Reading your Amazon returns data properly

There is a file in Seller Central that records, for every unit that came back, what the customer said was wrong with it, what condition it arrived in, and what happened to it afterwards. Thousands of rows of unsolicited product feedback, updated continuously, free. Most brands open it when something goes wrong and never again, because it presents as an accounting export rather than as research. Here is how to work it.

In this article:

• The 30-day lag that stops your return rate flattering you
• Four disposition codes, and why CUSTOMER_DAMAGED pays nothing back even when the item is fine
• The 45-day join that surfaces refunds you were never reimbursed for

Set it to arrive without you

The file is under Reports, then Fulfillment, then FBA Customer Returns in the Customer Concessions section. Seller-fulfilled orders are elsewhere, under Orders and Manage Returns, with different fields that will need joining by hand if you run both channels.

Rather than downloading it, schedule it. Amazon will generate the report automatically at daily, weekly, 14-day or 30-day intervals, and the export reaches back around 35 months. Pull eighteen months on the first run. Anything shorter and you will mistake seasonality for a trend, which is how brands end up rebuilding a listing in February because of a Christmas gifting pattern.

Before anything else, fix the denominator

Every conclusion you draw from this file rests on a return rate, and the return rate most sellers use is wrong in a specific, predictable direction.

Returns divided by orders shipped is the standard calculation and it misses timing. Returns lag sales by weeks, so the most recent month always contains orders that have not finished returning yet. The figure is not just imprecise. It is optimistic, systematically, and it is most optimistic exactly when you are looking at recent performance.

Use returns divided by orders shipped at least thirty days ago, with the lag set by category. Consumer electronics settle quickly. Furniture and apparel commonly run 30 to 45 days.

Expect the number to rise when you make this change, sometimes by several points, and tell whoever reads your reporting before it happens rather than after. Nothing will have deteriorated. You will simply have stopped flattering yourself.

The column that decides who absorbs the loss

Disposition is Amazon's condition assessment, assigned at the fulfilment centre, and it determines ownership, reimbursement eligibility and what happens to the unit next. Four values do most of the work.

SELLABLE goes back into your active inventory and needs nothing from you. DEFECTIVE and CARRIER_DAMAGED are unsellable but may be eligible for reimbursement.

CUSTOMER_DAMAGED is the one to understand properly. The unit stays in your unfulfillable inventory, no reimbursement is issued, and, despite the name, it does not require the item to be broken. Opened packaging qualifies. So a code that sounds like customer abuse frequently describes a customer who took the lid off, and it carries the harshest financial outcome of the four.

Two consequences follow. First, never analyse unsellable as a single bucket, because half of it may be reimbursable and half is not. Second, pull a sample of CUSTOMER_DAMAGED units and physically look at them once a quarter. Sellers do not always receive a complete explanation of each condition call, and repackaging a unit that needed nothing else is the cheapest recovery available to you.

Reason codes: where the signal is and is not

Amazon records 33 distinct return reason codes in this file, selected by the customer at the point of return.

Any individual code on any individual order tells you nothing. People misclick, they pick the first plausible option, they choose "no longer needed" because it requires no explanation. The signal is repetition on one ASIN, and the useful threshold is something like five or more occurrences of the same code on the same product inside ninety days. Below that you are reading noise and will spend a fortnight fixing a listing that was fine.

Above it, the codes point at genuinely different departments. Clustered "not as described" is a listing problem, which means images, bullets, stated dimensions and materials. Recurring "defective" is a supply chain problem, and the response is inspection reports or a supplier conversation rather than copy edits. Rising "arrived damaged" is packaging and prep.

Then do the thing almost nobody does, which is to cross the reason against the disposition. This pairing is where the misdiagnoses live. "No longer needed" arriving SELLABLE is ordinary retail behaviour and there is nothing to fix. "No longer needed" arriving CUSTOMER_DAMAGED at volume is a handling or packaging failure disguised as a demand signal, and reading the reason column on its own will never reveal it.

The join that pays for itself

The returns file records what came back. It does not record what you refunded. Joining the two on order ID produces the only number in this exercise that converts directly into cash.

You need quantity returned, disposition and the LPN from this file, matched against refunds, cash reimbursed, inventory reimbursed, refund date and transaction number from the reimbursement side. Filter to refunds issued more than forty-five days ago with no matching return record and no reimbursement credit.

That list is money that left your account for a unit that never arrived and was never compensated. Nothing in Seller Central assembles it for you, no alert fires, and the claim windows are finite. It is the single highest-return hour of work in this entire article and it needs running monthly, not annually.

Read what customers wrote

The comment field is optional, so it is often empty, and when it is not it contains the only unprompted language your customers will ever give you about the product.

Do not read it row by row. Extract themes monthly, in a session with whoever owns product and merchandising. Fit, sizing, colour accuracy, wrong item, arrived late. One vendor tool renders this as a word cloud of frequent terms, which is a blunt instrument that nonetheless beats reading four hundred rows.

The value here is qualitative and it is real. A reason code tells you a garment did not fit. A comment tells you it was two sizes small, which is a size chart fix rather than a photography fix.

Its blind spots

Three, and they matter more for European sellers than for American ones.

Units refunded without a return never appear here, because no unit was ever received, so your true return rate is higher than this file can show. On cross-border orders into the UK, Germany, France, Italy and Spain, low-value items default to returnless refunds where no domestic return address exists, which means the gap in your data is widest precisely where the losses are concentrated.

The €3 EU import duty is not in any column. On a returned cross-border order it was paid on the way in and it does not come back, so your real cost per return is higher than anything this file will calculate.

And the disposition arrives as a code with no reasoning and no photograph attached, which is why the physical sample check matters.

Two of those three gaps are physical rather than analytical, and they are the part we handle. Local return addresses across nine European markets turn silent returnless write-offs back into recorded returns with reason data attached, and every item is verified in Wrocław within 48 hours with a barcode scan and photo-documented check, so each condition decision has a dated image behind it rather than a code.

Send a ninety-day export and your returnless refund volume, and we will return your corrected return rate and the ASINs with the clearest signals in it.

Sources: Gorilla ROI, Hopted, Intentwise, Amazon Seller Forums, DataDoe, Sellercloud, eComEngine, ReturnPro

Not sure what your current setup is costing you?

Send us your numbers and we'll run them against the new regime: where the duty hits twice, what return freight costs against the goods you recover, and how much is avoidable.