Definition
Signed forecast bias is the average directional error across a defined set of forecasts. Using actual demand minus forecast, positive mean error indicates underprediction and negative mean error indicates overprediction. The sign convention, forecast horizon, and measurement unit must accompany the result.
Why It Matters
- Repeated underprediction can leave buyers short of stock even when a headline accuracy score looks acceptable.
- A Commerce Intelligence OS needs to distinguish error direction from error magnitude before recommending a planning adjustment.
How It Works
- Retain forecasts as issued and align them with later observations at the same horizon. Flag stock-constrained sales instead of treating them as fully observed demand.
- Subtract each forecast from its corresponding actual and average the signed differences. Keep units consistent and disclose any weighting.
- Inspect SKU and location groups separately. Positive and negative errors can cancel in a portfolio average, so show absolute error alongside bias.
- Investigate systematic direction before changing the model or buyer override. Evaluate proposed corrections on subsequent periods rather than correcting from one unusually busy week.
Ecommerce Example
Context: Illustrative example: three weekly forecasts are 100 units each, while actual demand is 110, 120, and 100 units.
Recommended move: The errors are 10, 20, and zero units, giving a mean error of positive 10 units per week under the stated convention.
Why it matters: Review the persistent shortfall before changing replenishment. These hypothetical figures describe a diagnostic, not an iKawn forecast result.
iKawn Framework
Align
The iKawn framework connects issued forecasts with later demand evidence.
Measure
Preserve direction, magnitude, horizon, and units.
Investigate
Locate the products and planning assumptions behind systematic misses.
Validate
Check whether a correction improves future ordering decisions.
Concise Summary
Signed bias identifies average error direction. Pair it with error magnitude and segment detail so cancellation does not conceal planning problems.