Improving demand and inventory forecast with data analytics techniques in a real manufacturing business case

The manufacturing industry has always had to deal with a contraposition in the context of the inventory management. On the one hand, it is necessary ensure the financial efficiency by trying to build up smaller inventories, in order to reduce costs of their management and to avoid company's assets slow and ineffective turn over. On the other hand, it is essential make sure that there is an operational continuity and, as a consequence, an high level of service. Therefore, companies are called to define a right and winning balance. In this fundamental task, data analytics techniques can help them, becoming a key factor in decision-making process and, more generally, an indispensable strategic weapon. Particularly, they may be a value added support to demand sizing and subsequent production forecasting, making them more effective and robust, translating, hence, into a competitive advantage. The purpose of this study is precisely to define, in a real manufacturing business case, the best data-based approach to demand and inventory forecasting. It starts from data exploration aimed at identifying the key insights that can lead to a better forecasting approach, taking into account also cost impact and periodicity of employment. Then a product clustering is carried out, to define clusters with similar behaviors, that must be treated separately and independently. Finally, the forecast modelling is implemented via 2 alternative and parallel methodologies (Multilinear Regression and Random Forest Tree regression) and results are compared to identify the best combination, leading to the highest accuracy for each cluster.