Building Custom Models for Potential Mailers
If you own a large active or inactive customer file and you’re not building custom list rental models for potential list users then you’re probably missing a huge opportunity to increase your list rental income.
What’s huge? Let’s say your customer file consists of 1 million names and you have 20 mailers that are using your hotline names, but not mailing your active file. It’s a pretty good bet you could make 10% to 20% of your active file work for them, at least twice a year. Assuming a list rental fee of $90 per thousand this means that you’relosing (1 million names; times 10% to 20%; times 2 usages; times $90 per thousand; times 20 mailers) between $360,000 to $720,000 in gross list rental income. Call it $500,000 to be conservative and adjust that number by the number of names on your file and your list rental rates and you can get a pretty good idea of the sums involved.
What’s more, building custom models for individual mailers is not a new idea. This database application has a proven track record, so it’s a pretty safe bet. The usual way this is done is to start with a match rate model and move to a test mailing, if the match rate model works.
So what’s a match rate model and what does “works” mean? In a match rate model you arrange to have your file matched against the file of the potential mailer.
Let’s say the match rate is 5% (5% of the mailer’s names match the names on your file). Now the question is, could you build a model to find the mailer’s names on your file, based on all the information you have about your customers? The way you go about this is by building a logistic regression model, the same kind of modeling technique that’s used to build response models, except that there’s no mailing and no response. Instead of a response rate we have a match rate. When you build a response model, using logistic regression, what you arrive at is a probability of response for each person mailed.
In match rate modeling what you wind up with is the probability of each person on your file being on the mailer’s file. If the model is a complete disaster, each person’s probability will be equal to the match rate. If the model is real good there will be a large spread between the probabilities.
For example, say the match rate is 5%. In the top decile the match rate might be 9% and in the bottom decile 1.3%.
Based on these results the mailer may decide to test mail any decile that has a match rate greater than, say, 125% of the average match rate. This choice of 125% is arbitrary, but the idea is that since we already know that the mailer can’t mail a random sample of our file, he is looking for a portion of our file where he expects the response rate to be significantly higher than the average. Remember, the premise behind match rate modeling is that names with a higher than average match rate probability will also have a higher than average probability of response when actually sent the mailer’s offer.
Given these criteria (a match rate 25% greater than average), together we would design a test mailing to our customers in deciles one to four of the match rate model, eliminating from the test population all names that already matched our file.
Say the number of names we’ve chosen to mail is 70,000. The logic behind this number is that we would like the mailing to produce about 2,000 responders and our best guess is that the mailing into these deciles will pull about 3% (3% of 70,000 = 2,100 responses).
If the mailer requires a 3% response rate to break even, the immediate reaction is that the mailer should mail into all four deciles of the file, based on the match rate model. But a closer examination of the model, built on the test mailing, indicates that mailing into all four deciles means that the mailer would be mailing into identifiable segments predicted to respond at less than 3%.
The bottom line resulting from this exercise is shown in the table on page 90. From the original population of 1 million names, the list owner has discovered at least 147,949 (about 15%) of his file that will work for a mailer that has a 5% match rate.
Based on a set of other assumptions about the value of a customer and the cost of promotion, the bottom line for the mailer is a contribution of $143,993 each time the model is used. For the list owner (assuming a $90 per thousand list rental fee) this translates to $13,315 in gross list rental fees per use. Assuming the mailer can use the model four times a year, we’re up to $53,262 per year per mailer. If we assume that there are 20 such mailers, well, pretty soon we’re talking about some serious improvements to the list owner’s bottom line.
If you would like a copy of this Lotus model, send a self-addressed floppy mailing envelope to David Shepard Associates Inc., 2 Micole Court, Dix Hills, NY 11746.
