Every chargeback forces the same choice, and most merchants make it by gut: fight the big ones, refund the small ones, argue about the rest. That instinct is roughly right and exactly wrong in the details — because the break-even point is not the order value, it is the order value times your win rate, minus what fighting costs. A $500 dispute you win a quarter of the time is worth less, in expected value, than a $100 dispute you win three quarters of the time, once fees and staff time are counted. This guide works through that comparison properly, then covers the cases where money alone should not decide.
The expected-value math, in plain terms
There are three cash outcomes, and the comparison needs all three before it needs any averages:
Option
Cash outcome
Refund now
You return the sale. Cost = the order value. No fee, no staff time.
Fight — and win
You keep the sale, paying the dispute fee and the staff time. Cost = fee + staff time.
Fight — and lose
The sale is returned AND you pay the fee and the staff time. Cost = order value + fee + staff time.
The fee and the staff time are paid whether you win or lose. Only the sale itself is at stake on the outcome. So the expected advantage of fighting is: win rate × order value − fee − staff time. If it is positive, fighting is worth more than refunding, on average over many similar cases. If it is negative, refunding is worth more. The companion fight-or-refund calculator judges the result against a margin of 5% of the order value: inside that margin, money alone does not settle it.
Two warnings belong with the formula. First, a single case is won or lost, never "expected" — expected value is an average over many similar cases, and a merchant fighting one dispute a month will feel the variance, not the average. Second, the formula is only as honest as the win rate in it. A rate borrowed from a vendor page is not your rate, and the next section explains why the published numbers disagree so wildly.
A worked example
Take a $100 order, a $15 dispute fee, and one hour of staff time at $20. Fighting costs $35 whether you win or lose. Refunding now costs $100. At a 30% win rate, the expected advantage of fighting is 30% × $100 − $15 − $20 = −$5: inside the margin, too close to call on money alone. At a 60% win rate, it is 60% × $100 − $35 = +$25: fighting is worth $25 more. The break-even win rate here is 35% — the $35 fighting cost divided by the $100 at stake. Any rate above 35% favours fighting; any rate below it favours refunding; within 5% of the order value ($5) either side of zero, the calculator calls it too close to call.
Notice what the example exposes: for small orders, the fee and the staff time dominate. A $20 order with a $15 fee and $20 of staff time needs a win rate above 175% to break even — impossible — so it should almost always be refunded, whatever the evidence. For large orders, the fee and time shrink to rounding, and the decision turns almost entirely on the win rate and the evidence. That is the whole discipline in one paragraph: small disputes are a cost problem, large disputes are an evidence problem.
Why published win rates disagree
Search for a merchant win rate and you will find 30%, 44% and 54% on respectable-looking pages, all presented as "the" rate. They are not contradictions; they count different things — and together they span roughly 30–54%:
Figure
What it counts
Source
30–45%
Self-managed average across merchants handling disputes themselves — the low end of the published range.
Fincoro, 2026 benchmarks.
About 44%
Average win rate on chargebacks merchants actually represented. Secondary summaries of the report vary slightly (43.8–44.6%).
Chargebacks911, 2026 Chargeback Field Report.
54% (US)
Share of representments won by US merchants who fought (UK: 49.1%).
Mastercard / Datos Insights, The Chargeback Window of Opportunity (2025).
About 10–20% net recovery
Share of all disputed value recovered, counting uncontested cases, second-cycle disputes and undetected friendly fraud.
Chargebacks911 field reports (10.7% net recovery in the 2026 report; about 20% in the 2024 report).
20–30% (one older playbook)
A figure one playbook still quotes. It sits below the current published 30–54% range — named here only so you recognise it when you see it, never as a rate to use.
Older playbook figure; below the current range.
The gap between about 44% and 10.7% is the point: merchants who fight, with evidence prepared, win nearly half the time; merchants overall recover barely a tenth, because most disputes are never fought at all. So the rate that belongs in your calculation is neither number — it is yours, for cases like the one in front of you, from your processor dashboard: the share of similar cases you fought and won. If you have no history, run the calculator at a low rate and at a high rate and see whether the answer changes; if it flips, your decision is really a decision about evidence, and the reason code decoder plus the evidence checklist builder matter more than the arithmetic.
One more trap: reason codes are not equally winnable. Fraud cases with strong device and delivery evidence behave differently from "not as described" cases that turn on product quality, and no honest page publishes a per-code win rate — the data is merchant-specific and the networks do not publish it. That is why the calculator's category selector deliberately changes nothing in the math: it records the kind of case you are deciding, and reminds you the evidence rules differ, without inventing a number.
Too close to call: the non-money factors
When the expected advantage sits inside the margin, money has done all it can, and four non-money factors decide:
Evidence strength for this reason code. Do you have the exact evidence the code requires — delivery proof, device data, prior undisputed transactions for fraud cases, cancellation records for recurring billing — or are you hoping the issuer is lenient? Weak evidence should push a close call toward refunding; the representment will cost the staff time either way.
Customer relationship. A repeat customer with a genuine confusion dispute is worth more than one order. Refunding fast can keep the customer; fighting hard can lose them and earn a public complaint. A one-time buyer with clear friendly fraud is the opposite case.
