Forecast accuracy is not the goal. Better capacity decisions are. Here is what changed the outcome in our logistics engagements.
1. Forecast the decision, not the number
Planners need to decide how many vehicles and drivers to schedule per zone and shift. A probabilistic forecast with a confidence interval maps directly to that decision; a single point estimate does not.
2. External signals beat model complexity
Weather, holidays, marketplace promotions and local events explained more variance than any change in algorithm. Get the calendar right before tuning hyperparameters.
3. Close the loop with the planner
The forecast improves fastest when planners can override it, and their overrides are captured and compared against reality. That feedback is training data for the next iteration and trust for the current one.
- #Logistics
- #Forecasting