On the accuracy of judgmental interventions on Statistical Forecasts

Forecasting at the Stock Keeping Unit (SKU) level in order to support operations management has proved a very difficult task. The levels of accuracy achieved have major consequences for companies at all levels in the supply chain; errors at each stage are amplified resulting in poor service and overly high inventory levels. In most companies, the size and complexity of the forecasting task necessitates the use of Forecasting Support Systems (FSS). The present study examines monthly demand data and forecasts for 753 fast moving SKUs, collected from three major U.K. suppliers. The companies rely upon FSSs to obtain baseline forecasts per SKU for each period. Final forecasts are produced at a later stage through the superimposition of judgments based on marketing intelligence gathered by the companies’ forecasters. The benefits of the intervention are evaluated by comparing the actual sales both to system and final forecasts as well as two simple benchmarks. The findings support that adjustments do improve accuracy with an overall gain of 3.72 (MdAPE) that is a percentage improvement of 18.9% (final forecasts being better in 55% of the cases). The accuracy gain differentiates according the size and the direction of the adjustments, and the level of noise in the series; two interesting findings: a) in almost one out of three cases the forecasters got the direction of the adjustment wrong, and b) small adjustments - less than 10% - should be avoided!