University of Arkansas

Walton College

The Sam M. Walton College of Business

Episode 18: Brent Williams

Brent D. Williams, Ph.D. is the Associate Dean for Executive Education and Outreach and Garrison Endowed Chair in Supply Chain Management in the Sam M. Walton College of Business at the University of Arkansas.







More Episodes

Listen on Apple Podcasts
Listen on Spotify
Listen on Google Podcasts
Listen on Amazon Music
Listen on iHeart Radio
Listen on Stitcher

Episode Transcript

[music]

00:07 Matt Waller: Hi, I'm Matt Waller, dean of the Sam M. Walton College of Business. Welcome to Be EPIC, the podcast where we explore excellence professionalism innovation and collegiality and what those values mean in business education and your life today. I have with me today Brent Williams who is an associate dean here, in the Walton College, and a professor of supply chain management, he has been doing research on forecasting and inventory management for many years, especially within the consumer package goods, and retail contexts and he has published many papers on forecasting that are published in top journals and he works with many companies on forecasting as well. Brent, thank you for joining me.

01:09 Brent Williams: Thanks for having me, Matt.

01:10 Matt Waller: So Brent, I know early on you did some... Your dissertation looked at forecasting orders from retailers, which is kind of a... You think that would be easy, you could just look at the orders and use a traditional time series method and forecast the orders, but then with the idea of bullwhip that's come in, this idea that uncertainty gets amplified as it moves up the supply chain, the thought always was, "Well, we should forecast at the POS level, the point of sale level, there's less uncertainty there." But your dissertation, sort of showed that neither are correct, that there's something in between is correct. Would you mind talking a little bit about that?

02:01 Brent Williams: Yeah, that's right. Those were at a time where point of sale data in retail had been available for probably a decade but there was a real effort to learn how could we integrate that data stream upstream into the forecasting process and you're spot on. The approach before that is take shipment data or order data, use a model and it was inaccurate to some degree. And so the question is, is it inaccurate within the bounds that you're okay with as the forecaster as the manager? But as POS data became available, you're right, it has less variability in it, so the thought is Well, it must then be a better predictor, and in some cases that's right, right? We found that in many cases that's correct, not always though.

02:58 Brent Williams: So the thought that spurred the research was, "Well, what if you could use both?" And so the research at the time focused on integrating both demand signals the order signal and the POS signal into a single model. There was also another fairly interesting piece to that. Again, this is really simple stuff, if you think back at this point. But one of the other questions that people always had about forecasting orders from a retailer, if you're a consumer package goods company is well, doesn't the amount of inventory held by the retailer affect the forecast? And the answer is yes. The problem was, "Well, how do you get reliable data on that, right? And so people were always thinking about, well, is the inventory data that a retailer might give me if I'm a supplier, they may or may not give it to me. If I got it, is it accurate?

04:03 Brent Williams: I think the little insight from that research was maybe I don't care so much about the actual level how much... Maybe I don't care about how much inventory the retailer has, what I care about within a given period is how much did it change? And if you think about orders and sales, right? POS data and order data, they represent the inflow orders of inventory and the outflow sales of inventory. So, in the model by taking by using the difference of those we were actually able to account for... To have a measure of how inventory changed by using those two signals.

04:45 Matt Waller: Well, I know you've had many other studies on inventory and forecast as well and you teach it both through executive education, and in four-credit courses here at the university, but what I'd like to talk to you about a little bit today is just forecasting in general and we'll look at some of the specific topics you've covered, but I wanted to talk about how disruptions that are occurring in industry, in technology, might affect forecasting in the future, and where, forecasting needs to occur.

05:25 Brent Williams: Well, what I would speculate is that I think technology, I think artificial intelligence as an example, will have an effect. But even if you just think about, if you think about even more simpler than that. When I mentioned the research that you and I were doing a decade ago, we were, at that point, integrating two signals into the forecast. Well, now with the computer power...

05:58 Matt Waller: And the two signals were shipment data and POS data?

06:03 Brent Williams: That's correct. But many other things affect the forecast, the weather, just as an example. We talked about inventory, but there's a myriad to these factors.

06:15 Matt Waller: The day of the week.

06:16 Brent Williams: The day of the week.

06:16 Matt Waller: The time of the day.

06:17 Brent Williams: Right. So what's happening in social media right, at the time, how is that affecting a specific... The sales of a specific product? Alright, well, that's where the opportunity really lies now, right? To start to integrate multiple signals, I think into a forecast to continue to make it more accurate. Now I think that, I think... So I think that's where we will see forecasting continue to go. That's been the trend for quite some time.

