Have You Ever Wondered How Modern-Day Distribution Analytics Developed?
Admittedly, the author is somewhat of a geography nerd, having graduated with a degree in Economic geography. He then spent his entire 40-year career mastering retail distribution analytics and strategies for financial services providers. This review of some of the underpinnings of this science/art provides a layman’s explanation of some complex calculations and why they are important for branch planners today.
To paraphrase Gene Roddenberry, Growth: the final frontier. These are the voyages of every bank or credit union. Their five-year mission: to explore strange new markets, to seek out new growth opportunities…
Today, every bank and credit union is trying to formulate a plan for greater revenue growth. For firms with existing branch and ATM networks, the questions are:
Are branches still relevant?
Should I go all digital like the neobanks?
Should I stay with branches since my customers still like to have a place to go for face-to-face advice?
If you have a branch network of any size, it’s hard to walk away from it. Your customers or members have been trained how to use your branches and ATMs. They still want them, and yes, they want online and mobile channels too. The key question if you already have a branch network isn’t “Should I keep it?”, but rather “How do I maximize its effectiveness Answering that question requires a basic understanding of the underpinnings of location theory and the field of economic geography.
In this article, I’ll explain some of the things I’ve learned in the 80,000+ hours I’ve spent studying and implementing these concepts and how TerraStrat applies them to branch/ATM planning. Let’s start with some history.
One of the early influencers in this science was German geographer Walter Christaller, who introduced the concept of “Central-Place Theory” for retailing in 1933. Christaller postulated that the primary purpose of any settlement or market town is to provide goods and services to the surrounding market area. Christaller was building off work done in the 1800s by another German, Johann von Thunen, considered the grandfather of spatial economics, who based his theories on the study of agricultural patterns in Europe.
The larger cities that provide the most goods are called higher-order central places. Smaller communities that offer fewer goods and services have small market areas and are considered lower-order central places. These lower-order central places are more focused on items purchased with greater frequency. Higher-order places are typically surrounded by lower-ordered central places. This hierarchical approach can continue to multiple layers with second-level communities surrounded by third-level towns, and so on.
The basic premise of this theory is that consumers are willing to travel farther to conduct more valuable, infrequent transactions. A good example is that a consumer probably won’t drive long distances to get a quart of milk but may to buy a car. In banking, this theory translates into some predictable patterns. If a consumer needs cash, they want to find the nearest ATM. But they are willing to drive a greater distance to get a deal on a mortgage as they value that transaction more highly. The shift to online mortgages and digital signatures may change this behavior in the future.
As you might expect, there are fewer higher-order places, and they tend to be surrounded by a greater number of lower-order places. You may recognize this concept in use today when you hear people discuss a “Hub and Spoke” distribution model for branches.
Expansion of Knowledge
One constraint to Christaller’s theory is that it was built on the assumption that central places are distributed over a flat geography, with a constant in population density and purchasing power, and no barriers to movement. The theory went on to state that movement is uniformly easy in any direction, transportation costs are linear, and consumers act rationally to minimize transportation costs by visiting the nearest location offering the desired good or service. Under Christaller’s model, the reach or trade area of any central place was a circle since travel was uniform in any direction. Anyone who’s ever visited a big city, like Pittsburgh, PA, knows there are flaws with some of these assumptions.
Other geographers followed up to expand or prove Christaller’s work. German economist August Lösch built on Christaller’s work by using a bottoms-up approach with a system of lowest-order (self-sufficient) farms, which were regularly distributed in a triangular-hexagonal pattern. He also illustrated how some central places develop into richer areas than others.
Edward Ullman introduced central-place theory to American scholars in 1941. Geographers have tested its validity and, in Iowa and Wisconsin, have come closest to meeting Christaller’s theoretical assumptions.
