You are covering a shift as a risk analyst. Ten card payments that the fraud model flagged are waiting in your queue, and for each one you read the details and the rules that fired before you decide. Wrongly declining a genuine customer costs money here as surely as missed fraud does.
The model gives every case a risk score, and you are free to overrule it. On some cases the score is wrong, and the signals underneath it show why. Every shift builds a new queue from the same kinds of cases.
Approvewhen the signals are clean and the payment should go through.
Declinewhen the pattern is clearly fraud and the payment must stop before it settles.
Escalatewhen the signals disagree. The order is held while a senior analyst confirms the payment with the cardholder.
You can also press A to approve, D to decline or E to escalate.
Results of your shift
AYour grade
0
Genuine payments you approved
0
Risky payments you stopped or escalated correctly
0
Fraud you approved or escalated instead of declining
0
Genuine payments you blocked or delayed
Net effect on the business€0
What an experienced analyst looks at first
Velocity tells you more than the amount. A tiny charge after several declined cards within a couple of minutes is card testing, and it should be declined even though it costs almost nothing.
A foreign IP address alone is not fraud, because people travel. Several soft flags together are a different matter, such as an account a few hours old, a new device and shipping that differs from billing.
When good and bad signals balance each other, escalate the payment. Declining on reflex loses customers who did nothing wrong.