Amazon settles most FBA losses on its own. Units go missing in a fulfilment centre, units get damaged in handling, customers get refunded for items they never send back — and a lot of the time Amazon locates the stock, charges the customer again, or reimburses you automatically without being asked. What it does not do is tell you about the cases it missed.
The residue is where the money is, and getting at it is almost entirely a question of timing. There are two claim windows, they apply to two different kinds of loss, and they behave in opposite ways. One starts closing the moment the loss is recorded. The other does not open until two months after the event, and a claim filed before it opens gets denied on the merits. Most of the money sellers leave behind is lost to that asymmetry rather than to any difficulty in filing.
Two Windows That Run in Opposite Directions
These are the two categories where a claim depends on a clock rather than on evidence. Amazon publishes both as guidance rather than as a contract, so treat the day counts as the shape of the rule and not as a guarantee:
| Loss type | Clock starts | Window | Failure mode |
|---|---|---|---|
| Lost or damaged in the warehouse | The transaction date of the loss event | Generally within 60 days | Filed too late — the event aged out |
| Customer refunded, item never returned | The refund or replacement date | Generally 60 to 120 days after | Filed too early — denied while the item may still arrive |
Read those two rows together and the practical consequence falls out. For warehouse losses, speed is everything and there is no penalty for looking early. For customer returns, patience is mandatory and the real risk is forgetting to come back. A single monthly habit cannot serve both, which is why reimbursement routines built around "check it once a month" tend to lose money at both ends.
Lost and Damaged: The Window Starts Closing Immediately
Amazon's inventory ledger records what happened to your units: lost, damaged, later found, later reimbursed. None of those columns on its own tells you whether you are owed anything. The number that does is what is left over:
When that resolves to zero or below, Amazon has already made you whole — it either returned the stock or paid for it, and there is nothing to pursue. When it is positive, those units were taken out of your inventory and never accounted for. That is the entire population of warehouse claims, and it is usually a small fraction of the gross loss events, which is why eyeballing the ledger for scary-looking Lost numbers is misleading in both directions.
The second filter is the one people miss. Of those missing units, only the ones whose underlying event falls inside the 60-day window are worth anything. A unit lost eight months ago is a real loss and a real hole in your margin, but it is not a claim, and the work of assembling evidence for it returns nothing. Separating the two is the whole job:
Missing, event inside 60 days
Claimable now. The reference data Amazon asks for — transaction IDs for damaged units, FNSKUs for lost ones — is all in the ledger event.
Missing, event older than 60 days
Gone. Worth knowing for your cost of goods, worth nothing as a claim. Do not spend evidence-gathering time here.
Filing goes through the Fulfillment Center Operations claim page in Seller Central rather than a general support case. Damaged units are submitted with their transaction IDs as the Reference ID, lost units with their FNSKUs. Amazon caps a single submission at 25 references, so an item with more eligible events than that needs splitting across submissions rather than trimming — the references you drop are not recoverable later once they age out.
Customer Return: The Window Opens Late
This category works backwards from the first one, and it is the one that catches experienced sellers.
When you refund a customer or send a replacement, Amazon expects the original item back at a fulfilment centre. If it does not arrive within about 60 days, Amazon will normally charge the customer again and reimburse your account automatically. That automatic path is why filing early fails: for the first two months there is genuinely nothing to claim, because the item may still turn up and the system may still settle it without you. A claim submitted into that period is not early, it is wrong.
So the window opens where the automatic path gives up, and then it closes:
Pending — roughly the first 60 days
Too early. The item may still come back, or Amazon may reimburse it unprompted. Most of these resolve themselves. Do nothing.
Eligible — roughly day 60 to day 120
Claimable, with a countdown. Nothing further happens on its own. An eligible order left alone until day 121 is worth nothing.
That second card is where the losses concentrate. Pending rows are self-correcting and forgiving; eligible rows are neither. They expire silently, there is no notification, and the only visible consequence is a reimbursement number that is quietly smaller than it should have been. Which makes frequency more valuable than thoroughness here: a shallow look every fortnight catches more money than an exhaustive audit twice a year.
Claims go through the Customer Return Claims page, keyed on the Order ID, with the SKU and the number of units that never came back. Compared with warehouse claims the evidence is trivial — the difficulty is entirely in showing up during the window.
Cash or Stock: Only One of Them Is Money
There is a second distinction inside the customer-return category that does not appear in most reimbursement advice at all, and it changes what your recovery number means.
