Few numbers drive as many decisions as the demand forecast. It sets production, purchasing, hiring, inventory and the cash plan. Yet in many companies it is a single line, revised by negotiation, that nobody fully believes. Sales add a buffer, operations add another, finance trims both, and the final figure reflects the meeting more than the market.
Show the range, not only the line
Every forecast has an error, and the error grows the further ahead you look. A forecast that shows its uncertainty lets each decision use the part it needs: production can plan on the central line, safety stock can be sized on the upper band, and the cash plan can be tested against the lower one.
Measure accuracy and bias, every month
- Accuracy, for example the mean absolute percentage error, tells you how far off the forecast usually is.
- Bias tells you whether it is consistently too high or too low. A small bias is more damaging than a large random error, because it accumulates in inventory or in lost sales.
- Forecast value added compares each step of the process with a naive forecast. If a manual adjustment does not beat the statistical baseline, it is adding noise, not insight.
A process that earns trust
- Start from a statistical baseline built from history, seasonality and known drivers such as price and promotions.
- Allow adjustments only with a documented reason, and track whether each one improved the result.
- Forecast at the level where decisions are made, and reconcile up and down instead of splitting a total by habit.
- Publish accuracy and bias next to the forecast, so its users know how much to rely on it.
The goal is not a perfect forecast, which does not exist. It is a forecast whose errors are known, unbiased and small enough for the decisions it supports, and an organisation that plans for the range instead of hoping for the line.
Bartor S.
Founder
