Bring 'Em Back
Once built, a reactivation model will predict which of your inactive customers are most likely to respond to an offer. This is achieved by first analyzing your inactive file and finding a subset of those records that have at one point or another reactivated.
The next step is to develop a regression model that discriminates between those two segments. "Those [customers] that reactivated as a result of promotions will have a special fingerprint relative to the group that deactivated and never came back," says Briley.
The model then can be used to score and rank your pool of inactives according to each customer's propensity to reactivate.
Paint a Data Picture
When building a model, says Rubin, you want to look at all behavioral data that existed when customers were active. In addition to RFM-related data, this may include products they bought and from which categories. When combined with other information, geographic and demographic data also can be helpful.
"You start to paint this big picture that tells you what they did when they were active from a purchasing standpoint, where they live, what their households look like, etc., so you can get a feel for what types of things they are potentially buying and what's important to them," says Rubin.
You'll also want to include any other data elements that "might help you understand their level of engagement," Rubin adds. This could include any customer-generated action such as Web site registrations, response to past promotions, call center interactions or e-mail inquiries, as well as any survey data that might pinpoint the reason for defection, such as a move or product dissatisfaction.
Briley also recommends looking at the relationship from "the other side of the fence." For instance, what kind of contact history did the company have with these individuals? Were they involved in frequent promotion campaigns? Were they targeted for thank-you letters? Did they receive special offers?