How a fraud score is calculated
The system gathers features: facts about the card, the device, the customer's history, the order and the network connection, and counts such as attempts per hour. A rules engine adds or subtracts points for each condition met. A model weighs all the features together and returns a probability, often scaled to a range such as 0 to 100.
Models are trained on past events labelled as fraud or genuine, mostly from chargebacks and issuers' fraud reports. Those labels arrive weeks late and fraudsters change tactics, so scores are monitored and models retrained regularly.
Turning a fraud score into a decision
Two thresholds are typical. Below the first the event is approved, above the second it is declined, and in between the customer is asked for more proof, through 3-D Secure, a one-time code or a manual review. Moving a threshold trades fraud losses against false positives, and the right point differs by product, country and margin.
A reason should travel with the number. Analysts and customer support need to know why an order scored high, so good systems return the main contributing factors.
Who produces fraud scores
Merchants and gateways score orders before authorisation, issuers score authorisation requests, and the card schemes supply network-wide scores to both. Banks and exchanges score logins, new accounts and withdrawals in the same way. A platform usually combines a vendor's score with its own rules, since only it knows its customers' normal behaviour.
Fraud score or credit score?
- Credit score
- Estimates whether a borrower will repay. A fraud score estimates whether the person is who they claim to be and means to pay at all.
- Risk-based authentication
- Uses a score to decide whether to challenge the customer. The score is the input, and the challenge is the action.
Where Fraud scoring shows up in a platform
Modules that do the work
- Fraud & riskReal-time fraud scoring, limits and case management, with the risk engines you already trust.
- Analytics & monitoringProduct analytics, attribution, engagement and crash reports, on a tracking plan you control.
- AuthenticationSign-in, passkeys and second factors for customers and staff, on your identity provider or ours.
- KYC / AMLIdentity checks, screening and ongoing monitoring, with the verification provider you choose.