Metrics question

How do you solve the problem about old data?

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

Ambiguous problem framing: can the candidate ask clarifying questions to turn a vague prompt into a concrete, solvable problem.

Ask the interviewer to clarify 'old data': stale user profile information, outdated cached content, or aging historical records affecting model or search quality.
State a likely interpretation, for example stale user preference data causing poor personalization, and confirm this framing with the interviewer before proceeding.
Identify the impact: old data can degrade recommendation relevance, search accuracy, or trigger outdated automated decisions if not refreshed.
Propose a solution: implement data freshness scoring and periodic re-verification prompts to users, or automatic decay weighting so older data contributes less to models over time.
Prioritize the decay-weighting approach first since it requires no extra user effort and can be deployed without disrupting the experience.
Define success as improvement in a downstream quality metric, such as recommendation click-through rate, tied to freshness of the underlying data used.

Leads with clarifying questions rather than guessing at a vague prompt, correctly treating this as an ambiguous framing exercise.
Proposes a concrete technical mechanism, decay weighting in a model, rather than a vague 'clean up the data' statement.
Ties the fix to a measurable downstream outcome, recommendation quality, rather than treating data freshness as an end in itself.

Answering without ever clarifying what 'old data' or 'the problem' specifically refers to.
Proposing a fix with no way to measure whether it actually improved anything downstream.

How would you decide the right decay rate for different types of data?
How would you handle data that is old but still highly accurate, like a stable preference?