Neural networks – empirically tested in DISKOVER 

In addition to statistical methods of sales forecasting, AI methods such as neural networks are becoming increasingly important. But even neural networks have to deal with the challenge of not being able to predict chaotic market behaviour. Unfortunately, the behaviour of markets and customers is almost always chaotic to a certain degree. 

Chaos cannot be forecast, but the extent of the chaos can be measured and compensated for by safety stock. The battle of prognostics is therefore won with safety stocks. 

Since forecasting methods and safety stock methods are intertwined, the success of a demand planning method depends on the forecasting method harmonising well with the safety stock method over the dynamic course of time.  

To find the right combination of forecasting method and safety stock method, DISKOVER therefore relies on a dynamic empirical simulation based on real data from the past. In DISKOVER, forecasts based on artificial intelligence must also undergo this simulation to test their resilience for practical use in the planning and scheduling process. DISKOVER regularly simulates, independently decides which process combination is best suited for an article and automatically applies it; a decisive success factor for reliable forecasts and thus automatable processes.