Many companies use experts to develop judgmental forecasting for use in their operations. Quite often forecasting is then based on the views of analysts or salespeople. However in general, human beings are not good at forecasting. There are several reasons, but the most important are:
- Carrying out endless forecasts is boring. People eventually cut corners leading to hasty assumptions greatly increasing the risk of errors.
- People tend to give undue weight to the most recent data. It is human to see a trend when sales of a product have been increasing for three consecutive days even though, in all probability, the increase is a purely random variation.
- People tend to look at too narrow selections of data. Many recurring situations have their own demand profile as seen with different seasons or days of the week, or even days of the week during a given season. For instance consumer behaviour around Easter can be modelled with a considerable degree of accuracy by using relevant comparable data from several years previously. Even if people had the focus and patience to do this they would struggle to absorb and process the sheer volume of data.
So when humans are carrying out large numbers of forecasts they have the tendency to be quite inaccurate and too reactive. Those errors can be quite large, and often miss behaviour patterns that show up in the data.
The best practice is to calculate and use statistical forecasts as a basis for all operations such as replenishment, promotions management, product introductions and so forth. Human input is only used to increase accuracy for specific situations and periods. However we find, with those of our customers that use a manual check and correction process, that their forecasts are considerably more accurate before any ‘expert correction’. The best results are received when manual input is restricted to times and situations where it can actually increase forecast accuracy. I’ll give you further details in my next blog post.