Monday, January 21, 2008

Case Study Monday: Acquisition Direct Mail

At RRW, we spend more than 50% of our time helping our clients acquire new customers. The rest of the time is spent in working on customer marketing (retention, segmentation, etc) and other direct marketing activities for our clients. Of course we spend time running the business (yikes--accounting!) and writing this blog :)

My point--over the years, we've become pretty good at figuring out how to acquire profitable new customers. So, we thought we'd share one of our own success stories today.

Situation Overview
Our client, a leading mortgage lender with aggressive new customer acquisition goals was faced with the following list targeting challenges:
  • They needed to maintain their historic high response rates to their ongoing direct mail campaigns, but at the same time,
  • Dramatically improve their approval rates. Over time, in a saturated refinance market, responsive applicants were increasingly failing to meet credit criteria, or they did not have the right equity requirements.
Solution Review
RRW Consulting recommended a data-centric, analytically based approach that leveraged available realty data as well as custom modeling services. We brought a new science to list selection that allowed this lender to achieve its response goals and yet dramatically improve conversion of loan applicants to customers.

We designed a three-prong data selection approach:

  • We utilized industry-leading sources of homeowner and realty data to ensure that the appropriate equity was available for each prospective homeowner. Specifically, we relied on the “Available Equity” and “Loan to Value” list selections offered by those data compilers that specialize in the compilation of first and second trust deeds, integrated with county tax assessor data. This initial pre-select was the first “cut” in determining that the prospect would ultimately qualify for the loan.
  • We built a Response Model that capitalized on prior prospect performance to predict future response performance. Essentially, we cloned prior responders to this lenders’ campaigns. We built a scorecard based on their demographic and psychographic characteristics that ranked the overall homeowner prospect universe from the most likely to respond to the least likely to respond.
  • Finally, we developed an Approval Model that isolated traits of homeowners likely to be approved for the refinance loan. Since we were not extending a firm offer of credit, we could not utilize individual level credit data, making the task of predicting loan approval especially difficult. However, the model was able to successfully rank order prospects in terms of their likelihood to be approved for the loan. The models’ success can be attributed to the wealth of demographics included in the model, and in particular the inclusion of credit data aggregated to the ZIP+4 level.
Implementation Overview
As can be imagined, the list selection approach quickly became fairly complex, due to the number of realty selections employed combined with the set of response and approval models. We developed a data selection and testing matrix that allowed us to be very responsive to changes in the clients’ goals and fluctuations in the marketplace. For example, if the CEO needed more applications flowing to the call centers, we relied heavier on the response model and less on the approval model. Conversely, if the applicant quality was being questioned, we’d select and mail those applicants more likely to be approved. Finally, we also considered frequency of prospect mailings. We managed the campaigns to reach qualified prospects more frequently than those less likely to respond, or to be approved.

The scientific data selection approach resulted in success. Our client was able to retain their high response levels and also dramatically improved loan approval rates.


Ted Grigg said...

Interesting post Suzanne.

I have one question. With all of the selection criteria you put against this file, and assuming some of your segments yielded the required revenue, were you left with a sizable rollout database?

I'm sure you warn your clients as I do that they can select themselves out of business if they're not careful.

How did you know going in that you would end up with a large enough target universe to keep the client happy?


Suzanne Obermire said...

So true--there has to be a balance between targeting and over-targeting (so that there is a teenie-weenie universe left).

In this case, our mailer was national. We used a list company that led the industry in mortgage/homeowner coverage, and our initial pre-selects (before applying the response/approval models) only weeded out those loans they couldn't make (i.e.: not enough equity). Wherever possible, we included homeowners where data was unknown, so we wouldn't lose important prospects.

Then, depending on mail volumes, we oftentimes used the models to suppress those homeowners who were NEVER going to respond to us. Thus, we didn't cut out 80% of the universe; instead we may have cut out 25%.

In regards to understanding the potential of the universe, it's quite simple: we ran counts! Sometimes, actual counts are not that easy to come by when you're using models. Yet, you can still get a pretty darn accurate estimate of the entire universe, even applying the model.

Again, it's always a struggle to find large enough quantities and yet still target the group you want to. A balancing act, for sure.