Insight
The forecast accuracy number that actually matters.
MAPE flatters and misleads. What to measure instead, and how to tell if your forecast adds value.
WERISCO editorial1 September 20267 min
Ask five planners how accurate their forecast is and you'll get five numbers computed five ways. Most of them are MAPE averaged across SKUs, a measure that over-weights tiny, erratic items and says little about business impact.
Three better questions
- Weighted error: is the forecast accurate where the volume and value are? Weighted MAPE (wMAPE) weights error by volume.
- Bias: does it lean consistently high or low? Bias creates excess or shortage even when error looks acceptable.
- Value added: does it beat a naïve forecast (last period, or same period last year)? If not, the process is adding cost, not accuracy.
A forecast that can't beat 'same as last week' isn't a forecast. It's a ritual.
Measuring over the right horizon
Accuracy one week out is easy. The decisions that matter (purchasing, production, capacity) are made further out. Measure accuracy at the horizon where you commit.
With 12 to 24 months of history, WERISCO can measure how your forecast error has behaved: wMAPE, bias and skill versus a naïve baseline.
Run the Risk, every week.
One supply-chain shock, decoded in numbers, and what a data-driven plant does about it.
