A refund that was actually worth taking

Three refund decisions, and the arithmetic behind each

A refund is not a verdict on whether a seller behaved badly. It is a comparison between the money that comes back and the money the return itself takes away, and the second figure is the one people leave out of the sum. The desk keeps completed case files for exactly this reason: the interesting part of a refund is never the complaint, it is the arithmetic that decided whether the complaint was worth filing at all.

All three files below sit on the same rung and started from the same position: money already paid, an item sitting in a warehouse, and no international postage spent yet. What separated them was not the size of the fault but which of the three return paths was still open on the day the decision had to be made. Two of the three buyers were surprised by that, and one of them nearly filed a request that would have cost more than the item.

The desk splits a return into three paths rather than one, because a single figure for the cost of returning hides the case where keeping the item beats every request you could file. That case appears most often on low-value orders, which is also where readers are least likely to look for it. Naming the paths separately is the whole reason the arithmetic becomes visible.

Footwear that failed QC before it was packed

The first file holds two pairs of one model, both inspected at the warehouse before anything was consolidated. The photographs showed a stitching fault at the toe of one pair and a visible glue line along the edge of the other, and the report reached the buyer while both pairs were still on a shelf instead of inside a sealed carton. Nothing had been paid to an international line, so the only money at stake was the goods value.

Because no parcel had been built, the rejection cost nothing beyond the original price. The goods value came back, the pairs went back to the seller, and the real loss was the waiting time on a model that never arrived. There was no postage to argue about, no transit claim to open, and no need to prove what had happened between two borders, which is where most refund cases become expensive.

The lesson in this file is about sequence rather than persuasion. A fault found before packing stays a seller matter, because the item is still inside the seller relationship and the warehouse has not taken on the shipment. The same fault found after the parcel leaves has to be argued as a transit matter, with a different evidence set and a much longer clock, and the two arguments are not interchangeable even when the fault is identical.

A wrong size that the listing never admitted

The second file started with a label. The listing named a size, the buyer ordered it, and what arrived measured closer to the size below. Nothing had been done deliberately wrong, which is what made the case difficult: there was no damage to photograph, only a number that did not match the number on the box, and a seller who could reasonably say the label was the label.

Measurement settled it. The desk runs a footwear tally that records insole length and widest point for each model, three pairs per run, and rows are retired when the three runs disagree by more than six millimetres. In this file the measured insole sat far enough from the labelled size that the gap looked like a property of the model rather than the luck of one pair, and that is a much harder claim to answer with a shrug.

With a measurement attached, the request turned from an opinion into a statement of fact, and the outcome was an exchange rather than a refund. That difference carries a cost that is easy to miss: an exchange pays postage in both directions, so it only wins when the item itself is right and the size is the only thing wrong. Where the item is also wrong, the exchange quietly becomes the most expensive of the three paths.

A parcel that was cheaper to keep than to send back

The third file is the one readers skip past, because it ends without a request being filed. The item was low value, the fault was cosmetic, and the return postage quoted came to a large fraction of what the goods themselves had cost. On paper the buyer had a legitimate complaint and a seller who had already agreed to cover part of the return leg.

Run through the return bench, the keep path priced on the share of the price that could be recovered locally, and that figure came in below the send-back path even after the seller contribution. The buyer kept the item, moved it on locally, and closed the file without filing anything. The refund that was worth taking in this case was the one that never had to be requested.

Nothing about that outcome is a defeat. Keeping carries no request, no window and no dependency on a seller agreeing to anything, which makes it the one path the clock cannot close. Its weakness is that it needs an honest local recovery estimate, and overstating that estimate is the quickest way to make a bad decision look reasonable for a week.

Threads that ran through all three decisions

Read side by side, the three files share more than they differ. The differences lay in which path happened to be available; the similarities lay in what made a path usable at all. Each buyer who got a good outcome had one piece of evidence that did not depend on being believed.

The strongest shared thread was timing relative to the warehouse. Every path that sends something back depends on a handling window the buyer does not control and cannot extend, and in the desk return model two of the three paths stop existing once that window closes. The window is therefore the first line to read, not the last, even though it is almost always the last line readers look up.

1. Evidence first: a photograph for a fault, a measurement for a size, the order reference for a value.
2. One path at a time: filing a rejection and a refund request over the same item usually produces one answer, not two.
3. Window before price: an expired window does not make the cheapest path expensive, it makes it unavailable.
4. International postage is sunk on every path, so choosing between them cannot recover it.

Writing a refund request that can be answered

A request that cannot be answered in one pass is a request that will be answered slowly. The practical aim is to leave the other side nothing to ask: name the item, name the fault in observable terms, give the date it was found, and state the outcome wanted as a choice rather than a paragraph. Requests that read as complaints get replies written as sympathy, and sympathy does not move money.

Messages are useful when the identifiers are already inside them, which means the order reference, the warehouse receipt line, and the photograph or measurement the claim rests on. Everything else in the message is negotiation. The desk records the asking side of this exchange as carefully as the answering side, because a case that stalls is usually stalled at the first message rather than at the fifth.

1. Contact the warehouse team first while the item has not shipped: the seller relationship is still open and the fix is cheaper there.
2. Ask one question per message and make it a choice — refund, exchange, or send back to the seller.
3. Keep the date the fault was found, the supporting photograph or measurement, and every reply, including the reply that says nothing.
4. If the seller agrees to cover return postage, keep the message that says so; it is the only record of who pays the second leg.

Window closure and what it removes from the menu

The handling window does more work in a refund decision than any other single line, and it is also the line readers look up last. Its length is a platform-side policy rather than a carrier promise, so it is whatever the policy states on the day you read it, and the desk records the policy date beside the figure for that reason. A window quoted without a date is not a window, it is a memory.

What the window closes is not the whole conversation. A local decision remains available after it expires, and a claim about transit rather than about the item runs on a different clock altogether. The table separates the three paths by what each counts as cost, how long each takes to complete, and what actually ends it.

The data point behind this note

Return paths are priced on money that does not come back rather than on the item price, which is how the desk return model was split into three paths during batch KB-2026-05; the same pass added the window-closure check after finding that an expired window silently invalidated two of the three paths.

◇KB-2026-05· 2026-09-30 Batch ledger · Challenge a figure

Notes in the same cluster

Related entries and gauges

• What happens when QC rejects an item?
• What can you do when a seller refuses a return?
• Return path bench

  What each path costs from where you are standing now.