Accuracy and bias answer different questions
Accuracy shows how far the forecast was from actual demand. Bias shows whether the team repeatedly forecast too high or too low. A portfolio can appear balanced while over-forecast errors in some SKUs cancel under-forecast errors in others.
Use a consistent scorecard
- Original forecast, updated forecast, and actual units for the same time bucket.
- Absolute unit error and percentage error where volumes are large enough.
- Signed error to reveal high or low bias.
- Promotion, price, stock availability, and campaign notes.
- Inventory value, margin, and lead time to show business impact.
Separate forecast error from constrained sales
Actual orders do not equal unconstrained demand when the item was unavailable, delivery slowed, or advertising was intentionally reduced. Flag these periods instead of teaching the next forecast that suppressed sales were normal demand.
Improve the process one cause at a time
Review the largest value-at-risk errors and assign a cause such as event lift, launch optimism, price change, stockout, or lead-time delay. Update the relevant assumption rather than applying one blanket adjustment to every SKU.