Diagnose sustained drop in executed trades

Context: You’re the analyst for a brokerage product covering onboarding → trading. Over the last 5 weeks, executed_trades per active user fell 22% and stayed low. No major outages are reported. How would you pinpoint the root cause and validate it? 1) Build a KPI tree from acquisition → KYC approval → first funding → first order attempt → execution quality → short-term retention. For each node, specify an unambiguous metric definition, the expected range, and guardrail thresholds (e.g., new-user day-7 retention, order rejection rate, deposit success rate, quote latency P95). Explain how you’d detect which node(s) are responsible for most of the variance in executed_trades. 2) Propose a drilldown plan and justify the exact cut order: acquisition channel, signup cohort, account tenure, funding method, platform (iOS/Android/Web), asset class (equity/option/crypto), order type (market/limit), market-hours vs after-hours, geo, and order reject_code family. For each cut, state the diagnostic you’d run (e.g., waterfall attribution, interaction effects) and the decision you’d make based on each possible outcome. 3) Disambiguate product vs external drivers. List at least five sanity checks with data sources: market volatility/indices, holiday calendar, competitor promos/fee changes, symbol halts, exchange connectivity, release calendar, instrumentation drift. Describe exactly how each check could confirm or refute a hypothesis. 4) Suppose an onboarding flow change shipped on 2025-07-10 across all platforms. Design a confirmation plan that doesn’t expose 100% of traffic: choose experiment unit (user, geo, or time-based switchback), outline rollout/rollback, define primary/guardrail metrics, and detail power analysis/monitoring windows to avoid novelty bias. 5) If instrumentation is insufficient, specify the minimum additional events/dimensions you’d add (e.g., KYC step_id, funding error_code, order entry latency, quote staleness, partial fill counts) and how you’d backfill or triangulate with logs to avoid a blind spot while the fix is in progress.