Payments / Intermediate / 5 min
Fraud detection: rules, scores and thresholds
Risk scores from simple signals. Move the block and review lines and trade fraud caught against honest customers refused.
Start the simulation → Change the load, break a part, watch what happens. Free, no sign-up.
In short
- What it is
- Fraud detection gives each payment a risk score from signals like device, velocity, amount and location, then allows, reviews or blocks it.
- Why it matters
- Fraud costs money and chargebacks, but blocking honest customers costs sales and trust, so the decision must be fast and tunable.
- The trade-off
- No threshold is perfect: lower lines catch more fraud and refuse more honest people, and manual review in between costs staff time.
- Say it in an interview
- "Real-time scoring with rules plus a model on features from a feature store, allow/review/block thresholds tuned on chargeback data, and step-up checks like OTP instead of hard blocks."
Check yourself
You lower the block threshold. What happens?
Answer: More fraud is caught, but more honest customers are refused too. Scores of honest and fraudulent payments overlap, so any line trades false positives against missed fraud.
Why add a manual review band between "allow" and "block"?
Answer: Unclear cases get a human decision instead of an automatic mistake either way. Reviewers handle the grey zone, at the cost of time and staff; the band width sets how many cases they see.
Which is a typical fraud signal?
Answer: Many payments within a few minutes from a new device. Velocity and new devices are classic signs of a stolen card being tested and used quickly.