A mathematician is a device for turning coffee into theorems. -Paul Erdos
We have a natural tendency to oversimplify things because it’s easier to make sense of simple things than complex things. The map on the right represents the simplest view of the relationship between a store and its customers. The two-mile radius around the store is a reasonable estimate of the trade area for the store, which means that most of the customers will be inside the circle. If this is true 90% of the time, we don’t care about the details of how customers make their choices. But what if only 70% of the customers actually are in the circle? 60%? 40%? At some point we have to consider a more complex set of factors that determine our true trade area and whether it contains enough potential customers to make a store successful.
Yesterday afternoon I got the urge to get some fresh air and a cup of coffee. I had some reading to do, so I wanted to go to a place where I could sit down and finish the coffee at the shop rather than bring it home. I’m omnivorous when it comes to coffee, meaning that I love any good cup of coffee whether it’s Dunkin’ Donuts, Starbucks, or a local shop. I hate to waste gas and time, so the closer the better. Considering all of this, I’m probably similar to nearly 50% of coffee drinkers in my area (although many people have a STRONG preference for Dunkin’, but that’s a New England thing that weakens as you move west in the US).
On the left is a map of my choices. The closest place is a Dunkin’ (#4) that’s just like most Dunkin’s: comfortable and well lit, good coffee, but not a “cozy” place to read a book, especially since most of my friends go there and would like to chat.
Starbucks has a store about 7 miles to the north(#1) and another one 3 miles to the south(#6). Although I like the one that’s closer, I tend to see people I know and that could distract me from reading. However, the extra 4 miles is a big price to pay for anonymity.
The other places are Heavenly Donuts(#3) and Aroma Café(#5). Heavenly is more like a Dunkin’, but I’m less likely to see someone I know. Aroma Café is more like a Starbucks, but I remember the price of coffee being high and the service was slow (the one time I went there). Besides, the Starbucks to the south (#6) is less than a mile further down the road.
After deep thought, I decided to go to the Starbucks (#6) because I was willing to trade-off the risk of seeing someone I knew for all the other pros and cons of the other places (I knew I could pick a table in the back and face the wall if necessary).
What’s the point of this story? Many consumer decisions are too complex to model with simple assumptions about trade area definition, competitor effects, trip patterns, and distance sensitivity. If I’m going to visit a client by car, I always go to the Dunkin’ closest to my house. If I wanted a breakfast sandwich served to me along with coffee, I would go to Aroma Café. If I wanted to get away from the locals, I would drive to the Dunkin’ or Starbucks in Andover which is 3X further than the closest Dunkin’ and I’d be driving past another Dunkin’ to get there.
Am I overthinking this? It’s true that often when we aggregate data in an area around a store (say a two-mile ring in this case for coffee customers) the details of their decisions average out to give us an overall probability of capturing customers. But that isn’t always true. Dunkin’ #4 is on a main feeder to Interstate 93. Starbucks #1 is in the center of Andover with nice retail shops that draw affluent people who like to drink coffee in connection with shopping. There are three Dunkin’s along this 10 mile stretch (plus a few more in gas stations) and they all do well. If Starbucks put a third store near Dunkin’ #4, what would be the impact on the other two Starbucks? It certainly is more complicated than simply looking at the impact of Starbucks on Starbucks, because the other coffee shops are already capturing much of the potential Starbucks business near the existing stores.
Every chain store operator (retailer, restaurant, service business) must face the complexity of the retail marketplace and develop methods of estimating sales, cannibalization, and competition effects that are as simple as possible, but no simpler (Einstein). At the end of the day, some good maps and demographic reports combined with local market info about competition and consumer preferences will give decision-makers a better basis for real estate planning and site selection than a mathematical model that is partially blind to key factors in the real world.
The key factors can be modeled if known. A trade area does not usually have a fixed position like the Great Wall or the Maginot Line. A trade area varies like the edge of the sea, driven by storm and pulled by the moon.
Have you ever tried Esri’s “Network Analyst” for site selection and to analyze competiton and cannibalization?
Yes, Adam. I was the founder of geoVue and we partnered with ESRI for a number of years. We reviewed Network Analyst but opted for another faster engine that was dedicated to computing shortest paths and didn’t have all the other transportation oriented functionality of Network Analyst. Shortest path calculations are good for measuring pure convenience from point a to point b, but in retail/restaurant modeling it’s even more important to understand the variety of trip patterns such as work to store (e.g. lunch at work), home to shopping to store (coffee while at the mall), and the unpredictable reasons that people trade off convenience for quality, selection, price, and other factors as described in this blog post. I’d like to hear your ideas about this.
I agree that it’s difficult and sometimes impossible to analyze certain factors.
Anyway, Network Analyst also computes shortest paths with time and distance as impedance. It supports multiple stops, various demand/facility points and additional geoprocessing such as routing, service areas or location-allocation. Therefore, it offers great functionality as far as site suitability is concerned. And it will become even faster with the next release of ArcGIS Desktop, 10.1, as far as I heard.
Maybe it’s worth a closer look?
I am aware of all the functionality you mentioned and it’s all good. We built location-allocation models for market planning using the Tietz-Bart vertex substitution algorithm with some modifications to constrain demand and competitive factors to realistic limits. For certain types of chain stores this approach works well and is probably the most accurate method possible. However, for some chains the trip patterns are so complex that they can’t be modeled with enough reliability to be worth the effort, which is VERY LARGE! There are other methods that use GIS to generate data and maps that can guide an experienced analyst or decision-maker to a “good enough” answer with much less time and cost, and the answer is probably as good or better than the output of a location-allocation model. I am interested in seeing if ESRI can speed up these calculations!
One more thing: Network Analyst can handle “barriers” other than the road network such as socioeconomic boundaries that affect consumer behavior and natural or man-made barriers that people simply don’t cross for psychological reasons (rivers, mountains, freeways), even when there is a road or bridge that makes it theoretically possible. This is a very important part of network planning and a powerful feature in Network Analyst.
Jim Stone, CRE Chain Store Advisors Reading, MA 781-608-0488 Blog: http://www.chainstoreadvisors.com Email: Jimstone52@gmail.com