# Balance Fraud Control with User Experience
Design a strategy to balance fraud control with improving the experience of legitimate users. Explain how you would choose actions for different risk levels, evaluate the trade-off, and detect when the strategy is causing avoidable friction. No particular fraud type or enforcement policy is specified, so state the context you would clarify.
### What a Strong Answer Covers
- A clear threat, decision unit, and harm model before selecting interventions.
- Risk-based actions with escalation or appeal for uncertain cases.
- Joint fraud and legitimate-user outcome metrics rather than a single rejection rate.
- Evaluation that accounts for delayed labels, selective verification, and adaptive attackers.
```hint Use more than two actions
Consider the trade-off between allowing, asking for additional verification, and blocking.
```
### Follow-up Questions
- When could a lower observed fraud rate be misleading?
- How would you assess the cost of mistakenly challenging a legitimate user?
Overview: Practice designing fraud controls that balance expected harm, risk-based verification, legitimate-user friction, and reliable policy evaluation.
Design a strategy to balance fraud control with improving the experience of legitimate users. Explain how you would choose actions for different risk levels, evaluate the trade-off, and detect when the strategy is causing avoidable friction. No particular fraud type or enforcement policy is specified, so state the context you would clarify.
What a Strong Answer Covers Guidance
A clear threat, decision unit, and harm model before selecting interventions.
Risk-based actions with escalation or appeal for uncertain cases.
Joint fraud and legitimate-user outcome metrics rather than a single rejection rate.
Evaluation that accounts for delayed labels, selective verification, and adaptive attackers.
Follow-up Questions Guidance
When could a lower observed fraud rate be misleading?
How would you assess the cost of mistakenly challenging a legitimate user?