Generative and Predictive AI for Decision Making and Retail Optimization

The increasing integration of Artificial Intelligence (AI) across many industries has radically transformed the way organizations access and act upon data. In the field of Business Intelligence (BI) this shift has significantly moved analytical practices from static reporting to dynamic, predictive and generative approaches. Among these impactful advancements, Generative AI (GENAI) and Large Language Models (LLMs) technologies are the most significant ones. This thesis projects aims at exploring how the implementation of a GENAI-powered chatbot called Genie (integrated within the Databricks platform) and its coexistance with predictive models is capable of enhancing decision-making processes and business optimizations in the retail sector. Specifically, this powerful tool has been deployed and tested for EssilorLuxottica, a giant leader in this domain. The research pursues two main objectives. First, to evaluate predictive performances of both a classical linear regression model and a Long Short - Term Memory (LSTM) network architecture, with the objective of checking whether the latter can better capture long distance patterns. Integrating such predictive models into the GENAI assistant, enables it to answer forecasting questions. To observe how the two models behave under different conditions, tests have been performed both for 13-week and 26-week horizon forecasts. Second, to assess how this AI-driven interface is capable of accelerating strategic decisions by making analytical and predictive insights accessible also to non-technical users across the company. Through this use case, this study highlights how the synergy between GENAI and predictive modeling can bring agility and real-time support. The results lay the groundwork for future developments, including finer-grained daily forecasting and the integration of broader informational capabilities into a single, intelligent assistant.