November 2008

Geodemographics: an idea whose time has gone?
We must be careful how we interpret analysis based on geodemographic clusters. Here’s why. By Colin Tener

If you saw something in a museum would that be a sign that perhaps its day had passed?
A few years ago, I had the opportunity to visit one of the Smithsonian museums in Washington and was amused to find a display of geodemographic Prizm clusters. (Okay, so only a museum geek with a marketing focus would find that amusing.) Finally, I thought, someone has put geodemographic systems where they belong—in a showcase of tools that marketers have moved beyond—like the rotary dial telephone. But I was wrong. Like a bad horror movie, these things just won’t die. Vendors still push them as if they were the greatest marketing tool ever.

Now to be clear, we need to distinguish between “raw” geodemographic data (like Statistics Canada’s Census data) and “processed” geodemographic cluster systems, of which there are many across North America. We will discuss the benefits of the “raw” data further on in this note. But for now, let’s focus on the commercially available geodemographic cluster systems.
Statistically based cluster systems seek to reduce large amounts of geodemographic data into manageable and understandable groups. Many use cute and dangerously simplistic names (Shotguns and Pick-ups anyone?) to both mask the underlying complexity of these groups and bring them to life for marketers.

Vendors of these systems claim that geodemographic clusters provide a wealth of information about your customers. But at the risk of belabouring the obvious, they do no such thing. Geodemographics tell you something about your customers’ neighbourhoods. If you think that distinction isn’t important, then think about where you live, and more specifically, your immediate neighbours. Are they like you? Do they share your tastes, interests and aspirations? If so, I’ll surmise that you live in a very insular commune. All the neighbourhoods I’ve ever lived in are a complex mosaic of cultures, attitudes and tastes. No amount of statistical hocus-pocus could result in a single description that fits all the residents. Yet that is exactly what proprietary cluster systems purport to do.

The premise under which the cluster argument is made, that “birds of a feather flock together,” simply isn’t true. Just because people have the same income, education level and even family formation as other people around them doesn’t mean they have the same needs and motivations. And that means we can’t market to them assuming that they do. It also means that we have to be careful how we interpret analysis based on geodemographic clusters.

Smoke and mirrors
A few weeks ago, I was speaking with a marketing expert who has deep knowledge of data-based marketing and cluster systems. He had been talking with a retailer who had done a postal code survey at point of sale. The data was then matched up to a national cluster system to see what could be learned about the retailer’s customers. Initially, the retailer thought the results were very informative, providing valuable insight into who was visiting his stores and where he should focus his marketing efforts. So my contact asked him if all the postal codes analyzed represented unique customers or if a customer could have come in multiple times, which would have over-weighted the cluster code and skewed the results. He then asked whether customer spend was captured for analysis at the same time as postal code. In other words, were high-spending customers getting the same weight in the analysis as low spending ones? On both counts the retailer admitted that he wasn’t sure. Moreover, he realized that these factors would significantly alter the interpretation of the cluster analysis.

Desperate Housewives or desperate vendors?
Another application to which proprietary cluster systems are put that is highly suspect is the attempt to link them to other databases. Consider for example, consumer media habits. I’ve been in client meetings where cluster vendors claimed they could tell us who was watching CBC news at 6 o’clock in Kitsilano in Vancouver or Forrest Hills in Toronto. This amazing media targeting opportunity was made possible by linking cluster data to media surveys via postal code. What the vendors don’t talk about is sample size. With media survey samples of 5,000, 10,000 or even 60,000, it is highly dubious that such claims have any statistical validity whatsoever.

All is not lost
So what is the proper role for geodemographics? First of all, if you have no data at all on customers or prospects except name and address and maybe some basic spending information, then geodemographic data is a place to start. But don’t use proprietary systems, get the “raw” data and perform the analysis yourself. Maybe your customers tend to live in above average neighbourhoods with respect to income. Or maybe income isn’t relevant but family formation is. In either event, you have a toolkit that will help you understand where you can find more potential customers based on dimensions that matter to you rather than being forced to accept the analytical straightjacket imposed by proprietary systems.
Secondly, even where you do have customer data, raw geodemographic information can help. We have often overlaid census data where clients did not have income information or other elements that the census contains. While imperfect in the same sense as clusters, since the data is only at a neighbourhood level, it has nevertheless often been the case that these elements prove to be statistically sound components of a predictive model or customer segmentation system. That said, in over 20 years of such analysis, we have never seen external overlay data be a stronger predictor of customer behaviour than internal data. Your own information will always dominate geodemographic overlay data. But the latter can help.

Geodemographic cluster systems may belong in a museum but geodemographic data is still a viable element in the marketer’s toolkit.

Colin Tener is V.P. Business Development for CVM Marketing Inc., a consulting practice that focuses on the art and science of identifying which customers represent the greatest potential value to your organization and then helping to realize that potential. He can be reached at (416) 572-7682 or colin.tener@cvmmkt.com.