Material forecast is the process of deciding which items to stock in the inventory, how much, and when. Aim of the forecast is to increase parts availability with the less possible impact on inventory, having enough stock in the warehouse to ensure the business keeps moving but not enough stock to drain its limited cash reserves. This decision process is being profoundly revised in its foundational concepts, thanks to new classification methodologies enabled by Machine Learning. We integrated domain knowledge and ML to create a new classification and level setting process, leveraging on 6 years of data and new statistical indicators for demand patter. These new features are used to run the machine learning algorithm that classify Make To Stock / Make To Order items in a single flow approach. The validation phase is reduced at each iteration as ML model can be re-trained to incorporate past validations, increasing efficiency and performances. Level setting problem is addressed benchmarking ML methods (Reinforcement Learning), Montecarlo simulations and traditional statistical methodologies. Regarding RL and Montecarlo we established punishments for letting an particular inventory item run out of stock and we also punish the model for stock too higher value for too long. For rewards, we primarily focus on ordering items within a safe window before the demand. First application of this new methodology brings a 20% reduction of inventory, without impact on sales, and a workload reduction of about 70%.
Feedback form: https://python.it/feedback-1554
in __on Saturday 4 May at 12:15 **See schedule**