A refunded job can still cost you money

A refund feels like an undo, and it is not one. The buyer's money goes back; several of the charges that were taken along the way often do not, and the difference lands on you.

The size of the gap is small per job and it is the reason a run of refunds costs more than the fees it appears to cost.

The three layers, and what each does on reversal

Commission. Usually returned in full on a genuine refund, because the platform's cut is calculated on a completed sale and the sale is no longer completed. Not universal, and worth confirming in the terms rather than assuming.

Processing. This is the one that bites. Many processors keep the transaction charge on a refund, on the reasoning that the work of moving the money was done either way. Stripe's refund documentation is explicit: its processing fees from the original transaction aren't returned. Where that applies, a fully refunded job returns the buyer's full payment and leaves the processing charge as your cost.

Chargeback fees. A different category entirely. If the buyer goes to their bank rather than asking you, a fixed fee is typically applied on top, and it is usually charged whether or not the dispute is decided in your favour. Stripe's UK pricing lists £20 for each dispute received, and its dispute documentation says that fee is never returned unless a contract states otherwise. That asymmetry is why chargeback rates and what drives them matters more than the raw refund rate.

What that looks like in practice

An illustration, not quoted rates.

The job took your time, produced nothing, and cost you a few units of currency on top. Across a format that refunds at any meaningful rate, that is a line item most earners never put in their spreadsheet.

Flat components make this worse at low prices, for the same reason set out in minimum viable price when fees are flat: a fixed charge is a small fraction of a large job and a painful fraction of a small one.

Partial refunds are usually the better instrument

A partial refund returns some of the payment and typically leaves the transaction intact, which means the processing charge is already accounted for and you keep proportionate value for work actually done.

It is also the honest answer in most real disputes, because most real disputes are about part of the job rather than all of it. When to offer one at all is covered in the post on giving a refund, and the short version is that speed matters more than the amount.

Why offering the refund first is usually cheaper

Given the fee asymmetry, a refund you initiate costs you the processing charge. A chargeback costs you the processing charge, the chargeback fee, the hours of evidence-gathering, and a metric on your account that platforms watch.

So the arithmetic favours conceding early on anything genuinely contested, which is uncomfortable and correct. What evidence actually wins a dispute is worth knowing for the cases that are worth fighting, and those are fewer than instinct suggests.

Price it in rather than reacting to it

If a format refunds more often than your others, its true take-home is lower than its price implies, and the only way to see that is to record refunds against format rather than against the month. That is one extra column, and it is the column that decides which formats survive a review of take-home per format.

Buyers rarely understand any of this, and the buyer-side account of how a commission is expected to go is a better guide to what they think they are owed than your terms page is. Where a dispute turns on whether the work met a stated standard, a documented method is the difference between evidence and opinion, which is the argument made on the measurement side. A fair share of refunds start with a buyer who expected what an automated tool gives instantly - what those tools produce is worth knowing so your scope can rule it out before payment.

For the specific refund window, who bears which charge, and how partial refunds are processed on the platform, Rate Cock's terms for judges is the binding version, and it is the one that decides the arithmetic above.