's a great point. We oftentimes sit around and wait for results to come back, and we never know what happened. With this area, it’s almost instantaneous. Very nice. Thank you again to Dr. Mark Ferguson from University South Carolina and Dr. Ron Menich from Catalina marketing for speaking to us today about leveraging data science to optimize promotional display.
Thanks for listening to Analytics Out Loud, presented by Center for Applied Business Analytics at University of South Carolina Darla Moore School of Business. Have a great dataset.