Interview Questions › AI Product

AI warnings for UPI payment scams

Design an AI feature that warns PhonePe users before they pay a likely scammer. How would you build and evaluate it?

Context

Scams often use urgency ("your account will be blocked"). False alarms annoy users and slow payments; missed scams cost users money and trust.

Clarifying questions to ask first

• What signals do we have (payee history, reports, message context)?
• Should we block or warn?
• What is an acceptable false-positive rate?

How to approach it

1. Start from the user problem, not the model: is AI the right tool here?
2. Describe the experience and where the model fits in it
3. Define what "good output" means and how you would evaluate it before and after launch
4. Name the failure modes (wrong answers, bias, privacy, cost) and how you contain them
5. Close with a launch plan and success metrics

Framework: User and problem -> Is AI the right tool? -> Experience design and human-in-the-loop -> Quality bar and evaluation -> Risks, cost and latency -> Staged launch and metrics

Sample answer

• Warn with a reason, block only on high confidence
• Precision vs recall trade-off set by loss size
• Feedback loop from user reports

1. Backtest on labelled past scam reports
2. Measure precision and recall by amount bucket
3. Holdout to measure prevented losses

Risk-scored warnings with clear reasons, stricter thresholds for large amounts and new payees, blocking only at very high confidence. Measure prevented losses against a holdout, with false positives and payment completion as guardrails.

What interviewers look for

• Scam losses prevented
• False positive rate
• Payment completion