Farmers Insurance's Shiv Gupta on How Predictive Models Can Improve Direct Mail ROI
Considering Farmers Insurance Group has been in business since 1928, it definitely has a lot of historical information about its auto, home and life insurance policy holders. The Los Angeles-based company that began with the intention of insuring the vehicles of rural farmers decided it was time to build on that data and take an intelligent look into the future through predictive modeling aimed at retaining customers.
The process already has yielded a 14 percent increase in Farmers' return on investment for its direct mail efforts. Shiv Gupta, Farmers' director of insight and innovations, says the insurance provider decided to tap a resource it had been using for years—Cary, N.C.-based software provider SAS Institute. About two years ago, Farmers began gathering data and using SAS Analytics to build the predictive model that came online nine months later.
"Our objective was to develop a predictive model for retention and lifetime value of every customer we have in our book of business," Gupta says. "And we have about 7 million households, and the transactions are about 20-fold of that. ... [The predictive model] provided us not only the ability to cleanse large volumes of data quickly and easily, restructure the data so that we could use it for analysis, but then also seamlessly go into the analysis stage and complete that analysis and batch work so that we could actually develop lifetime value numbers for every single household in Farmers—historically, five years back and today."
Gupta explains more of Farmers' thinking:
Target Marketing: How did Farmers Insurance reorganize its direct marketing strategy to have predictive modeling be the foundation?
Shiv Gupta: So, when we initially started asking the question, "Who are our best customers?" we were able to identify them rather easily, looking at history. Then we said, "Can we predict, day one, who our best customers will be?" And that's when the predictive modeling started to come in. ... Using the data that we collected when they walked in the door, we were able to develop a very strong model that helped us separate those customers who were going to be with us for a significantly lengthy amount of time and those who were likely to detrite quickly. Using that internal model and that data, we were able to map to external data that was provided to us by mailing vendors for our own customers. And through that mapping, we were able to develop a mailing campaign model. Obviously, that mailing campaign model doesn't provide as strong of a separation as our internal model. But, nonetheless, it's powerful enough to where we've been able to get significant improvement in ROI ...
TM: Once Farmers Insurance learned that 20 percent of its customers provided 80 percent of its revenue, how did the company design its predictive model to maximize this segment's customer lifetime value?
SG: Once we found out that we had significant separation among our customers ... it was like getting a lottery ticket. You had to cash it in. ... What that allowed us to do is to develop, one, an acquisition strategy ... Can I start asking those customers better questions about why they came to Farmers? Why they choose insurance or how they go about buying insurance? So we conducted several studies: attitudinal, segmentation studies and buying behavior studies, and also media habit studies. And, based on those studies, we were able to understand ... how they go about buying insurance and what they watch, in terms of media, so that we can advertise better. Furthermore, our copy testing has also become aligned towards the high- value customer, so that everything that we go put out there has been tested and resonates well with the high-value customer. ... So here's the way to look at the ROI and how that increased. Before, we would look at one major factor, which was a probability to respond to our direct mail. And that would be the decision to mail or not mail. ... If that percentage was hitting our threshold, we'd mail. With the introduction of customer lifetime value, which really is primarily driven by retention and then also driven by the number of relationships that customer has with us, we were able to put in a second dimension. And that second dimension now said, "How much revenue or how much value do we expect this customer to provide us once we acquire them?" And then our analysis no longer became the probability to respond, but also the probability to respond and the potential revenue that comes in. So now, when we have lower probabilities to respond, but we [feel] that the revenue [justifies] taking that risk, we … mail to that group. And there, by adding that group, we found that we had an increase in ROI ...
TM: Demographic and behavioral characteristics of the most valued customers are included in this predictive model. How does that help Farmers Insurance target these consumers?
SG: We do not use demographic variables for our modeling aside from the ones that we are allowed to collect for rating [or policy pricing levels]. ... We are able to now understand behaviorally how these customers go about buying insurance ... Do they first check on the Internet and get a quote on the Internet? Do they get referrals from friends? ... And how is that different from the lower lifetime value segments? ... How do they go about consuming media, and where should we be advertising? ... Furthermore, in the near future, we can start discussing, "Can we design products to meet their needs better? Are there things that they would like their insurance company to provide that is unique to that segment that isn't being provided today? And can we find ways to service them better?"
TM: How does predictive modeling aid in retention?
SG: One thing we've done, when we looked at our customer lifetime value model, we said we wanted to measure what the intrinsic stickiness is of a customer. ... So we wanted to understand, "What if the effect, by measuring intrinsic value first, we were actually able to separate out the importance of acquiring the right type of customer and then the importance of servicing the right type of customer?" And that was very powerful in that we found out that, for us, 70 percent of the gain really came from acquiring the right customer. ... Only 20 [percent] to 30 percent of the loyalty that a customer exhibits is actually coming from the service that we're providing. ... And that means a lot of economic implications for our company, in that we are now much more careful in terms of spending ... on service in a more appropriate way ...