Refund Fraud: CEO Agent Triage with Evidence Summary

Gives every refund request a reasoned, evidence-backed risk write-up instead of a raw score. The agent gathers signals across systems, reasons about whether the pattern looks like genuine dissatisfaction or coordinated abuse, and hands the support agent a recommendation they can trust and explain.

Reach for this when refund decisions are nuanced and a binary rule produces too many false positives. The agent narrative helps frontline staff approve legitimate refunds fast while spotting the subtle multi-account cases a threshold would miss.

1. 1A Zendesk refund ticket triggers the agent.
2. 2The agent queries Stripe for the customer's charge, dispute, and refund history.
3. 3It pulls the customer's prior Zendesk tickets to read the complaint pattern.
4. 4It checks Postgres for linked accounts and lifetime refund totals.
5. 5The agent weighs the evidence and decides: clean, watch, or escalate.
6. 6It posts a cited risk summary and recommended action as an internal Zendesk note, escalating to Slack only when the verdict is escalate.