Metrics question

How would you A/B test a new feature for Uber drivers without negatively impacting the platform?

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Experimentation design that accounts for marketplace interference, a common trap in two sided marketplace testing.

Clarify the feature, such as a new earnings display or route feature, since the risk profile differs by feature type.
Flag the interference risk explicitly: driver side experiments can bleed into the rider experience through shared supply.
Propose geo based or time based randomization instead of pure driver level randomization to reduce cross contamination.
Define primary metrics on the driver side, earnings and acceptance rate, plus guardrail rider side metrics like wait time and cancellation rate.
Run a small scale test in a few cities first, monitoring guardrails closely before wider rollout.

Explicitly names marketplace interference and spillover as the core risk, showing marketplace specific expertise.
Proposes geo or time based randomization as the fix, a real technique used in ride hailing experimentation.
Includes rider side guardrail metrics even though the feature targets drivers, correctly treating this as two sided.

Randomizing purely by individual driver ID, ignoring that treated and control drivers share the same rider pool.
Measuring only driver metrics with no guardrail on rider experience.

How would you decide between geo based and time based randomization for this specific feature?
What guardrail would make you stop the test early?