Pre-launch new product demand forecasting : from the Bass diffusion model to a new machine learning-based model

Nowadays the manner in which advanced analytics can help companies to reach a higher efficiency as well as a higher effectiveness is increasingly more evident. There is a growing attention to machine-learning techniques that allow both the formalization of previous knowledge and the extraction of information from data. At the same time, there is an increasing need of companies to know better the final customer and to forecast the final demand. The most relevant but, at the same time, the most difficult forecasts are those before the launch of new products. They are the most difficult because there is not historical data to extract information about the final customers that will buy the new products. However, they are the most relevant because, knowing the final demand, all the pre-launch decisions can be taken in a less uncertain scenario, optimizing the resources allocation and better managing the entire Supply Chain. This thesis aims to develop a new machine-learning based model, able to predict the demand of ne products before their launch. The results of this study reveal the accuracy of the model, tested on real cases, as well as the added value that it could generate in a company.