← Lessons

quiz vs the machine

Platinum1780

System Design

Fraud Detection in Orders

Scoring orders for fraud risk with rules, signals, and a review pipeline.

6 min read · advanced · beat Platinum to climb

Catching bad orders

Fraud detection evaluates each order for the risk that it is fraudulent, such as a stolen card. The goal is to block bad orders while approving the overwhelming majority of legitimate ones without friction.

Signals and scoring

The system combines many signals into a risk score:

  • Velocity: many orders from one card, device, or address in a short time.
  • Mismatch: billing and shipping addresses far apart, or a new device on a known account.
  • Reputation: known bad cards, emails, or IP ranges.

A model or rules engine turns these into a score. The model often runs in near real time during checkout.

Acting on risk

  • Low risk: approve automatically.
  • High risk: block or step up with verification.
  • Gray zone: send to a manual review queue rather than guessing.

Decisions feed back as labels to retrain the model, and the system must balance false positives, which annoy good shoppers, against false negatives, which cost money.

Key idea

Fraud detection scores orders from velocity, mismatch, and reputation signals, then approves, blocks, or routes to manual review while balancing false positives and negatives.

Check yourself

Answer to earn rating on the learn ladder.

1. What does a velocity signal capture?

2. What is the gray zone handled with?

3. What trade off does fraud detection balance?