Fraud scoring.
In plain English
Fraud scoring runs a transaction through a model that weighs dozens of signals and produces a risk number before the payment is approved. Inputs include the amount, the merchant category, the distance from your usual pattern, device and network fingerprints, the speed of recent activity, and whether the billing details match. The score is compared against thresholds the issuer or merchant sets: approve, challenge with a step-up check, or decline outright. Where that threshold sits is a business tradeoff, because tightening it blocks more fraud and also blocks more real customers.
01Why it matters
A false decline on a legitimate purchase is the direct result of where a bank set its threshold, which is why confirming through the app or calling the issuer usually clears the block in minutes.
02The math, step by step
Say a model reviews 100,000 transactions containing 200 fraudulent ones. A tight setting catches 180 of them but also declines 1,500 good purchases. A looser setting catches 140 and declines 400. The bank is choosing which of the two errors costs it more.
Illustrative example. The amounts here are hypothetical, chosen to show how the math works, not real quoted rates or figures.
03What this is NOT
The first decision is not made by a human. A model scores and decides in milliseconds. People get involved later, on a small share of flagged cases and on disputes, which is why an initial decline usually arrives with no explanation attached.
04Receipts
Every figure on this page is sourced to a primary document. Tap to open the original.
Plain-English answers from our glossary. Receipts included. Never advice.
Educational tool. Answers come only from ClearMoneySchool's published glossary and are not advice. Why we never give advice