The company used a demand forecast that had been optimized over many years and achieved good performance. However, the solution used was not able to process the large number of factors influencing demand. In particular, the existing solution was inadequate for slow-moving items and items with asymmetrical fluctuations around the demand forecast value. As the demand forecast is the cause of procurement and production, an improvement was sought.

The solution created uses pattern recognition methods - technically neural networks. By using these, all relevant factors influencing demand could be taken into account. In addition, the method is better able to deal with slow-moving items and asymmetrical fluctuations.

The first step was to compare the forecasting accuracy of the created solution with the current forecast.

After successfully establishing the increased precision and the associated benefits, the solution was installed in the customer environment. The forecast results are written back to the customer's systems, meaning that hardly any procedural adjustments were required. The installation at the customer's premises means that no data leaves the company.

The model is regularly retrained by us to take account of new market conditions.