Forecast accuracy is not enough
Accuracy compresses different planning failures into one number. Bias shows which direction the system is consistently getting wrong.
The problem
A forecast can look acceptable in aggregate while repeatedly creating the same operational failure. Average error measures size; it does not reveal direction.
Decision principle: use accuracy to understand magnitude and bias to understand behaviour.
A practical model
Track weighted absolute percentage error alongside signed forecast error. Segment both by product family, planner, horizon, and demand pattern.
| Signal | What it reveals | Useful response |
|---|---|---|
| High error, low bias | Volatility or weak signal | Review model and aggregation |
| Low error, high bias | Consistent directional miss | Correct systematic assumptions |
| High error, high bias | Structural forecasting issue | Rebuild process and ownership |
Recommendation
Review bias as a distribution, not only as a portfolio average. Opposing errors can cancel each other and hide where intervention is required.
Key takeaways
- Accuracy and bias answer different questions.
- Segmentation matters more than a single headline score.
- Metrics only create value when paired with a decision rule.