AAA Auto Club South’s Ernesto Diaz and Kristin Rahn on Efficiently Building Predictive Models
TM: What data did AAA Auto Club South input to build these predictive models?
ED: We are using all of our active members' membership information. We are adding to that demographic information that we have available from our demographic data providers. We're also taking behavior information—so purchases, active policies and those kinds of things that they may have with us. ... There's another aspect of it that we have that I would probably subcategorize under behavior information, like things that they purchased that we call our "relationship index," which is really measuring engagement of members with different product lines.
Kristin Rahn: Well, there's two pieces of it. And the piece that my team is using to do the statistical modeling and the profiling and targeting is fully implemented. ... And then there's another piece that's going to give a quick count and self-serve analytics to the business, and we're just rolling that piece out now.
Right now, we have two main tables that we've pulled together. One is at the household level. And we have 2.5 million records in it. And then we have one at the individual, or customer, level and there's about 4.2 million records in that. And we probably have 200 different fields in each of them. ... We have all kinds of behavioral and purchasing data at the household and customer level. And then we have a lot of demographic data that's external data that we purchase and append. And then we also have all of our marketing preference, opt-in or opt-out, for different business lines ... And with those, we can do pretty much anything we want to do without having to build new data all the time.
TM: How is AAA Auto Club South building segments, then determining how to use them in predictive models?
ED: The aggregate of the deciles would be ... the segments. ... So we're using the top deciles as the segment we want to target. In the case of the attrition model, we're trying to use the middle deciles as the segment that we're trying to pursue. Because the higher deciles would be people that are most likely to renew, anyway. So we want to focus our attention on the ones where we might be at risk of not renewing.