By the Onsites AI team · Last updated · 5-minute read
Refund replies are where desks lose the most with the least effort — not by refunding (refunds are cheap and honest) but by improvising: the full refund for one buyer and a fight for another with identical cases, the tone that reads like suspicion, the decline that detonates into a chargeback. The fix is written policy plus a four-rung ladder plus replies drafted to match the rung — and every one of them ends with the same sentence structure: what you get, when it lands, what changed. This guide writes the ladder, the eligibility checks that take minutes, the tone rules for declining without detonating, and the special cases that break desks when unhandled: chargebacks, repeat refunders, goodwill offers. The template library carries the replies; this guide carries the policy behind them.
Rung one — full refund, same day, no argument: our fault (not arrived, damaged, wrong item, mis-billed) or the buyer's clean regret inside the return window (14 days is a fair published line). The reply is short: "Done — [amount] back to [method] today, your bank shows it in [window]. Nothing to return; if the item shows up, bin it or pass it on." The margin argument against rung one loses to arithmetic: the refund is the cost of the story the buyer tells — five-star reviews, repeat orders, and the alternative (the chargeback, the review, the ghosted brand) all cost more. Rung two — partial refund, offered first: shared fault (usage, partial delivery, cosmetic damage accepted), return-cost trade-offs (keep the item, take the difference — the return loses money for both of you). Offer first, sized to the wound: "Keep it as-is if you like and I'll refund the difference today — your call." Rung three — store credit, offered, never swapped: full value plus a point or five ("€50 credit rather than €48 refund") — the desk's margin plays but only as a choice; swapping a refund for credit without asking is how desks get reviews. Rung four — the decline: outside window, used goods, stated policy — the honest no, with the reason in one line and the escalation path offered (see below). The ladder's whole power is consistency: the same fault, the same rung, every time — and the reviews can tell.
Before the rung, three checks — the desk's order data in the thread makes each a lookup. One — the order's life: delivered date, condition reported, the window's math; "bought in March, it's August" is rung four unless the fault is latent (then it is rung one — defects don't expire politely). Two — the account's history: the buyer's orders, their refund history — a first-time buyer with one prior order is rung one at their word; a fourth return in a quarter gets the pattern note (below), not a fight. Three — the cost book: what full refund, partial, credit and return each cost the desk (goods, shipping both ways, the story) — sizing the rung is arithmetic once the ledger is written. The checks take minutes and remove the improvisation that reviews punish; the reply drafted from the rung (the copilot drafts with human review) then carries the desk's voice without its hesitation.
The decline is the hardest message in the library, and the desk writes it with four sentences. One — hear them: "I've read the thread and I see the [frustration/expectation]" — the decline that skips the reading reads. Two — the honest no, once: "The return window closed four weeks ago" — the policy stated plainly, without the apology-for-existing that invites negotiation. Three — the yes that exists: "What I can do: [the option — the repair contact, the goodwill credit at my discretion, the escalation]" — every decline carries a real option or it is just a no in a nicer coat. Four — the path forward: "If you want [owner] to look at it fresh, say the word — one word is enough." The tone rules that surround the four: no suspicion theater ("we've checked with our warehouse team" when you haven't), no policy-pedantry (the buyer doesn't care when the policy was written), never the second "no" for tone's sake (the first is the answer; the second is the detonation), and never the public-correctness play ("as stated in our terms..." — terms are for contracts, threads are for people). The declined buyer who got four honest sentences charges back at a fraction of the rate of the one who got one.
Chargeback incoming: refund immediately, through the ladder's rung one. The fee fight (case preparation, evidence, the waiting, the chargeback rate) costs more than most refunds and reads worse in reviews — the reply: "I see the dispute opened this morning. Before it runs its course: I can resolve it directly today — [the fix]. If that works, close the dispute and it lands as a refund." The desk's memory is the evidence anyway. Repeat refunders: the pattern note, not a fight — the account's history shows four returns; the desk's written pattern is the goodwill conversation ("happy to keep making it right — and let's look at what keeps going wrong, because I'd rather fix the causes") — refund fourth time as first, note the pattern, and let the account data decide rung five. Goodwill: exists, logged, never routine — the €20 credit offered on a case that earned it is a repair; the same credit offered weekly is a budget line, and the desk's ledger knows the difference. The B2B refund: invoice-level, not checkout-level — the corrected invoice (the billing conversation's sibling), the credit note, the terms; same ladder, different documents, and the loop closes in the same thread.
One desk's refund month under the ladder: 23 refund requests. Rung one: 15 (five late deliveries, six defects, four clean regrets inside window) — all refunded same day, average resolution 3 minutes from request to refund, zero follow-up disputes. Rung two: 4 (two keep-the-item-and-the-difference, two return-with-cost-absorbed) — all accepted, no negotiation. Rung three: 2 (credit chosen by buyers, worth the +5% margin play twice). Rung four: 2 decls (a window-closed buyer who took the goodwill credit offered at discretion; a used-goods case whose four-sentence decline ended in "understood, thanks for looking" — the sentence every decline should end at). One chargeback, caught at the phrase and refunded first — the case never filed. The month's review cost: zero escalations, one pattern note written, zero improvisation. The ledger: refunds are the desk's cheapest repairs, and the ladder is why they stay that way.
What refund policy should a small desk run?
The four-rung ladder: rung one — full refund, same day, no argument (our fault or clean regret inside the 14-day window); rung two — partial refund offered first (shared fault, keep-the-item trades); rung three — store credit offered as a choice (+5% value, never swapped silently); rung four — the honest decline with reason and an escalation path.
How do we decline a refund without a detonation?
Four sentences: hear them ("I've read the thread"), the honest no once with the reason, the yes that exists (the real option at your discretion), and the path forward ("say the word if you want it reviewed fresh"). Skip suspicion theater and policy pedantry — never write a second "no."
How should chargeback threats be handled?
Refund immediately through rung one: the fee fight costs more than most refunds and reads worse. Reply the same day — "I see the dispute; I can resolve it directly today" — and close in the thread where the desk's memory is the evidence.
What about repeat refund requesters?
Same rung the fourth time as the first, plus a pattern note in the account. The right frame is the fix-the-cause conversation ("let's look at why this keeps going wrong"), not a fight over the fifth refund. The account data decides the next step.
Do full refunds always lose money?
No — they buy the ending: reviews, repeat orders and fewer cases. The rung-one refund is the desk's cheapest repair; chargebacks and detonated buyers cost multiples. Consistency (same fault, same rung, every time) is what the reviews can see.