Katkat
Back

Diagnose a Sudden Revenue Decline: Analyses, Metrics, and Root-Cause Tests

CoinbaseData Scientist · Technical Screen
MediumData ScientistTechnical ScreenAnalytics & Experimentation

Last updated

This interview question evaluates metric design, causal reasoning, experiment setup, diagnostics, SQL/statistical checks, and recommendations in a realistic interview setting. A strong answer for Diagnose a Sudden Revenue Decline: Analyses, Metrics, and Root-Cause Tests states assumptions, handles edge cases, explains trade-offs, and shows how to validate the result clearly.

Scenario

A key revenue metric on Coinbase's dashboard suddenly declines—for example, weekly trading-fee revenue drops noticeably week-over-week. You are the data scientist asked to investigate.

Question

Describe a structured, step-by-step approach to investigate and diagnose the root cause of the revenue drop. In your answer, address each of the following:

  1. Triage and validation — How would you confirm the drop is real (not a measurement/data-quality artifact) and pinpoint exactly when it began?
  2. Revenue decomposition — How would you decompose revenue into its underlying drivers (e.g., volume × take rate, or traffic × conversion × price) and quantify which driver moved the most? Show the math.
  3. Priority metrics — Which metrics and KPIs would you inspect first, and why?
  4. Data cuts / segmentation — Which segments and dimensions would you slice by to localize the drop, and how do they help you avoid Simpson's paradox?
  5. Funnel diagnostics — How would you use the trade and funding funnels to find the failing step?
  6. Releases & experiments audit — How would you check whether a recent deploy, config/fee change, or running A/B experiment caused the drop?
  7. External / market factors — How would you separate macro/market-driven effects (e.g., crypto volatility) from internal product/ops causes?
  8. Hypothesis testing & attribution — How would you turn observations into testable hypotheses and quantify each cause's contribution (with methods like ITS, diff-in-differences, synthetic control)?
  9. Action & monitoring — How would you communicate findings, choose remediations, and add monitoring to prevent recurrence?
Hints

Think in terms of decomposition (volume × take rate; or sessions × conversion × AOV), segmentation (asset, geo, platform, cohort, payment rail), funnels, recent changes/experiments, seasonality, and external market factors. Remember this is a crypto exchange: trading volume is highly sensitive to price volatility.