Bought for the table

The second order from the Ramadan customers is due about now, and it is not coming.

Ramadan filled March this year, so the people who arrived in it reach their first repeat window this week, and the repeat rate on that group comes in as the worst of the year. Set beside the March acquisition number, the two read as a business that bought a great many customers who did not like the product.

That is almost never what happened.

Think about who was buying. Somebody ordering for a table of twelve rather than a household of four. Somebody buying a gift, two days before it was needed, because they were going to a house they could not walk into empty-handed. The basket was large, the delivery date could not move, and the thing was chosen because it would look adequate in front of people whose opinion would be heard later.

None of that describes a person who needs the product again in six weeks.

Retention analysis assumes the buyer is the user. It is built that way. A cohort is a set of people who first bought in the same month, and the month is the only thing they are credited with having in common. So the Ramadan group gets held against the February group, and the February group was buying for itself.

I have watched teams answer that number by rebuilding the product page.

The cut that would have answered it is not the date of the first order. It is the reason for it. Gift or self. One table or one household. Occasion or Tuesday. Most systems already hold enough to guess: a delivery address that is not the billing address, a message field that was filled in, a basket four times the usual size.

The Ramadan customer is a different person wearing the same email address.