Geography And Demographics
In this chapter, the author will look at the role that geographic and demographic data (sometimes mushed into the term "geodemographics") can play in may areas of retail strategy, from site selection to direct marketing.
A caveat: these techniques are "as much a science as it is an art." (EN: this is a terse way of saying that there is not a magic wand - the practice uses data to inform or confirm intuition, which remains a significant factor.)
A second caveat: seasonality is often a significant factor in consumer behavior (see previous chapter for an extended discussion), and geographic location is a determining factor in seasonality. For example, people buy different goods in winter than in summer - but location determines the impact of the seasons: when "winter" begins and ends and how much it impacts the consumer depends on location (it's a completely different time of year in Australia than in England and in the US, southern states have shorter and milder winters). The author intends to omit that issue to focus strictly on location, but it is considerable.
He also forewarns that this chapter provides "detailed and technical information" to illustrate the advanced techniques many retailers are leveraging - the technical details are a necessity to understanding the process and its potential value.
Understanding The Tools And The Data Requirements
The various tools and utilities used by analysis are generally referred to as "GIS" for geographic information systems - though many of them also leverage demographic data and as such are not purely geographic.
Many such tools were borrowed from other disciplines, specifically engineering. (EN: My understanding it was originally military, so even engineering is an adoption.) A GIS tool was a way of mapping terrain, creating a three-dimensional grid (longitude, latitude, and elevation) for a task such as building a highway, where the important data was the lay of the land and objects that would impeded construction (hills and valleys, rivers, forests, etc.)
When people are included as a data factor - plotted on the map in the same manner as a forest of trees, and additional data about the people are cross-referenced (their age, income, ethnicity, etc.) - it becomes easy to see how GIS can help to analyze the customers in a given geographic area in a way that is more specific and detailed than census tables.
There are many different types of mapping tools, from the commercial ones found on Web sites (using Google Maps to find a shop close to your home) to specialized tools licensed by marketers: MapInfo, ESRI, ArcInfo, among others, that provide access to vast amounts of data and provide tools to research, explore, and model.
A caution is that there is a great deal of data, and the problem has become one of too much rather than too little. To avoid being cluttered and overwhelmed, it is necessary to sort out inessential factors to be able to make a meaningful analysis based on the most influential ones.
How Geography Fits Into Retail
For brick-and-mortar retail, geography is of critical importance: the success of a store depends on the proximity of a sufficient number of customers who will visit the store, and the location of multiple stores, yours and your competitors, largely determine how the customers in a given trade area will be divided. In this sense, GIS analysis is critical.
The most basic first step is geocoding the location of a store, which is a very straightforward process: the digital equivalent of putting a pin on a map.
From there, the question becomes what other data to include, as there is a vast array of choices, which are often included in "layers" (think of a clear plastic sheet overlaid on a map).
- Data about population, the physical location of peoples' homes.
- The locations of competitive stores (whether other chains or your own stores that unintentionally compete for customers)
- The road networks that connect the store to its customers and suppliers
Each of these factors can be cross-referenced to an enormous amount of data.
- Population information can be cross-referenced to demographic data such as age, income, gender, ethnicity, lifestyles factors, and other details.
- A store location can reflect its size, number of employees, types of merchandise, volume of sales, etc.
- Road networks can be referenced to traffic levels and patterns that individuals take between many different locations.
The GIS systems can be cross-referenced to literally any form of data that can be coordinated to a physical location. The author makes specific mention of TIGER data (Topologically Integrated Geographic Encoding and Referencing) sourced from the US Census Bureau, which provides several more layers of data.
Each of these data sources can provide a separate layer that can be displayed on a map produced by a GIS tool. The level of complexity is astounding, as there are potentially hundreds or thousands of layers that can be added to a map.
Retail Data: Internal Data Collection
All of your internal data can also be loaded into a GIS system: anything that can be correlated to a physical location, most often indicated by ZIP code, can add specificity to your analysis.
For example, you could plot the locations of households where a given promotion has been sent, then time-sequence it against transaction data that includes customer ZIP codes from the transaction data, to have a clearer sense of how the promotion led to a change in sales.
In past years, retailers had little information about customers at the point of sale - but much has changed. Many customers pay by credit card, a transaction that is includes information about the billing ZIP code, and customers have been amenable to loyalty card programs, that enable each transaction to be associated to a specific customer (whose address is in your database).
Retailers are frequently curious about the source of their sales and the factors that define their most profitable customers - and without GIS, it's something of a general and vague idea. With GIS, it can become extremely specific and reliable.
Retail Trade Areas: Differing Methods for Debate
The concept of trade areas was previously discussed: a map of the locations from which a given store draws its customers. This is important in store retail because 60% to 80% of business comes from a small geographic area. Marketing can attempt to increase the size of the trade area, but promotion is most effective within its existing borders. For example, you wouldn't want to pay to send a sales brochure to people who are so distant that they are unlikely to visit your store at all.
