Regression Modeling Turns CareerTrack's DM Campaign From Static to Dynamic

Seminar-style learning company CareerTrack, Boulder, CO, has been a successful direct marketer since its founding in 1982 — but executives decided last year that they wanted more.

The company sends out more than 1 million seminar and tape training catalogs each year. And although an in-house recency-frequency direct marketing program was producing increased annual response rates on the company's quarterly catalogs, the marketing department knew that a more sophisticated database program could increase rates and achieve more.

It was right. A regression modeling-based database marketing program begun in February 1997, with the help of database marketing company Database America Co. Inc., Montvale, NJ, gave CareerTrack a 10 percent increase in revenue, a 7 percent decrease in overall mail quantity and an 18 percent increase in the average order size of mailings this year vs. last year.

“We had been using a recency-frequency program for several years, and while it did increase response, it did not do it to the extent that we would have hoped for,” said marketing director Susan Guerrero. “We wanted to not only increase our revenue, but we wanted to have the ability to reduce our mail quantity as well.”

The companies decided that regression modeling would give CareerTrack a more sophisticated way of looking at its customer data and that this would lead to better mailings.

Regression modeling looks at relationships between responders and relationships between nonresponders and attempts to weed out those that are significant and those that aren't.

CareerTrack took several steps: First, Guerrero and two other staff members took a close look at whom they had mailed to and who had responded to a catalog they had mailed earlier that year.

The company had extensive information on its customers: Its in-house database included names and addresses of customers, previous purchasing activity with CareerTrack, details on the types of seminars and audio/videos they had purchased and how much money they had spent with the company. It also knew the businesses customers worked for and the location of these companies, thanks to one of Database America's SIC databases.

Database America ran a regression model on the information, looking for variables that might suggest future response. As expected, the model came up with locations that had spent more than $300 and had previously purchased audiotapes and videos, for example. It also found people who recently had inquired about a seminar but did not purchase anything and those who had not purchased anything for more than a year.

“Basically, it came up with a variety of combination criteria that ultimately ended up telling us who we should mail to for the next promotion,” Guerrero said.

After three months, Database America returned the scored data, with 10 being the best and 1 being the worst in terms of the material that CareerTrack considered important. In July, CareerTrack tested the model responses against its old recency-frequency-type segmentation by sending out 100,000 September catalogs in both the old and new ways.

In August, responses began coming back — and since the catalog has a 32-week lifecycle, they continued through the end of the year.

“We mailed less, we got more responses at a higher average order size and, therefore, more revenue and more profit,” Guerrero said.

CareerTrack is using the regression model as a regular part of its marketing strategy. Guerrero said she is particularly happy with it because she thinks it is something that most marketers can understand.

“We are at a point in the industry where modeling techniques are getting very complicated, but regression modeling is something that is very easy for me to understand,” Guerrero said. “While humans tend to develop segmentation models based on logical criteria, the regression model pulls together information that is not obvious to someone just looking at the data, which usually brings the best results.”

Related Posts