06:46 Matt Waller: When you think about forecasting too, there are times you need to forecast and there is times you need to affect the forecast. In other words, I can take all of the givens, as being exogenous, exogenous means they're not in my control. But in some cases, I can actually affect them. The most obvious one is price. If I really want sales to increase, sometimes I can just simply decrease the price and I'll sell more.

07:18 Brent Williams: That is right or you could push it harder, on social media or in advertising or change your display strategy, if you think about a sales and operations planning process, that's Part of that decision process at the end of the S&OP process is that if I have a gap where I'm expecting to have more capacity than I have demand then I may choose to stimulate demand. And that's what you're talking about.

07:48 Matt Waller: We know there is different levels of forecasting. Your comment about sales and operations planning, made me think of this. If you're trying to forecast sales for a company for next year annual sales, you couldn't do a fairly good job of that, but if you say, Let's take Walmart for example, suppose you wanna forecast sales for next year for the company, you could take several years of historical, several quarters of historical data, and forecast it but if you wanted to forecast bags of Purina dog food, are going to sell in the Super Center in Fayetteville off of Joy Street on Thursday of next week, that's gonna be very difficult. You could be way off...

08:43 Brent Williams: Yeah, that's right, right, you are right. What a seemingly insurmountable challenge to do that, but in many cases, in our supply chain, we're challenged with doing that... So we're talking about millions of forecasts that have to be generated.

08:58 Matt Waller: I wonder, I kind of wonder, the reason I brought up the ability to change things. I think that in the future we are going to have forecasting systems that are not only looking at what's happening and trying to project forward but they're also using artificial intelligence to flip switches so to speak, to start affecting it in other words... Yeah, we want to sell this many this is the range we wanna be in. And so, you could, for example, I conceivably have social media including Google ads, Facebook ads, etcetera, etcetera, where the system is monitoring sales across the country monitoring inventory levels, and you're saying, "Hey we need to sell this inventory down", and then they could all of a sudden start utilizing social media ads and say, "Well where is this working?" Where is it's not where should we put more investment? I think maybe in the future, it's gonna be more holistic and dynamic.

10:10 Brent Williams: Rather than today make a forecast and then start making business decisions based on that forecast.

10:17 Matt Waller: Yeah, I really think it's gonna be more experimental, you and I have talked about this before. I think inventory management, forecasting. In the future, are gonna be completely integrated, and it's gonna be less about applying these models we've been teaching for years and more about experimentation, where you're in real-time changing things, and you still do need to understand the forecasting models 'cause even if you experiment and you get to steady state, where you are optimizing you still have to forecast.

10:55 Brent Williams: That's right. I think that's a really good point. The fundamentals still are important.

11:01 Matt Waller: Yes.

11:04 Brent Williams: But as you said, what things are affecting the other things, understanding, well how do you measure that? How do you know that? I think those are some really important things. We continue to teach those things, so we do continue to teach the models. I think they're important. I think, one, we do also have to think about the level of investment required to access the kind of technology right now that we're talking about. It's a large investment and lots of companies aren't gonna make that investment right now, so some of the fundamentals continue to be important.

11:38 Matt Waller: The other thing I think is it seems to me that people still need to understand the broad supply chain logistics concepts because I think there is a lot of people who are actually doing forecasting, for operational purposes, but don't understand the interactions between say. Inventory returns, days of supply, gross margin return inventory investment, and even return on investment. How does this all work together?

12:14 Brent Williams: I think that's really important as well. I think having a foundation that says If I make this business decision as it relates to forecasting or inventory management, or logistics decisions ultimately, how does that roll up to the financial metrics that the company is focused on? I think that's critical for anybody in one of those roles to be able to at minimum to be able to tell that story when they make decisions, how does it ultimately roll up and affect, financial metrics?

12:46 Matt Waller: Yeah, and really just to understand what is it they're forecasting? They're forecasting demand. People need to understand the components...

12:50 Brent Williams: They do.

12:50 Matt Waller: Of demand and what are those components from a forecasting perspective?

12:50 Brent Williams: When I think about it, I think the first ones that come to mind, is the level, think about it as the base demand, but then we tend to Take a given piece of sales data and try to lack of a better word. Pull apart, right? When I think about it graphically, that's the way I think about it where there's a base demand. There's trend so, are sales growing? Are they declining over time? And then of course you mentioned day of the week or whether we mentioned those things that were in some way they're alluding to seasonality, which is another component of demand, some of the other things that managers are often really interested in, are things like outliers lumpiness in data. So take outliers. The question often is, Well, is this spike or is this large decrease? Is that an outlier? Is it an anomaly? Is there a business reason for it? Should I include it in my forecast going forward? Those are still important questions that managers are grappling with.

