You work at a payments company processing 15,000 card transactions per second at peak. Fraud currently costs the company roughly $40M per year in chargebacks, and the existing system is a set of hand-written rules maintained by a risk team.
You need to build a real-time fraud detection system that scores every transaction inline: before authorisation is returned to the merchant, and decides to approve, decline, or send to step-up authentication (3DS).
Fraud is roughly 0.15% of transactions. The business is emphatic that false positives are expensive: a declined legitimate transaction costs an estimated $12 in lifetime value, while the average fraudulent transaction that gets through costs $180. Confirmed fraud labels arrive via chargebacks, which lag the transaction by 15 to 90 days.
Fraudsters adapt within days of a rule change. Design a system that stays effective as they do.
Build the architecture on a canvas: place the components, configure them, connect them into a data flow, and write a short reason for each one. The AI reviewer grades your design against a rubric written specifically for this problem.
Chargebacks take up to 90 days. How do you evaluate a model you shipped last week?
Your model outputs 0.07 for a transaction. Walk me through exactly how that becomes an approve, decline, or step-up.
A fraud ring starts making $9 test transactions to find your threshold. How does your system notice, and how does it respond?
Minimum 8 components · needs a wide desktop screen