Set refund and chargeback policy thresholds
A refund is your choice and your cost; a chargeback is the customer's choice, your fee, and a fight you win or lose on evidence with a hard deadline. The policy must name both, with numbers.
Procedure
Set the refund window in days from the charge date, e.g. 30 days self-serve, 14 for a consumed digital good. Past the window, refunds need a named exception path.
Decide refund authority by amount tier: auto-approve under threshold_minor = 2000 (USD 20), human review 2000-50000, manager above 50000. Keep every amount in integer minor units.
Compute the true cost before granting: cost_minor = amount_minor + gateway_kept_fee_minor. Processors typically do not return the original transaction fee, so a "full refund" still costs the 2.9% + $0.30.
For digital goods on partial use, pro-rate by unelapsed days: refund_minor = amount_minor * remaining_days // term_days.
Treat a chargeback as a separate workflow: it arrives with a respond_by deadline (often 7-14 days). Missing it is an automatic loss.
Assemble evidence per chargeback: order record, IP and device, delivery proof, customer communication, and the accepted terms at purchase.
Track the chargeback ratio cb_ratio = chargebacks / transactions per month. Networks watch this; above roughly 0.9% risks a monitoring program.
Reconcile refunds and chargebacks against the settlement report monthly so both hit the ledger as negative revenue, never as an unbooked adjustment.
Name the owner: support auto-refunds under the threshold, finance reviews above it, and both write the reason to the ledger.
For a cancel-and-refund, pro-rate from the cancellation date, not the request date, when the contract allows forward cancellation only.
Trend a monthly refund rate by product; a spike is a quality signal before it is a cost signal.
Refund to the original tender; refunding a different card or account is a laundering vector, not a courtesy.
Reconcile refunds to the settlement report line, not a separate spreadsheet, so the reversal carries the same currency.
Keep a make-good policy for an outage: a credit is cheaper than a cash refund and stays in the billing system.
Set a monthly refund-rate alarm; a jump beyond a set band triggers a review before the quarter's recognition closes.
Pitfalls
Treating a chargeback as a refund and not responding: you pay the amount plus the chargeback fee and lose by default.
Forgetting the retained processing fee: refunding a $49 charge made at 2.9% + $0.30 still costs about $1.72 that does not come back.
A refund window longer than the dispute window invites a double payout if the customer also disputes.
Auto-refunding above the threshold tier without review is a fraud vector; the tier exists to stop it.
Tracking refunds in a spreadsheet apart from the ledger hides them from revenue recognition.
Refunding to a different card or account than the original is a laundering vector; refund the tender.
A refund rate trended month over month catches a product defect earlier than a growing support backlog.
Issuing store credit instead of cash without saying so on the policy page invites disputes.
Make-good credits applied as cash refunds leak margin; keep them as credits unless the customer closes the account.
Verification
python3 -c "amt=4900; fee=amt*290//10000+30; print(amt+fee)"
# 5072 -> the customer gets 4900, you are out 5072 minor
Report: the refund-window days, the authority thresholds, and this month's chargeback ratio read from the processor's dispute report.
1---2name: set-refund-and-chargeback-policy-thresholds3description: Use when defining when money is given back. Sets refund windows and authority thresholds tied to cost, and treats a chargeback as a processor evidence task with a deadline, not a refund.4metadata:5 source: https://skillmd.com/skills/jiayaoqijia/set-refund-and-chargeback-policy-thresholds6---78# Set refund and chargeback policy thresholds910A refund is your choice and your cost; a chargeback is the customer's choice, your fee, and a fight you win or lose on evidence with a hard deadline. The policy must name both, with numbers.1112## Procedure13141. Set the refund window in days from the charge date, e.g. 30 days self-serve, 14 for a consumed digital good. Past the window, refunds need a named exception path.152. Decide refund authority by amount tier: auto-approve under `threshold_minor = 2000` (USD 20), human review 2000-50000, manager above 50000. Keep every amount in integer minor units.163. Compute the true cost before granting: `cost_minor = amount_minor + gateway_kept_fee_minor`. Processors typically do not return the original transaction fee, so a "full refund" still costs the 2.9% + $0.30.174. For digital goods on partial use, pro-rate by unelapsed days: `refund_minor = amount_minor * remaining_days // term_days`.185. Treat a chargeback as a separate workflow: it arrives with a `respond_by` deadline (often 7-14 days). Missing it is an automatic loss.196. Assemble evidence per chargeback: order record, IP and device, delivery proof, customer communication, and the accepted terms at purchase.207. Track the chargeback ratio `cb_ratio = chargebacks / transactions` per month. Networks watch this; above roughly 0.9% risks a monitoring program.218. Reconcile refunds and chargebacks against the settlement report monthly so both hit the ledger as negative revenue, never as an unbooked adjustment.22239. Name the owner: support auto-refunds under the threshold, finance reviews above it, and both write the reason to the ledger.2410. For a cancel-and-refund, pro-rate from the cancellation date, not the request date, when the contract allows forward cancellation only.2511. Trend a monthly refund rate by product; a spike is a quality signal before it is a cost signal.2612. Refund to the original tender; refunding a different card or account is a laundering vector, not a courtesy.2713. Reconcile refunds to the settlement report line, not a separate spreadsheet, so the reversal carries the same currency.2814. Keep a make-good policy for an outage: a credit is cheaper than a cash refund and stays in the billing system.2915. Set a monthly refund-rate alarm; a jump beyond a set band triggers a review before the quarter's recognition closes.3031## Pitfalls3233- Treating a chargeback as a refund and not responding: you pay the amount plus the chargeback fee and lose by default.34- Forgetting the retained processing fee: refunding a $49 charge made at 2.9% + $0.30 still costs about $1.72 that does not come back.35- A refund window longer than the dispute window invites a double payout if the customer also disputes.36- Auto-refunding above the threshold tier without review is a fraud vector; the tier exists to stop it.37- Tracking refunds in a spreadsheet apart from the ledger hides them from revenue recognition.3839- Refunding to a different card or account than the original is a laundering vector; refund the tender.40- A refund rate trended month over month catches a product defect earlier than a growing support backlog.41- Issuing store credit instead of cash without saying so on the policy page invites disputes.42- Make-good credits applied as cash refunds leak margin; keep them as credits unless the customer closes the account.4344## Verification4546 python3 -c "amt=4900; fee=amt*290//10000+30; print(amt+fee)"47 # 5072 -> the customer gets 4900, you are out 5072 minor4849Report: the refund-window days, the authority thresholds, and this month's chargeback ratio read from the processor's dispute report.