Top-Down Order Forecasting

Top-Down Order Forecasting


Through a whiteboarding session, redefined the methodology for the demand forecasting system used by Gorillas to predict hourly orders per location. The new approach incorporated macro factors such as marketing activity and weather, significantly improving forecast accuracy and leading to Gorillas UK becoming the most efficient country within the Quick Commerce industry.

Business Impact:

  • Significantly increased the accuracy of order forecasts, leading to a 20% reduction in RMSE.
  • Enabled Gorillas UK to achieve the highest operational efficiency within the company and the broader quick commerce industry.
  • Supported global teams in the international rollout of the improved forecasting methodology.

Key Features:

  • Replaced the previous hourly store-level time-series model with a top-down approach, producing country-level estimates based on marketing spend, weather, and seasonal factors.
  • Distributed orders based on historical splits of discounted orders, organic orders, and referrals to generate daily estimates per store which were then cascaded to hourly forecasts based on historic order distributions.