Mail Model of the Year
It’s a truism not to pitch consumers who don’t want to hear from you. U.S. Bank, the nation’s fifth-largest commercial bank, had the reverse problem—spending money and resources mailing individuals who would’ve purchased anyway. The bank believed this group accounted for as much as 50 percent of a given marketing campaign and nearly $400,000 of pointless expense. The bank wanted to maximize return on investment and reduce waste, using models to predict campaign uplift—yet its models were falling short on both points.
“We had a custom-built response product for years,” explains Jane Muelhaupt, the bank’s vice president of consumer direct. Stability is a gauge of how the accuracy of a model built for one period carries over to the next, but these models had no monthly or quarterly consistency, Muelhaupt says, which meant they could never identify someone as a “sure thing.”
“The performance of our direct-mail population over the control was very volatile—at times, negligible,” she says. The bank’s direct-mail efforts would succeed one month, only to backfire the next.
In 2007, U.S. Bank implemented Portrait Software’s Uplift Optimizer to remedy these woes—but Muelhaupt was skeptical. “At first, we weren’t convinced [that Uplift would work],” she recalls. “We put [it] through a trial process over about a three-month period to test the application before we made the investment.” Even after pulling the trigger, Muelhaupt rolled out the software slowly, beginning with the bank’s credit products, expanding to the deposit side in early 2009.
Accuracy relies on the ability to prepare significant amounts of data and potential variables. “Uplift models are somewhat fragile and don’t have a long shelf life,” Muelhaupt says. U.S. Bank’s previous models, though, were so cumbersome that they typically only changed once a year, sometimes less often. Uplift enables U.S. Bank to constantly refresh its models using an extensive testing methodology comprising almost three years of results. Creating a model used to require four to six weeks; now it takes a couple of days.
With the old models, average uplift ran between 5 and 10 basis points; with the new ones, uplift is between 30 and 50 basis points. “Sounds like a small number, but it makes a big difference,” Muelhaupt says. “It moves you from unprofitable or marginally profitable to profitable.”
Campaigns for deposit products have improved 108 percent year-over-year. One campaign, targeting direct-deposit accounts, reduced mail volume by 32 percent and still increased incremental sales by 73 percent. Credit products fared even better: a 189 percent year-over-year increase in sales while mail volume dropped by 20 percent. What’s more, the sheer number of respondents has increased in each instance now that U.S. Bank is able to remove what the company calls “sleeping dogs”—the 10 percent of pre-Uplift recipients who would respond negatively.
The success earned U.S. Bank the Portrait-user Uplift Award and Portrait itself an award from the Direct Marketing Association. Despite the glory, Muelhaupt says she’s aiming higher. “This is a process. You need to build your models. You need to invest in mailing and getting great data sets,” she says. “We’ve made quantifiable improvements in our own performance so the goal is to keep moving.”
After deploying Portrait Software’s Uplift Optimizer, U.S. Bank was able to:
- end mailings to the 20–50 percent of individuals who would have purchased anyway, saving $400,000 per campaign;
- increase uplift from between 5 basis points and 10 basis points to between 30 basis points and 50 basis points;
- increase sales on deposit products by 108 percent;
- garner a 73 percent increase in sales (and a 32 percent decrease in mailings) on a campaign for direct deposits; and
- increase year-over-year sales on its credit business by 189 percent, while reducing mailings by 20 percent.