Plan a send
Measured results come from the test. Margin, email cost and list size are your assumptions. Results cover two weeks after the email.
Each rule against emailing the same number of random customers. Ranges come from 2,000 resamples of the hold-out. The share snaps to the nearest of 10% to 60%, where the ranges were worked out. Eight rules at six shares make 48 comparisons, so one or two can land above zero by luck. Only the frozen pick was chosen in advance.
What each email did for each kind of customer
All 64,000 customers. Each dot is the extra purchase rate the email caused, with its 95% range. Patterns that hold on the hold-out are marked in the report. Most differences here are within chance.
Files
The send list has one row per customer and no personal details. It recommends the men's email for everyone, ranks customers by chance to buy in case you must cut the list, and marks a random 10% to hold back so the next send can be measured.