Define Marketplace Success Metrics and Investigate Their Movement
WhatnotData Scientist · Technical ScreenLast updated
A product analytics interview about defining success for a two-sided marketplace and diagnosing an unexpected metric movement. It tests metric decomposition, cohort and funnel analysis, instrumentation checks, and the ability to separate product effects from mix shifts.
Define Marketplace Success Metrics and Investigate Their Movement
A two-sided live-commerce marketplace asks you to define success metrics. A primary metric changes unexpectedly, and the team needs a diagnosis that distinguishes product impact from changes in buyers, sellers, inventory, and measurement.
Constraints & Assumptions
- Buyers can transact with multiple sellers and sellers can list multiple items.
- The metric may be a rate, so numerator and denominator must be investigated separately.
- Promotions and supply mix can change at the same time as product behavior.
- Late transaction and refund events can revise historical values.
Clarifying Questions to Ask
- Which marketplace decision will the metric support?
- Is success short-term conversion, healthy repeat behavior, liquidity, or contribution margin?
- At what grain and delay are orders, refunds, sessions, and listings considered complete?
Part 1 — Metric framework
Propose one primary marketplace metric and a compact set of buyer, seller, liquidity, quality, and economic guardrails.
What This Part Should Cover
- A defined unit, denominator, window, and inclusion rule
- Metrics connected through a causal funnel
- Counter-metrics that prevent one-sided optimization
Part 2 — Diagnosis
Lay out a query and analysis sequence for a sudden change in the primary metric.
What This Part Should Cover
- Logging and definition checks before causal stories
- Numerator-denominator decomposition
- Cohort, mix, supply, promotion, and latency analyses
Hints
Hint 1
Decompose before segmenting: For a rate, first determine whether the numerator, denominator, or both moved. Then hold definitions fixed and decompose by stable cohorts and marketplace sides.
Follow-up questions
- How would you detect that a metric improved by starving low-converting users?
- Which metric should be used for experiment power calculations?
Solution walkthrough is for members
Get the full walkthrough with reference code and complexity analysis, plus Kitty Coach at 300 messages a day. Hints, the code runner and community answers stay free.