14:08 Matt Waller: I think too, if I'm forecasting the demand for bottled water at an e-commerce site, suppose I'm an e-commerce retailer, and people are crazy enough to order bottled water that way. The transportation costs would be high. But let's suppose they do that, forecasting that is very different than forecasting the demand for say Halloween costumes, because bottled water is something that is ongoing, you're always gonna have demand for bottled water but you're only gonna have demand for... Well, 99% of your demand for Halloween costumes will be before Halloween.

15:00 Brent Williams: Yeah. Yeah, that's an interesting example, right? So, it is more... I think it's more challenging that the Halloween example. I think an even more challenging example, of that might be, let's imagine cold and flu medications that you buy over the counter. Well you and I, we both know that that's driven the initial increase in sales for a given season is gonna be driven by when people start getting sick, and that's not always perfectly predictable, right? Various factors are affecting that. The weather is probably affecting that.

15:39 Matt Waller: The effectiveness of the flu vaccine.

15:43 Brent Williams: Exactly. So that's an example of, you know you're gonna have most of your demand within a given period of time, a pretty short period of time. The real challenge is you don't know exactly when that's gonna stop, you also don't know going into a season, you don't know how long that season is gonna persist and how strong the demand is gonna be. Is this gonna be a really bad flu season, is it gonna be a moderate flu season? So that's an even more challenging in my mind, seasonal business to forecast.

16:19 Matt Waller: You might, it's easy for someone to think. Okay, well all of this artificial intelligence is coming out. Why will people need to understand forecasting? But I don't believe that there's ever gonna be a time people can be completely divorced of understanding forecasting, because what happens is if you just use mathematical models, you wind up over-fitting them. That's the problem right now, you wind up, you take a historical data set, if you have a polynomial that is high enough degree, you can fit it perfectly. Any data set. And it's gonna do terrible than just using the average going forward.

17:09 Brent Williams: Yes, you're gonna fit it really well, within what we would call the in-sample data.

17:13 Matt Waller: Yes.

17:14 Brent Williams: And then once you get in the out-of-sample or the future if you will, it's not going to perform as well.

17:19 Matt Waller: Yeah, I speculate with all this talk about artificial intelligence and machine learning and cognitive computing that there's gonna be a lot of people fall for false claims about what software can do and what technology can do to forecast.

17:41 Brent Williams: This is why I think... And I know I keep going back to this, I think that's why still understanding the fundamentals are important. I feel like when you have that base of knowledge you can make these judgments or at least a more informed opinion on how a particular software system is gonna perform for your business if you understand how these things work. It could be as simple as really understanding. Well, how does a regression model work, and how do you put causal variables into that in a way that's gonna predict sales? Understanding those basics, just give you enough inside, I think to be able to make those judgements.

18:21 Matt Waller: Yeah, that's a good point. I think forecasting too is getting more difficult. And the reason I say that is that there's several reasons for this: If everyone but only one kind of car, we could forecast demand on a daily basis fairly accurately, we could use a model of it, looks at trend and seasonality and level and we could do pretty darn good, but people buy lots of different kinds of cars. And what we see in every industry is a proliferation of variety and increasing customization and all of that makes the forecasting of any one item more complicated. I was listening to something this morning about a new type of shoe I had never heard of, and it's taken off, it's doing the shoe is doing really well. And so, I Googled it and I found it and it looks really comfortable, very reasonably priced. And a lot of people have them and I don't know, I wanna say they've only been around a short time, and I wanna say their sales are like at 50 million or something like that.

19:53 Brent Williams: Wow.

19:55 Matt Waller: But they introduced it on the internet, but the ability to create a new brand and get it out there now is easier. You don't need to be a big company with a huge advertising budget anymore. And so, the big brands are constantly going to be seeing more error in their forecast, because they never know when these scrappy little entrepreneurs are gonna come up with something that nips at their heels.

20:24 Brent Williams: Yup, yeah, so you've got the demand spread out over more unique items, SKUs if you will, more SKUs, stock keeping units in the world in general. So you've got the demand spread out over that. And you're probably increasing the variability because of that additional variety.

20:41 Matt Waller: Yes.

20:42 Brent Williams: That's right, that's a great point.

20:45 Matt Waller: Thanks for listening to today's episode of The Be EPIC podcast from the Walton College. You can find us on Google SoundCloud, iTunes, or look for us wherever you find your podcast, be sure to subscribe, and rate us, you can find current and past episodes by searching Be EPIC podcast, one word that's B-E-E-P-I-C podcast. And now, Be EPIC.

[music]