Companies in the field of sales and distribution of reading material, which have the problem of great variability in the demand for their products, which causes higher costs and customer dissatisfaction. Deep Learning, which is a branch of Machine Learning which is responsible for training a computer to learn on its own, works by recognizing patterns using many processing layers and learns through repetition in its training. Using the KDD methodology, we have implemented a deep learning algorithm to improve the logistics management of a reading material distribution company through demand forecasting and other aspects of logistics. It was shown that with the use of certain functions in the neural network of the algorithm it can be predicted with a low deviation and a minimum level of error, giving us results for 85% of improvement in demand prediction, also an 80% of improvement in material devolution, increasing the precision of the company on buying material aiming to get the exact amount of material to distribute. However, we can improve the results for a more accurate prediction by adapting the algorithm to the situation. Also, we share our challenges, and other opportunities for future research. This paper can help other investigators being a guide for beginners in deep learning application in logistics management.