the-chargeback-pattern-finderlisted

Takes a dispute or chargeback export and finds whether the disputes cluster by product, channel, geography, order value band, or payment method into a fraud pattern, or whether they're mostly service failures and friendly fraud, then weighs that against what over-blocking would cost in rejected good orders. Use when a dispute rate is rising, a payment provider has sent a warning, or the store wants to know how much of its chargeback problem is real fraud. Boundary: `the-anomaly-alert` flags a generic single-metric time series moving outside its normal range. This skill dates every dispute to the sale it belongs to, not the filing date, and clusters across the dimensions an export actually has instead.
sidchaudhary/gtm-skills · ★ 1 · AI & Automation · score 74

Install: claude install-skill sidchaudhary/gtm-skills

# The Chargeback Pattern Check Take a dispute export and work out how much of it is real fraud versus a service failure or friendly fraud, using a stated dating rule and a stated set of clustering dimensions, not an impression of "chargebacks are up." ## How to run Ask the user for these inputs. If any are missing, ask before analyzing. 1. **Dispute export**: one row per dispute, with amount, reason code, product, and ideally both the transaction date and the filed date. 2. **Order volume for the same period**: needed to turn a dispute count into a rate. 3. **Whatever supports clustering**: channel, geography, order value, and payment method per disputed order, plus card BIN range, shipping/billing address match, and customer order velocity if the export happens to carry them (most don't; say which of these couldn't be checked rather than treating their absence as a gap in the analysis). 4. **Declined-order export, if over-blocking is a question**: orders the fraud rules rejected, not just disputes that got through. Disputes show what went wrong after approval; declines show what the rules are already stopping. Sizing over-blocking without the decline side is guesswork. ## Method 1. **Date every dispute to the original transaction, not the filing date.** A dispute filed in March belongs to the sale that happened in January. If only a filed date exists, use it but say explicitly the rate is now dated to filing and will understate a rising problem, since disputes take wee