A refund is money, a return is a box arriving at a warehouse. Why Amazon's two numbers disagree, how to read reasons and dispositions, and what to ask Claude.
Ask two people on the same Amazon team for last month's return rate and you will often get two different numbers. Both can be right. One counted refunds, the other counted returns, and on Amazon those are different events that happen on different days, in different reports, and sometimes not at all for the same order.
This post explains the difference, the reasons the two numbers drift apart, how to read return reasons and dispositions, and the questions to ask your AI so that the answer you get is the one you meant.
The returns report carries the operational details: the return reason, the request status, and for FBA the customer's comment and the condition Amazon's warehouse assigned to the unit. The refund line carries none of that. It only carries the money.
None of this is an error in the data. It is the reason to decide which question you are asking before you compare anything.
Pick one per question and say which one in the prompt. A common mistake is to sum the quantity on settlement refund lines and call it refunded units. In DataDoe's settlement table that field is filled from the returns reports, so for refunded units on the sales timeline use the sales and traffic data instead.
The return reason is what the customer said: unwanted item, not as described, defective, arrived damaged and so on. The disposition is what Amazon's warehouse found when the unit arrived: sellable, customer damaged, defective, carrier damaged or damaged. Disposition exists only for FBA returns.
The gap between the two is where the useful findings are:
One trap: FBA and FBM use different reason codes. FBA reports UNWANTED_ITEM where FBM reports CR-UNWANTED_ITEM, and NOT_AS_DESCRIBED on FBA matches AMZ-PG-BAD-DESC on FBM. Map the equivalents before you combine the two channels, or the same reason shows up as two.
The refund is only part of it. A unit that comes back as customer damaged or defective cannot be sold again, so its product cost is lost on top of the refund. In DataDoe, each return row carries the product and shipping cost you uploaded, valued as of the return date. In the profit tables, only units that come back sellable are treated as restockable. Every other disposition counts as a lost unit.
That makes a simple ranking possible: not "which ASIN has the most returns" but "which ASIN loses the most money to returns once unsellable units are counted at cost".
With DataDoe connected to Claude or another AI client, returns, settlements, sales and product costs sit in one schema, and your AI joins them when you ask. Questions that work:
For the reasons data itself and what patterns to look for, see reading Amazon's returns and refund reports. For the money side, see reconciling Amazon settlement reports.
Usually timing and refunds without a return. Amazon refunds many orders before the unit arrives, and some units never come back. Compare a longer window, such as a quarter, and match on order ID rather than on month.
No. The warehouse condition exists only for FBA returns. FBM returns have a request status, a carrier and label details instead.
It shows up in the FBA returns report with the status "Reimbursed", meaning Amazon paid you for the unit instead of returning it to your stock. The payment itself is in the reimbursement report.