A client recently asked AccuData to develop a predictive model to help target a large national direct mail and email campaign. Notice the request was to develop “a” predictive model - we actually wound up building ten, and despite the incremental cost of building ten vs. one, they drove sky-high ROI for the campaign.

Why did we do this and why did it deliver economically?

Predictive models built for a national prospect universe assume that individuals or households with similar characteristics behave the same way in each MSA. They don’t. Many marketers approach this problem by asking their analytics team to include a geographic factor, essentially asking their team to make sure the predictive model includes a factor for geography. But this is often a mistake.

First, in many cases the geographic factor is weak compared to other key elements (e.g., demos, lifestyle factors). In this case the geographic factor falls out of the model. Alternatively the geographic factor could be so strong that other important variables fall out of the model.

What works?

In our experience the best way to address the challenge is to build distinct models for each key market. In this way a weak, but potentially important geographic factor, is incorporated into your targeting. At the same time a strong geographic factor is incorporated because you will develop models at the MSA level (or some other geo level).

If you are a large marketer you must fight for every 1/100th of a percent of response. In our experience, building localized models for key markets delivers critical competitive advantage.