A fast refund is sometimes the cheapest outcome

Give the refund whenever winning would cost more than the fee, which for a small job is usually. A refund is not an admission; it is a purchase of your time, your review score and your dispute record for the price of one fee.

The question is never whether you were right. It is whether being right is worth what winning will cost, and for a small job the answer is usually no.

The arithmetic that decides it

Put the numbers next to each other before you argue.

On Stripe, for instance, the dispute received fee is never returned, even when you win; only the separate countered fee comes back.

That last row is the one people miss. Platforms track dispute rates, not refund rates, and a rising dispute rate can affect ranking, payout timing or account standing regardless of whether you won any of them. A refund you granted is an outcome; a dispute you won is still a dispute that happened.

So the threshold is not "was I wrong". It is roughly: if the fee is smaller than an hour of your time, refund and move on unless something else is at stake.

Refund immediately in these cases

You cannot deliver. Illness, a scheduling collapse, a job that turned out to be outside what you actually do. Refund the moment you know, not on the deadline. The buyer who is refunded on day one is mildly inconvenienced; the buyer refunded on day four has lost their own deadline too.

You delivered something faulty and cannot fix it fast. A corrupt file, an inaudible recording, a job done against the wrong brief. Offer the fix first, refund if the fix would be slow.

The buyer is distressed rather than difficult. Somebody who clearly expected something else and is upset about it will not be argued into satisfaction. The refund ends it in one message.

The fee is small and the conversation is heating up. This is the pure cost calculation. Cheap jobs generate a disproportionate share of trouble for reasons set out in why the bottom tier attracts the hardest buyers, and the fastest way to stop paying that tax on an individual job is to hand the money back.

Hold the line in these cases

The work was delivered to the agreed scope and the buyer simply changed their mind. A delivered assessment cannot be returned. If you refund on regret you are offering a free trial, and word of that travels.

The refund request arrives with a demand attached. "Refund me or I will leave a one-star review" is not a refund request, it is a threat, and platforms treat it as one. Do not pay it. Report it, quote the confirmation, and let the process run.

The request is really about the conclusion. Somebody unhappy with what you found is not unhappy with the service. This is the case where refunding teaches you nothing and teaches them that a paid assessment can be unwound by objecting to it.

It is the fourth request from the same buyer. Patterns are data. A buyer with a habit is a buyer you decline next time.

In all four, answer once, calmly, in three sentences, and then stop responding to repetition. Escalating word count is what turns a refusal into a dispute.

Partial refunds, and why they are usually a trap

A partial refund feels like the reasonable middle. In practice it satisfies nobody: the buyer who wanted their money back now has an open grievance with a smaller number attached, and you have conceded that the work was defective while still charging for it.

There are two cases where a partial genuinely fits. One is where you delivered part of a multi-part job - two of three assessments, the written half of a written-and-recorded pair - and the split is arithmetic rather than judgement. The other is a late delivery where the work was fine, and you are discounting the failure rather than the deliverable.

Say which of the two it is when you offer it. An unexplained partial reads as haggling.

The mechanics, which are not what you assume

Money moving back is not symmetrical with money moving forward.

Processing charges are frequently not returned when the payment is, so a refunded job can leave you out of pocket by the card fee - the detail differs by platform and is covered in what actually happens to the fee on a refund. Refunds issued after a payout can be clawed back from your next balance, which is how a refund in a quiet month becomes a negative balance. And a refund does not always close a chargeback, because a buyer can refund and dispute, and the card network does not know the two are the same job. Stripe's refund documentation flags this case, recommending that a business accept or challenge a dispute instead of refunding, to avoid duplicate reimbursements.

Keep the refund conversation on the platform and in writing for exactly that reason. The record of what you offered is what stops a resolved refund becoming an unresolved dispute, and what a moderator actually weighs is the same evidence in both processes.

One last thing worth naming. Refund requests cluster around jobs where the buyer expected a number to mean something it does not, which is a definitional gap rather than a quality failure - what a score can and cannot claim is the clearest statement of it, and a sentence of framing at delivery prevents more refunds than any amount of care afterwards. The same buyers frequently arrive having already run an automated tool and expecting your conclusion to match it; the honest account of what automated accuracy means is why it sometimes will not. On the commissioning side, refunds are seen as unremarkable and are asked for less often than earners fear - the buyer-side account of how a finished review is received is a useful corrective to the assumption that everyone is looking for their money back.

Whether a refund is yours to issue at all depends on the platform's order flow, and on some it is a moderator action rather than a button. Rate Cock's judges page is where its own version of that is set out.