Business Context

Airbnb’s Core Stays marketplace serves ~6M active listings and ~150M monthly site/app visitors globally. The Finance team uses a weekly GMV forecast (gross booking value) to set marketing budgets, staffing for customer support, and liquidity planning for host payouts. Over the last 8 weeks, forecast accuracy has deteriorated: the median absolute percentage error (MdAPE) increased from 6% to 14%, with several weeks missing by >20%. This is now a board-level issue because a 10% GMV miss can translate into ~$80–$120M variance in quarterly revenue recognition and marketing ROI.

Metric Scenario

You own the analytics for forecasting and KPI instrumentation. Stakeholders are debating whether the issue is:

• a true demand shift (macro seasonality, competitor pricing, travel restrictions),
• product changes (new checkout flow, pricing transparency),
• supply constraints (host churn, calendar availability), or
• measurement issues (delayed event ingestion, cancellations not reflected).

You have access to historical forecasts and actuals, plus event-level product logs and booking lifecycle data. The VP of Finance asks: “What strategies would you use to improve the accuracy of your forecasts, and how would you prove they work?”

Data Available

Your Task (Hard)

1. Define the primary forecast accuracy metric you will use (e.g., WAPE vs MAPE vs MdAPE) and justify it for a marketplace with cancellations, outliers, and varying region scale.
2. Diagnose the recent accuracy degradation by decomposing forecast error into actionable drivers (demand, supply, conversion, cancellations, mix shift, data quality).
3. Propose at least 5 concrete strategies to improve forecast accuracy. Your strategies must cover:
   • metric/definition fixes (what “actual GMV” means, treatment of cancellations and FX),
   • model/process improvements (retraining cadence, backtesting, horizon-specific models),
   • leading indicator instrumentation (what to monitor daily to anticipate weekly GMV), and
   • operational controls (data freshness SLAs, alerting, model versioning).
4. For each strategy, specify how you would measure impact (offline backtest design, holdout periods, segment-level evaluation, guardrails).
5. Recommend a forecast review dashboard: which 6–10 metrics you’d show weekly to Finance and Product to explain why the forecast moved.

Constraints:

• You must deliver an initial plan in 10 business days.
• Forecasts are used for budget decisions; explainability matters.
• Changes must not increase “accuracy” by simply smoothing volatility (Finance needs responsiveness to real shocks).