Retail Sales Models: An Interview with Dr.Richard Fenker

By Joe Francica

Retail Sales Model: An interview with Dr.Richard Fenker

Directions Magazine conducted a brief interview with Dr.Richard Fenker, Vice President and Chief Software Architect of Market Insite Group.Dr.Fenker, formerly the founder of Tangram Corporation, which recently merged with Market Insite Group, has developed sales forecasting systems for over 100 major US corporations and Tangram's models have been used to help select approximately 100,000 retail locations for clients such as McDonald's, Blockbuster, Brinker International, Lowe's Companies, Pep Boys, Applebee's, Eckerd Corp., Valvoline Instant Oil Change and others.

JF: What is the status of retail models in the industry?

RF: From my perspective, almost no one seems to understand all the issues that relate to modeling.And the biggest of these issues, is that any modeling that I've every seen, only scratches the surface of how people are actually behaving.And it doesn't matter whether its Huff or a regression model or a neural network model.Any model that I've every seen seems to be dealing with a "micro topic" within the whole scheme of trying understanding human behavior.

JF: Huff believes that the models do not factor the human behavior on the competitive side.Many of the models in use employ sales figures from the store in which the model was being created and did not factor the relationship to competitive environment.Huff did not believe that that model could be calibrated accurately without sales from the competition.

RF: Well, now I would disagree with that. That's the way a 'demand-oriented modeler' thinks about modeling.You divide the world into those that buy your products and those that buy your competitor's products.If you look at the history of the Huff model and its application within all the firms that do gravity modeling as a principal base for modeling such as MPSI, you get some version of an approach which takes a roughly fixed demand and divides that demand into a set of rules.I was in Europe about one year ago at a conference essentially on gravity modeling and Huff modeling.What was fascinating was that I watched the papers that were presented at the conference was that almost every single paper was about an adjustment to one of these classic models to explain some very normal aberration that was simply caused by the fact that the model didn't fit much of what was happening in the real world.So, they were saying that 'demand changes so I must make some adjustment to the real world.' The model does not work very well in a situation which fluid where most of most my business is driven by transients, such as at convenience stores. There are some pretty interesting issues out there and none of these classical models, none of them, even come close to explaining the complexity of behavior except in very unusual situations.The only time I have seen a successful a Huff sort of model or gravity approach to modeling, really work well, and probably Thompson has noticed this too, is if I take a Lowe's or Home Depot or that kind of business, and I look at the behavior or people who live in a rural town, where demand is controlled by who lives around the store, then I see tremendous relationships.As soon as I move any modeling into real world situations in markets where I've got city stores, suburban stores, mall stores, lots of fluid behavior within the market, all of a sudden the gravity components no longer work particularly well.

As we began to test gravity models over the last 15 years, we could see that it worked in some cases but in most cases, it had limited utility.The approach we adopted in the early '90s was to begin to pull from the shelf of all possible models, and to use each model selectively to describe certain kinds of behavior.If you look at our development of models as a company, we became more and more complex because we had more and more tools in our toolset.What we learned is how to apply these tools in fluid way depending on the type of problem we were presented with.

JF: So, you have categorized the models in a way that fits a customer or fits an industry?
RF: Actually, both.We would start with a generic situation.Based on my experience, I would say that there are about twenty different types of models that seem to explain most retail sales situations.I would also say that which model you use and where will depend on the retailer.Then, depending on the types of circumstances you are modeling, within the world, I would say that to model any retailer, you need all twenty models.

JF: Why? Because of all the demographic or geography characteristics that could be factored?

RF: That's right.Now, stretch that one step further.Because of the complexity of who human beings use the world, part of that is demographics, part of that is segmentation data, part of that is because I have different sources of demand, when I go from one situation in the world, when I go from an urban location to a suburban location, etc., the rules change.And so even simple variables, such as demographics...well which demographics matter to me? In some situations, I care a lot about who lives around me because they are my customers.And others, I don't really care! Because the draw is not based on who lives around me.It is based on the retail dynamic that is associated with my store.Take a good mall store.All malls for example, to some degree, all depend on affluence.So, if I have affluence within 10 miles of a mall, it is going to be a better mall.But to large degree, the people who live within 2 or 3 miles of it don't mean anything to it.

Another quick example: If I am in a competitive retail environment, my site characteristics, such as visibility and access, matter a great deal to me.On the other hand, as I become less competitive, more of a destination or move more into suburban areas where I am implicitly a destination, the importance of those characteristics to me really falls off.So my modeling system has to understand it all.It has to understand how gravity feeds work; it has to understand how the rules change each time I move to a different situation; it has to understand when pattern recognition models work, and so forth.

JF: Is Cheetah always a custom implementation?

RF: That was part of the reason we merged with Market Insite Group.We are developing a whole suite of web products that provide real time customization.

Published Friday, May 17th, 2002

Written by Joe Francica

If you liked this article subscribe to our bimonthly newsletter...stay informed on the latest geospatial technology

Sign up

© 2017 Directions Media. All Rights Reserved.