Precedent and signalling. Refunding every small friendly-fraud dispute teaches abusers that your store pays out. Fighting a few visible ones, even at a small expected loss, can be a prevention expense, not a dispute expense. Price it honestly as such — do not pretend the calculator blessed it.
Ratio position. This one deserves its own section, because it can override a positive expected value entirely.
When refunding first protects your ratio
A lost dispute still counts toward your chargeback ratio even when you accept it, and a won dispute does not erase the original count. But a refund issued before the customer contacts their bank creates no dispute at all — and disputes resolved through pre-dispute products such as Visa RDR are excluded from VAMP. That asymmetry is why refunding first is sometimes the smarter ratio play even when fighting looks profitable on the single case.
The thresholds that matter: Visa's VAMP program flags merchants at a 1.5% ratio from April 1, 2026, with a 1,500 fraud-report-plus-dispute monthly floor (CEMEA stays at 2.2%); Mastercard's ECM program flags at 100+ chargebacks in a month with a 1.50% chargeback-to-transaction ratio, with HECM at 300+ and 3.00%. If your ratio is close to either line, a small expected-value edge from fighting is not worth the monitoring risk — check your position with the chargeback ratio calculator and the monitoring program checker before you let the calculator's Fight band overrule ratio safety. And remember: ratio damage has no honest dollar figure, which is why the calculator shows it only as a flag, never as a dollar line.
Arbitration is a second decision
If representment fails and the case escalates past pre-arbitration, the network rules on it finally — and the losing party pays the filing fee. Published figures put Visa at $500 and Mastercard at $400, though one April 2025 update reports Visa's fee rising to $600, so confirm with your acquirer before relying on either number. Arbitration is deliberately excluded from the fight-or-refund comparison: it is a later stage, with its own evidence bar and its own costs, and folding a $500 fee into a first-cycle decision would distort every small dispute. When a case reaches that stage, run the comparison again with the arbitration fee in place of the dispute fee — for most order values, that alone flips the answer to refunding.
Run your own numbers
Your order value, your fee, your staff time, your win rate — win, lose and refund outcomes side by side, with the ratio-risk flag kept separate from the dollars.
Compare expected values from your own figures: refunding now costs the order value; fighting and winning costs fee plus staff time; fighting and losing costs order value plus fee plus staff time. The expected advantage of fighting is win rate times order value minus fee minus staff time, judged against a 5% margin.
Because they count different things: published benchmarks span roughly 30–54% and vary by category and evidence (Fincoro 2026 self-managed average 30–45%; Chargebacks911 2026 Field Report about 44% on represented cases; Mastercard/Datos 2025 54% US representment wins); net recovery across all disputes is only about 10–20% because most disputes go uncontested. No single published rate is a fact — use your own rate.
When you are near a monitoring threshold. A refund before the customer contacts their bank creates no dispute at all, and pre-dispute resolutions such as Visa RDR are excluded from VAMP — so protecting a ratio near Visa's 1.5% line or Mastercard's 1.5% / 100+ line can be worth more than a small expected edge from fighting.
Only at a later stage. Arbitration fees (published figures: Visa $500, Mastercard $400, loser pays; one April 2025 update reports Visa at $600) apply if a case escalates past representment, so they are kept out of the first-cycle comparison and judged as their own decision.
Sources (checked October 27, 2026): Visa VAMP 1.5% merchant Excessive threshold from April 1, 2026 (1,500-event monthly floor; CEMEA 2.2%) — Chargeflow, "Visa VAMP rules 2026" and Dodo Payments, "Visa VAMP Explained", summarising Visa's Acquirer Monitoring Program; Mastercard ECM 100+ chargebacks and 1.50% ratio (HECM 300+ and 3.00%) — JPMorgan Chase, "Mastercard Excessive Chargeback Merchant Program Guide" (acquirer reproduction of Mastercard's rules) and Chargeback.io's Mastercard monitoring guides; arbitration filing fees, loser pays, Visa $500 / Mastercard $400 — Justt, "Pre-Arbitration Chargebacks", World Financial Review, "What is Chargeback Arbitration?", and Chargeback Gurus' pre-arbitration guide, with the caveat that Chargebacks911's April 2025 update reports Visa's fee rising to $600; win rates — Fincoro 2026 benchmarks (self-managed average 30–45%); Chargebacks911, "2026 Chargeback Field Report" (about 44% represented-case wins — secondary summaries vary slightly, 43.8–44.6%; 10.7% net recovery) and Mastercard / Datos Insights, "The Chargeback Window of Opportunity" (2025) via Chargeback.io (54% US representment wins; UK 49.1%); the 20–30% figure one older playbook still quotes sits below the current published 30–54% range and is named only as an outlier; $128 average — Chargeflow, "True Cost of a Chargeback", summarising Mastercard research with Javelin (2026) ($46 + $82, excluding the sale); processor dispute fees $15 Stripe / $20 PayPal card chargebacks — each processor's own published page, verified October 27, 2026 on the chargeback cost calculator page. Figures vary by country, currency and merchant contract; check your own processor dashboard before acting.
General information only — not financial advice. This guide compares expected values from figures you supply. A single case is won or lost, never "expected". Enter your own win rate from your representment history; benchmarks are not predictions. Verify thresholds and fees with your acquirer.
Chargeback Clock