A public 1998 direct-mail file, rebuilt as a scored decision and judged once on 96,367 donors the rule never saw. It beat mailing everyone and every simpler cut tried, by 28% to 66%.
A direct-mail manager had to decide how deep to mail a lapsed-donor renewal. The rule was to mail every lapsed donor, at $0.68 a piece. It felt fine because it made money. Mailing everyone netted $0.11 a piece, and that profit hid the half of the file that lost money.
We named the baseline first. Mailing everyone stayed in every comparison, and we registered a summed RFM score as the simple-cut baseline before the test.
We wrote down how this decision behaves in direct mail, and what the score should find, before we opened the outcomes.
We built the rule on the 95,412 donors whose results were known. Each name got two estimates multiplied together: the chance they give, and the gift if they do. Inputs were chosen by rule, 61 of 479 columns.
We froze the rule and wrote down a forecast of $17,285. Then we ran it once on the 96,367 donors it never saw. The file returned $16,926, 2.1% under.
We tried every simpler way we could find. Six no-model selections were scored on the same unseen donors, and one model instead of two was tried across 20 splits. The summed RFM score was registered before the run. The other five cuts were added after it, and the record says so.
We handed back the rule, the memo, the record of every check, and the two test cells for the next drop.
The largest-gift fifth of the file responded at 3.61% with an average gift of $30.71. The smallest-gift fifth responded at 7.13% with $7.59.
So what Rank on expected dollars, because response alone fills the mailing with small gifts.
The 44,988 names below the line lost $6,366 between them, and every tenth of the file below the line lost money.
So what The cut is a gain as long as mail never lowers giving.
At $0.79 a piece, mailing everyone only breaks even, while the rule still nets $11,792 on 37% of the file. This check was added after the run.
So what Recut the depth whenever postage or print moves.
No organization acted on this, because no engagement took place. On a live file the change is a recut list pull: mail a name when its chance of giving times its likely gift covers the piece cost. Two test cells are designed in. A random unmailed cell among the names the rule mails measures how much of their giving the mail caused. A mailed cell from the names the rule cuts, read a year on, says whether they return more than $0.14 a piece beyond the cut names left unmailed, and so whether a later mailing should go deeper.
Every donor was mailed, so the file shows who answered, not what the mail caused. Only a holdout cell can say how much of the $16,926 would have come anyway.
The method transfers. The numbers don't.
The rule was judged on unseen donors from the same June 1997 mailing, not on a later one. The stronger test is to build on one drop and score the next.
We'll tell you what your own history can say about it, and which offering fits. If none does, we'll say that too.
Start a conversation