A Comparative Analysis of LLM-Enhanced Business Intelligence Tools at Amadori

In the contemporary data-driven landscape, the usability gap in Business Intelligence tools represents a critical point. Despite vast information availability, non-technical stakeholders often lack the knowledge to extract insights independently using complex analytical tools. This thesis investigates the potential of Generative Artificial Intelligence and Large Language Models (LLMs) to bridge this divide by enabling Natural Language Querying (NLQ). Conducted within the operational context of \textbf{Gruppo Amadori}, a leader in the Italian agri-food sector, this research performs a rigorous comparative evaluation of three distinct AI architectures: the commercial ``ready-to-use" solution Strategy Auto (from Microstrategy), the custom cloud-native prototype Dinova and the ecosystem-integrated assistant Google Gem. The methodology relies on a foundational Data Governance framework, demonstrating that the creation of a Business Glossary and Data Dictionary is not merely a preparatory step but the essential ``semantic engine" required to mitigate model hallucinations and ensure accuracy. The tools were stress-tested against a realistic dataset modeled on a Snowflake schema, utilizing a standardized protocol of 24 business queries of increasing cognitive complexity. The empirical results indicate that while commercial tools offer immediate stability for executive reporting, custom solutions provide superior cost-efficiency and architectural flexibility. In conclusion, this work demonstrates that the effective democratization of BI depends less on the specific AI model and more on the maturity of the underlying semantic layers, proposing a strategic hybrid approach and identifying ``Agentic AI" as the future evolutionary path for autonomous enterprise analytics.