Fraud Detection · Documentation

Business Problem & Solution

The problem Fraud Detection solves in the hospitality ecosystem and how this downscaled demo proves it.

Fraud Detection
a bookingthe systema risk callnew-account high valueBIN ≠ IP countrydevice burst · watchlistrisk scoringrules + modela transparent scorenamed reasonsauto-hold ≥ 75catch risky bookings before they cost
Live diagram — the deterministic signal rules are authoritative; the metered model rides on top, and a score ≥ 75 auto-holds.

Catch risky bookings before they cost

Booking and payment fraud is a chargeback and trust problem: a high-value first booking from a new account, a card whose BIN country does not match the guest IP, a burst of bookings from one device, a watchlisted email domain, a billing-versus-stay-city mismatch, a rapid cancel-rebook. Fraud Detection scores bookings against this signal set, surfaces the high-risk ones, and lets an owner hold, clear, or escalate a signal into a refund/dispute case to limit exposure.

What this proves — and what it simplifies

It proves a transparent risk score with named reasons, an auto-hold threshold, and a clean escalation loop into the refund/dispute app (#21). The deterministic signal rules are authoritative; the metered model rides on top. It simplifies by scoring synthetic bookings with a recorded OSS model at $0 by default; no real cards, no real guests, no PII.

Business Problem & Solution · Fraud Detection · Abhishek Saxena