If you have ever sat through a quarterly planning meeting and wondered how seriously to take the projections on the slide deck, you are not alone. Studies from the Institute of Business Forecasting suggest that demand forecasts at the product level carry a mean absolute percentage error of between 20% and 40% in most consumer goods sectors.
Where the numbers tend to hold up
Aggregate forecasts - think total revenue for a division rather than unit sales for a single SKU - perform considerably better. At that level, errors often fall below 8% to 12%, which is genuinely useful for capital allocation decisions. The problem is that most operational decisions require granular accuracy, not aggregate comfort.
Three patterns that show up repeatedly in the data
- Forecast error increases sharply beyond a 13-week horizon in most retail and manufacturing datasets.
- Models that incorporate external economic indicators outperform pure time-series models by roughly 15 percentage points in volatile periods.
- Human overrides of statistical models introduce bias in about 6 out of 10 cases studied by Fildes and Goodwin in 2007.
The honest takeaway is not that forecasting is useless. It is that the value sits in understanding the error range, not in treating the point estimate as a plan. A forecast with a clearly stated confidence interval of plus or minus 18% is more actionable than a precise-looking number with no stated uncertainty.