The unit on your desk works. That tells you about that unit, and the complaints are about others.
For cases we accept: the failure mode from the complaints, a test that matches the symptom, the affected population by batch, and containment with records.
A real person reads your case and replies within about an hour.
ASA Compliance Group operates Amazon Sellers Appeal. On defect files we isolate the failure mode, test it against the affected lot, and evidence the containment in Seller Central.
Include the reported failures and return reasons, the batches or dates involved, and any testing already done. Four short steps. Or Shamosh reads every intake.
Or Shamosh · Founder and lead strategist
5,500+ enforcement cases since 2016. Every intake is personally assessed before acceptance.
Send what you have, including prior DIY or agency appeals. We say what is viable in writing before you commit.
Questionnaire
If the questionnaire does not load, open intake on the account page.
* Across ASA’s accepted and fully managed enforcement cases (all notice types we take on), not defect matters alone. Details under Proof.
Customers reported that the product did not work as represented. It failed early, arrived damaged, could not be assembled or used, was missing a component, or behaved inconsistently. The notice describes the customer experience, and the cause behind it is what the response has to establish.
That cause can be any of six things, and they are not interchangeable: an isolated damaged unit, a batch problem, a design or component fault, unclear instructions, a compatibility gap, or transit damage. Defaulting to improved quality control answers none of them, which is why it reads as an absence of investigation.
The sequence that produces a usable answer:
What exactly failed, in whose hands, under what conditions, on which batch. Taken from the complaints rather than assumed.
A protocol that reproduces that symptom on that lot, with pass and fail defined before testing starts.
What now stops it, bounded to the population the evidence supports, with records showing it operates.
One branch leaves this page entirely. Where the reported failure could cause injury, fire, electric shock, choking, poisoning or structural collapse, it is a safety matter even if nobody was hurt, and it belongs with product safety compliance under Amazon's safety framework. Amazon's defect prevention guidance covers the rest.
The same case appears under several names in Seller Central, which is worth recognising early: Voice of the Customer, NCX, product condition customer complaints, and a listing flagged as at risk. They are views onto the same underlying complaint data rather than separate problems. Where the condition complaint is about age rather than handling, the shelf-life requirements are set out on the expired products page.
The complaint text usually distinguishes these better than any internal record does, because it describes what happened rather than what should have.
Dead on arrival, or non-functional on first use. The cleanest signal to investigate, because the failure is unambiguous and usually reproducible if you have the right unit.
Get returned units rather than pulling fresh stock. A unit from the complained-about lot is evidence; a unit from current stock is a different sample answering a different question.
The hardest to test, because the failure needs time or cycles to appear and a bench test rarely reproduces it. This is where a passing sample misleads most reliably.
The product is complete and does not fit, or is incomplete and cannot be used. Often this is not a manufacturing problem at all: the listing created an expectation the product was never going to meet.
Where the product conforms and the page describes it wrongly, the correction is to the page, and the lane is incorrect detail page. Correcting content is only legitimate here when the product itself is genuinely conforming.
The product left intact and arrived broken. Test the packaging against the route rather than the product against the specification, and look at the damage pattern: consistent damage in one place is a packaging finding, scattered damage is a handling one.
Where units arrive intact but visibly handled, the complaint may be about condition rather than function, which is used sold as new and a different file.
Complaints cluster on one production run, one supplier change, or one inbound date. This is the most tractable finding available, because it bounds the population precisely and makes containment provable.
It only appears if the complaint data is mapped against lot, supplier, production and inbound dates. Complaint volume alone hides it, which is one more reason a rate is the wrong thing to argue about. Where the complaints turn out to describe a different product entirely, the lane is wrong item complaints.
Reviewers want a cause the evidence supports and a control that exists. Most rejected responses supply a reassurance instead.
A low complaint percentage offered as the defence. There is no published safe number, and severity and product risk weigh independently of volume.
Misuse asserted without instructions, use conditions, complaint evidence or testing that supports it.
A unit that works, presented as disproving the complaints, with no match on lot, model, condition or test method.
A letter saying the product is not defective. That is an input to an investigation, not the investigation.
Inventory replaced without establishing whether the issue was design, manufacturing, batch, transit, instructions or listing expectation.
A hazard handled as ordinary quality. Injury potential changes the obligations, whether or not anyone was hurt.
Symptom, then test, then control. Skipping to the control is what makes a response read as a promise.
Map the complaints and returns by product, code, order, batch or lot, date, fulfilment channel and reported failure mode. The shape of that data answers the first real question, which is whether this is concentrated or spread. A cluster on one lot and a scatter across two years need different responses.
Then state the failure mode in testable terms. Not does not work, but what stopped working, when, under what use, with what preceding it. That sentence is what the test protocol is built from, and a vague version of it produces a test that cannot fail.
The protocol has to reproduce the complained-about condition, on units from the complained-about population, with criteria fixed in advance:
A test designed after the results are known is not evidence, and it is visible as such. So is a protocol that could only ever pass.
Bound the population from the data: which lots, which dates, which supplier runs, which fulfilment paths. Then state which cause the evidence actually supports, and say what it excludes. Naming what was ruled out and how is what separates an investigation from an assertion.
Supplier input belongs here as one source among several. Their analysis, batch records and any corrective action they took are useful, and they do not substitute for evidence you hold yourself, because Amazon is asking what you established.
Quarantine, stop-sale, removal, replacement, revised inspection, packaging change, corrected instructions or listing content, and monitoring afterwards. Each with what changed, which population it covers, when it started, who owns it, and what the current state is.
A drafted corrective action, a supplier promise and a planned test are all legitimate things to be doing and none of them is an implemented control. Say which state each item is in.
If no testing has happened yet, that is fine and worth saying. Designing the test around the reported symptom is most of the value, and a test run before the symptom was defined usually has to be repeated anyway.
Scoped historical outcomes on cases we accepted and fully managed. One seller voice below; deeper galleries live on account reinstatement.
Population: historical cases ASA accepted and fully managed (we prepared the compliance response and ran follow-ups in Seller Central). The figure combines reinstatement and restoration work across notice types in that managed population, not defect matters alone. Declined intakes and self-serve template buyers are not in the rate.
Success: reinstatement or restoration of the privileges Amazon had restricted for that file (account, listing, verification, funds, or related, depending on the case). Amazon decides every outcome; the rate is not a guarantee for future cases.
Review: the underlying records were independently reviewed. Verified as of .
More methodology context and galleries: account reinstatement proof.
“In less than 24 hours after the appeal, my account was reactivated with all products available for sale.”
More outcomes and video proof: account reinstatement proof.
For sellers who already have the Performance Notification. Broader questions: full FAQ library.
If Amazon removed a listing after defective-product complaints, start with the questionnaire.
Listing removed rather than mis-described: listing reinstatement. Other notice names: all suspension types.
