For more than two decades, multimedia publisher Rodale has been relying on modeling as a way to unite and maximize its very large customer database. Today, the company has more than 26 million active customers on file, and its modeling program is even more powerful than ever, enabling Rodale to send out hundreds of millions of promotions each year—including more than 140 million direct mail efforts—in support of its constantly evolving repertoire of magazines, books and DVDs.
Here, Todd Leiser, the Emmaus, Pa.-based organization’s vice president of database marketing, pauses to discuss what it takes to keep Rodale modeling strong.
TG: How has modeling enhanced Rodale’s direct marketing?
TL: Rodale first started regression modeling in the early ’80s, primarily to improve response rates of direct mail promotions to existing customers. Today, we continue to expand modeling applications into new media and the variables we predict. …
[Currently] 65 percent of all outbound direct mail is coming from our regression models. The predictable nature of the models enables us to roll out products with a higher degree of accuracy, allowing us to expand the reach of marketing programs (by modeling new internal and external universes). With the models in place, marketers also can spend more of their time on other aspects of the programs.
TG: What are some of the key variables you use?
TL: They are similar to any customer segmentation scheme—recency, profitability, long-term value … prior purchase activity. We expand the internal data with external overlay, focusing on behavioral/transactional traits, then general descriptors (age, gender, income). More important than ever is the channel of the last purchase. We [used to] run a more homogenous business, where direct mail dominated our activity and alternative media was only a small portion. [Our] source mix has changed over the years [and] that variable is becoming more of a driver. If someone is an online customer, [he or she] isn’t going to be as responsive to direct mail, and we have to incorporate that into our models.
Our business is primarily a free-trial, bill-me business, so prior payment activity, prior return activity [and] prior collection activity is always important in our models.
TG: How often do you update your models?
TL: We build new models for every new product. … We also build new models for different subsets of the customer base for those products. The model will be built when the product is new, and we’ll update it with a back test around the second rollout, six to nine months later. Typically, the life cycle of our products is two to three years, so that model won’t be updated unless that product looks like it’s going to have … continued life beyond that point or if there’s a significant creative or offer change.
For products with longer life spans, magazines for example, those models will be updated twice a year. We are always mailing back tests or regression test samples when we do direct mail, and we incorporate the new results with the older data.
TG: Where has modeling had the biggest impact for you?
TL: On the retention side. Our magazines have very large customer audiences. Prevention, as an example, has 3.1 million active subscribers … and there’s several unique customer segments within that audience. We built the marketing database and the modeling environment in order to help us sell books and other products to our big mass consumer audience. The biggest effect has been being able to take that large audience of readers—and expires and cancels, which number in the tens of millions—and mine into those segments to find the nuggets of customers who ultimately will respond to a book promotion, pay for a book product and become long-term customers across the organization.
TG: What is required to run a successful modeling program?
TL: If I had to pick two key aspects of the modeling program, they would be the people and the data. On the analytics side, you need folks who love sifting through data to find predictive characteristics and are statistically savvy, yet can translate the complex analytics into a marketing P&L. To be successful, they need to bridge the gap between stats and marketing. On the data side, the modeling program will only be as good as the data that is available (independent and dependent variables). Start with internal transactional activity (as much as possible), and then expand in promotional/contact history, aggregate geographic (ZIP code penetration), external transactional overlay (cooperative data from your industry), individual/household overlay, and the other area-level characteristics (census, cluster codes).
TG: What advice would you offer?
TL: Focus on a developing a few key wins for your marketing departments, and start with the data being used within the business everyday (customer transactional detail).
[From the January 2007 issue of Inside Direct Mail, a sister publication to Target Marketing magazine. To learn more about Inside Direct Mail, visit http://www.insidedirectmail.com ]