The core data that drives the model of a trade area is transactions: the ZIP code or address of customers who currently shop at a store, based on the assumption that their neighbors are similar to themselves and are likely to become customers if approached with the right offer.
A traditional and inaccurate method is merely calculating the distance between a customer's home and the store, and drawing rings around the store, in a bulls-eye target fashion. This identifies customers as living a certain number of miles from the store, in any direction - the conclusion being that people who live within five miles of the store are its most likely customers, five to ten miles are less likely to shop there, ad so on.
The chief problem with this method is it assumes customers to be evenly dispersed in the "circles" around the store ... consider the problem of a store located on the coast, which assumes that there are an equal number of customers who live five miles into the ocean as five miles inland. It also fails to account for population density in the areas, natural barriers such as rivers, expressways, and the like, and the layout of roads in an area (a customer on the other side of a rive may be one mile away, but must drive down the shore to a bridge, and back up the other shore to your store).
Another method, heavily used by grocery stores, is the gravity model: it's better than the distance model, but still has limitations. This model plots the locations of customers and judges the distance to the stores they shop - it is still a radius-based design, but based on the location of customers rather than the location of the store.
A more common an accurate model is to use the location and shape of actual ZIP code areas rather than arbitrary rings or circles. It is not as geometrically clean (there ill be odd-shaped areas, gaps, and places in which a more distant location is "hotter" than a closer one), but it is highly accurate. Where distance is necessary, the centroids (geographic middle of an area) can be calculated for each ZIP code territory.
The calculation can be made even more accurate if a store has a credit card or loyalty card program, in that the exact location of each customer's home, based on billing address, can be plotted on a map such that the clustering of customers can be visually appreciated.
The author speaks of a client who wanted to understand where credit card applications and new accounts were coming from, set realistic store-level goals for acquisitions, and understand the kind of customer (income, education, home value).
Based on the ZIP codes associated to accounts, the author was able to map them using GIS (question one), consider the population of those areas in terms of number of potential customers (question two), and cross-reference the ZIP codes against market data to provide demographics of people who lived in certain ZIP codes (question three)
Planning Based on Geodemographic Analyses
A vast amount of data can be gathered using the techniques described in this chapter, but data for data's sake yields no profit. It must be used to inform business decisions. It can do so in a myriad of ways, and it's not possible to detail every possible use - but by way of a case study, the author will illustrate ways in which the data might be used.
The challenge in this case was to identify opportunities for credit card growth - specifically, for each store location, to forecast the maximum population base that could likely be sold a specific credit card. There were a few different kinds of "card product" that could be offered.
There were some limitations - specifically that market basket data was not available, and the analyst had to base his study on the addresses (ZIP) of current cardholders.
The first step of the project involved gathering all applications for 18 months and assigning them to specific stores (10 mile radius for standard stores, 25 for superstores) by ZIP code, classifying them as approved or unapproved. Next, considering demographics (income, education, home value, etc.) to determine the number of creditworthy households that did not have the card in each store's area. This was done for each type of card product to determine which cards would fare best in each area.
(EN: The author is a bit sketch on how the analysis was done, but I expect it's a regression analyst that determines, based on the various factors, the number of cards people in a given ZIP code can be reasonably expected to obtain based on the past 18 months behavior.)
The author mentions that prior to this work, the store did a single-factor analysis, and accepted that the number of credit products a store should sell would be based on the sales revenue (the more sales you have the more credit cards you should be able to sell) without accounting for other factors (a store in a poor area might have high sales volume, all cash sales, from customers who aren't creditworthy.)
He also describes the different card products: a private-label card (used in the store only), a full-function card (a "normal" visa card that is store branded) and an all-occasion card (no clear idea what this is, maybe prepaid gift).
Switching channels to describe one of the tools he created: a "sales penetration map" that created a display for each ZIP code, indicating how many households that are eligible for a card presently have one, as a data layer on a map.
It's immediately relevant because it's difficult to sell more cards to an area in which most people already have them. It can also be overlaid with trade areas and other layers to see where correlations exist between card ownership and shopping behavior.
Additional Uses of GIS Tools
The author asserts that even the "best and most successful" operations that leverage GIS tools are still using less than half their potential.
Most tend to leverage GIS on a distinct purpose or need, using it to consider and address a specific problem, discarding it after the project has concluded. Very few use it as a method to monitor on an ongoing basis or inform decisions on the store level.
Particularly for credit products, it is worthwhile to consider trade areas based on tender type (cash, check, credit, or debit) in order to track changes over time and identify very specific targets where there is high potential. You cannot assume that once a person has a card that they will actually use it.
And while the author's experience, hence his focus, is marketing credit products, the same techniques can be applied to other situations. For example, logistics can be very well informed if it knows how much of a given product people are going to buy, based on past behavior, and make inventory decisions accordingly.
Another example: a store that sells home-improvement goods can hone its inventory by knowing the types and ages of homes in the area where it is located, which can predict the kinds of maintenance and materials those homeowners are likely to need.