Templates for the questions a payments team asks: how to catch fraudsters without angering honest customers, what looks strange in a transaction flow when nobody labeled it, where money moves in circles — and whether the data can be trusted before you build conclusions on it. This section has 4 templates.
Every template is a live walkthrough on demo data: open it in the Templates section of Analytics Agent and press “Use this template” to run it on your own transaction flow.
Card Fraud Detection (The Catch-And-Anger Trade, Measured)
Shows: it finds rare fraudulent transactions without blocking half your customers: three models compete on history, the alert threshold is set by the price of a missed fraud against an angered customer, and every flag comes with reasons you can check by hand.
You get: the “catch or don’t anger” trade measured in numbers: a lot of fraud caught, few honest payments blocked.
Transaction Anomalies (No Labels, Two Detectors, Reasons Attached)
Shows: it finds the strange stuff in a transaction flow when nobody confirmed in advance what was fraud: two independent detectors, high severity only when both agree, a plain-language reason on every flag, and a ready-made notification rule.
You get: suspicious payments get noticed continuously, not once in a while — and a person can verify every finding in the payment system in seconds.
Transfer Graph (Where Money Moves in Circles)
Shows: it finds “collect and spread” rings and fans in the transfer log: a clear search for fast circles with a stable amount, a fan rule for new counterparties, and centrality so a “central” account is not confused with a “suspicious” one.
You get: instead of an answer to “who is the fraudster”, a hint on where in the flow to look: rings and fans are tracked month after month.
Data Quality Passport (Can I Trust This Feed?)
Shows: eight simple data checks before you start: a “pass / attention / fail” passport for each one, and every defect gets a date and a plain-language explanation.
You get: half of what looks like fraud is a broken export — checking the data and getting the passport is cheaper than investigating something that never happened.