Amazon can settle one of these claims two ways. It can pay you, or it can put a unit back into your sellable inventory. Both are recorded as a reimbursement. Only one of them is cash:
| Settlement | What you receive | Effect on the claim |
|---|---|---|
| Reimbursed in cash | Money into your account | Closed. The unit is paid for. |
| Reimbursed in inventory | A sellable unit back in stock | Closed. You have the goods, not the proceeds. |
| Neither | Nothing | Open. This is what a claim is for. |
Being made whole in stock is a legitimate outcome — you are not owed money for a unit you now have back and can sell again. The problem is arithmetic rather than fairness. If your tracking adds cash reimbursements and inventory reimbursements into one "recovered" figure, that figure overstates the money you actually received, and it does so by an amount you cannot see. A quarter that reads as $3,000 recovered might be $1,900 in cash and $1,100 in returned stock, and only one of those paid for anything.
Two Claims, Timed
Both effects are easier to see on specific numbers. The figures below are illustrative.
A warehouse loss, nine months of history
One SKU, cost basis $11.40 a unit, with a long ledger. Over nine months Amazon recorded 14 units lost and 6 damaged, later found 9 of them, and reimbursed 5:
Half the outstanding value is unreachable, and it was unreachable before anyone looked. That is the cost of a quarterly cadence on this category, and it is invisible if you only ever read the item-level total: $68.40 is the honest measure of what went wrong, $34.20 is the honest measure of what is worth your afternoon.
A customer return, through the corridor
An order for 2 units at $34.99 ships on 3 March and is refunded in full on 18 March. One unit comes back. Neither cash nor stock arrives for the other:
| Date | Status | What the window says | Right action |
|---|---|---|---|
| 18 March | Refund issued | Clock starts | Nothing |
| 1 May | Pending | Opens in 16 days | Nothing — the unit may still arrive |
| 17 May | Eligible | Window opens | File from here |
| 1 June | Eligible | 45 days left | File now |
| 16 July | Expired | Closed | Nothing left to do |
One unit, $34.99 of revenue already refunded, and a 60-day stretch in the middle of the year during which doing nothing is correct, followed by a 60-day stretch during which doing nothing is a write-off. A seller who checked this order in April would have concluded there was no claim. A seller who checked it in August would have been right and too late. Neither of them did anything wrong except look once.
What Filing It Yourself Is Worth
Recovery agencies will do all of this for you and typically charge 15 to 25% of whatever they recover. That is a real service and the percentage is not unreasonable for it — they are paid on results, and if they find nothing you pay nothing.
It is worth knowing the arithmetic before agreeing to it, because the fee scales with your success rather than with their effort:
| Recovered in a year | Fee at 15% | Fee at 25% |
|---|---|---|
| $2,000 | $300 | $500 |
| $5,000 | $750 | $1,250 |
| $12,000 | $1,800 | $3,000 |
| $30,000 | $4,500 | $7,500 |
Set against that, the honest case for doing it yourself is narrower than "save the commission" and rests on one point: for the two categories in this article, the hard part is knowing which cases are claimable and when, not filling in the form. Once a row tells you it is eligible and hands you the Order ID or the transaction IDs, the submission is a few minutes of pasting. The commission buys you the diagnosis more than the labour.
There is also a floor under this decision that has nothing to do with the fee. An agency working on contingency has no reason to pursue a $34 claim, and the long tail of small warehouse losses is where a lot of the aggregate sits. Those are worth filing only if filing is nearly free.
A Routine Built Around the Windows
Because the two clocks run opposite ways, the order you work in matters more than the time you spend. Expiry risk first, value second:
- Customer returns, filtered to eligible, sorted by time remaining. These are the only rows that vanish. Anything inside a couple of weeks of closing gets filed in this sitting, regardless of size. Pending rows get no attention at all — they are not yours to action yet.
- Warehouse losses, filtered to unresolved, sorted by value. Work down from the largest, but check the eligible event counts before opening anything: a big missing total made up of old events is not a claim. Here value ordering is right, because nothing on this list is about to expire in a way you can predict from the item row.
- Batch the submissions, up to 25 references each. Filing six items in one sitting costs far less than six singles across a fortnight, and the cap is per submission rather than per day.
- Reconcile cash separately from stock. When the outcomes land, keep paid-in-cash apart from returned-to-inventory so your recovery figure stays a cash figure. The settlement analyzer is where the cash side shows up against the rest of the payout.
Fortnightly is enough for both. The windows are measured in weeks, so a cadence shorter than that buys nothing, and a cadence longer than a month starts losing eligible customer returns to expiry.