Purpose: A company creates an inventory in order to be able to continuously produce and fully meet the demand for its products. It thus covers the time and quantity discrepancy between articles of the supplier-buyer chain, and if applicable, between the different phases of production. But the very existence of inventories equals financial costs for their acquisition and maintenance, as well as economic costs stemming from the fact that the funds bound to inventories cannot be used more effectively by the company. Therefore, inventory management searches for the best volume of reserves that generates the lowest possible cost while being able to satisfy customer demand. There are currently a number of methods of inventory management (e.g. ABC), which are more or less effective. Neural networks also appear to be one of the methods of optimizing company inventory. They are useful for the prediction of time series. The aim of this article is to verify the possibility of using neural networks for inventory management using an example of a particular company. Design/methodology/approach: The data file contains the time series of a particular company's inventory covering the period of the last two years. Data are fitted to a curve that deviates the least from the actual data and that can also predict the future development of the state of inventories. The differences between reality and the obtained regression curve are measured by the method of least squares. To obtain the regression curve, neural networks were used, specifically, 10000 multilayer perceptron networks were generated. The best five were selected using the least squares method. These subsequently underwent a factual test of reliability and the most appropriate curve for the given company was determined. Findings: A neural structure that describes the development of company inventory very accurately was obtained, and it is therefore apparent that it also has the ability to predict the development of company inventory in the future. The acquired perceptron network was subjected to factual analysis and arguments for and against its use in practice were given. Research/practical implications: On the basis of the neural structure, the business can predict not only the future state of inventory, but also its movement in time. By doing so it will be able to reflect not only the needs of production and producer demand, but also fluctuations in the supply of inventories. It will be able to determine the optimal safety inventory, and if applicable, the optimal technological inventory, and obviously also the appropriate delivery cycles of inventories. The procedure for obtaining the neural structure is also applicable to other businesses. Originality/value: The additional value of the article can be seen in the use of neural structures in inventory management, and therefore in achieving a higher degree of accuracy of the obtained results.
