Fraud scoring
What is fraud scoring?
Fraud scoring is a risk-assessment method that gives each transaction a numerical score representing how likely it is to be . The score is calculated from signals such as transaction history, customer behavior, and anomaly-detection patterns, so a or merchant can act on the riskiest transactions first.
The score maps to low, medium, and high risk bands depending on where it lands. A low score lets a payment clear without friction, a medium score can trigger an extra check such as , and a high score can block the transaction outright. Because the score is produced during checkout, it works on where the cardholder isn't there to verify identity in person. Fraud scoring is one layer of a wider fraud-prevention stack, sitting alongside tools like and the .
Key facts
- Also known as: transaction risk scoring, risk scoring
- Inputs: transaction history, customer behavior, device and location data, and the velocity of recent payment attempts
- Output: a numerical score that maps to low, medium, or high risk bands
- Applies to: online checkout, card-not-present transactions, and recurring billing
- Typical actions: approve, challenge with step-up authentication, or decline
How fraud scoring works
- Data collection – At checkout, the system gathers signals tied to the transaction: amount, currency, the cardholder's device and location, and the history linked to the card or account.
- Pattern analysis – Each signal is compared against normal behavior and known fraud patterns. Deviations such as an unusual amount, a mismatched location, or several rapid attempts raise the risk reading.
- Score calculation – The model combines the weighted signals into a single score that reflects the transaction's overall risk.
- Risk categorization – The score places the transaction in a low, medium, or high band.
- Action – Based on the band, the transaction is approved, sent for a step-up check, or declined.
Why it matters
A fraud score turns a yes/no call into a graded one, so low-risk payments clear without friction while only the riskiest ones get extra scrutiny. That protects revenue on two fronts: it blocks transactions likely to end in a , and it spares good customers who a single blunt rule would otherwise decline.
Scoring also scales in a way manual review can't. A risk team can't inspect every order, but a score can be applied to all of them and route only the borderline cases to a human. False positives (legitimate payments flagged as fraud) drop when the decision rests on a weighted score rather than one rigid rule.
Common issues
- False declines – An overly aggressive threshold blocks legitimate cardholders, who see a generic decline and often abandon the purchase.
- Missed fraud – A threshold set too loose lets fraudulent payments through, which resurface later as chargebacks or .
- Stale models – Fraud patterns shift, so a model that isn't retrained loses accuracy as attackers adapt, for example through a that tests stolen card numbers.
- Data gaps – A score is only as good as the signals behind it; thin data on a new customer weakens the prediction.
