Our return rate says 68%. Is the product broken?

> WHAT WE WALKED INTO

The customer-service team at a baby-category partner reported a 68% return rate on a newly launched product. At 68% you stop selling the product, and that conversation was about to happen. Amazon’s own page for the same product showed a rate in the mid-twenties, which was the first clue: two systems disagreeing that badly means one of them is dividing the wrong things. The 68% was that week’s processed returns over that week’s sales, two populations in one ratio. We rebuilt it from the raw order and return exports, every returned unit joined to the order that sold it. The new launch returns at about 27%. The US account as a whole returns at 14.7%.

> THE MOMENT · THE REBUILD, AS LOGGED

> STEP 01

Traced the 68% to its formula

One week’s processed returns over the same week’s sales. Returns arrive weeks after the sale, so a slow sales week against a normal returns week reads as a catastrophe, and a strong sales week reads as a miracle. Neither is information.

> STEP 02

Measured the lag

The median return lands 26 days after its order; just 1.9% arrive inside 7 days. That lag is the whole reason a weekly ratio cannot work, because the returns in any given week belong to sales from a month ago.

> STEP 03

Rebuilt on matched cohorts

February to May orders, with time to mature, every return joined to its order. The new launch: about 27%, matching what Amazon’s own page showed. The US account: 14.7%, about 25,900 units sold and about 3,800 returned. The Canadian account: 19.2%. Three numbers, three questions answered, where the 68% had answered none.

> STEP 04

Kept the real problem

About 27% on a newly launched product is high, and the early read of customer return reasons points at fit and use-case mismatch: listing problems before product problems. That is a reason to work the listing, not to stop selling. The rebuild also surfaced about 900 more units refunded than physically returned, a gap with several possible causes (returnless refunds, returns in transit, reimbursements) that we logged rather than named.

> THE NUMBERS, DRAWN FROM THIS FILE ONLY

> WHAT CHANGED · FLIP IT YOURSELF