The forecasting of future sales volumes is critical to be successful in many industries. Due to an ever increasing complexity in the time series under consideration, traditional forecasting methods are reaching limits in their performance. Therefore, new approaches are needed to continue this vital task. The goal of this master thesis is to evaluate the use of neural networks to forecast logistic time series, specifically the determination of primary demands in material requirements planning. The operational processes and weekly customer demands of an Austrian food company will form the basis of this approach. The main objective is to decrease inventory while still maintaining the minimum delivery capability desired by the company. To evaluate the feasibility of using neural networks to determine the primary demands, the effects of the forecast on inventory and delivery capability were compared to the results of the current dispatch planning. The effects of the forecasts were mesured for the year 2021. A simulation of material requirement planning was used to calculate the inventory. The results of the evaluation show potential for the use of neural networks in material requirements planning, both to reduce total inventory and in relation to the different disposition procedures used by the company.