All this brings us to William J. Reilly, an American economist and professor who applied one of Isaac Newton’s laws to the retail trade. Newton’s law of gravitation states that “any particle of matter in the universe attracts any other with a force varying directly as the product of the masses and inversely as the square of the distance between them.” In other words, a larger object will generate more gravitational pull than a smaller one, and that gravitational pull weakens the further you are away from it.
Reilly’s “Law of Retail Gravitation” developed in 1931 borrows Newton’s theory and applies it to retailing. According to Reilly's law, customers are willing to travel longer distances to larger retail centers given the higher attraction they present. In Reilly's formulation, attractiveness of the retail center becomes the surrogate for size (mass) in the physical law of gravity.
Like the work of Von Thunen and Cristaller, Reilly’s law presumes the geography of the area is flat without any rivers, roads, or mountains to alter a consumer's decision of where to travel to buy goods. In Reilly’s approach, mass was represented by the size of the retail center.
Some 30 years later, Texas geographer and professor David Huff modified Reilly’s law by calculating gravity-based probabilities for consumers at each origin location patronizing each store in the store dataset. Huff’s spatial interaction model uses these probabilities (built on multi-factor analyses) to generate probability surfaces and market areas for each store in the study area. Huff’s model was originally designed for the placement of large malls. TerraStrat can fine-tune the model’s flexibility to analyze hub and spoke strategies, adjusting for urbanicity, segment strategy, and even branch formats.
To account for differences in the attractiveness of one store relative to others, a measure of store utility such as sales volume, number of products in inventory, square footage of sales floor, store parcel size, or gross leasable area is used in conjunction with the distance measure. In other words, Huff broadened the definition of a retail center’s “attractiveness” to be more than just its size. Huff’s approach allows you to add potential new store locations as input to determine new sales potential as well as the probabilities of consumers patronizing the new store instead of other stores. This fact is key in new branch planning.
If you made it this far into the article, I thank you. Now let’s look at how TerraStrat applies the concepts to branch/ATM planning.
Structural Applications
When you think about markets, it’s important to understand their basic structure. Generally, there is a large “core” area that offers anything you want. That core area is surrounded by smaller communities that offer most of what you want, but for some things, you’ll need to go to the core. Those secondary communities are surrounded by even smaller towns, suburbs, or exurbs that offer even fewer goods and services, typically those purchased most frequently. We see this model carried out in how the Census Bureau defines MSAs or CBSAs, with a larger core city surrounded by smaller counties that have some inherent connection to the core. A “hub and spoke model” is exactly that concept.
The Reilly/Huff models explains the impact of distance on the probability that consumers will visit a specific retail (or branch) location. The further a consumer is from a retail location the less likely they will visit it. Those distances vary depending upon the “gravity” or “attractiveness” of the retail location. These models help determine the optimal distances between retail branch locations to cover a market.
Combining these two approaches, you can now design an optimal branch/ATM network. The greatest challenge is the sheer size of the databases necessary to perform these analyses. That’s where artificial intelligence and machine learning programs come into play. These modern tools were designed to analyze large scale databases to perform complex calculations and find patterns. These programs can calculate the attractiveness of any retail location, relative to any other retail location in the market, and can be filtered for pretty much any target segment(s).
TerraStrat leverages the latest street network data to account for traffic patterns and barrier to travel time. We also analyze the retail density and quality to make sure the model is calibrated uniquely to each market. Here is how an integrated branch and remote ATM network might look in a mature market, where you have developed a large customer base.
Waldo Tobler, a geography and cartography professor at UC Santa Barbara, coined the first law of geography in 1970, that “everything is related to everything else, but near things are more related than distant things.” Some 50 years ago that was hard to measure, but modern tools make it possible today.
Final Thoughts
Keep in mind that the goal of gravity modeling is to minimize the number of branches while maximizing coverage, optimal for thin branch networks.
If you are thinking about transforming your branch and ATM network for a post-Covid19 world, you’ll want to keep these models in mind as they help explain how consumers make choices of where to bank. Thanks for letting my geographer nerd out!
