LUCA, the AI agent that transforms 78 years of basketball statistics into editorial content
The Lega Basket Serie A has preserved championship statistics since 1948: results from the first forty years, game summaries from the 1980s, and play-by-play data since 2008. This knowledge base had remained largely inaccessible until yesterday: to reconstruct a historical comparison between two teams, a journalist had to request a query from a league representative or manually create updated Excel sheets game by game. This asymmetry between data availability and the ability to query it gives rise to LUCA, the artificial intelligence agent presented on May 5 in Milan by Lega Basket Serie A (LBA), Infront, Microsoft Italia, and Impresoft 4ward.
An agentic system, not a single model LUCA is not a single generative model but a multi-agent architecture built by Impresoft 4ward on the Microsoft Foundry platform, which leverages the GPT-5.1, 5.2, and 5.3 models from Azure OpenAI. Three families of agents collaborate: one searches for correlations and patterns within the statistical data, another verifies that the outputs are genuinely grounded and not the result of hallucinations, and a third transforms the validated insights into ready-to-use content, such as articles, social media posts, or commentary sheets. Ultimately, there is human validation, as LBA has mandated that the published content is certified 100%. The data does not leave the Azure perimeter.
For Daniele Grandini, Chief Innovation Officer at Impresoft 4ward, the difference between a conversational assistant and a true agent lies in operational autonomy: a goal to be achieved, freedom in choosing tools, and work times measured in minutes or hours rather than seconds. "Artificial intelligence gives humans superpowers," Grandini summarized, marking the boundary between automation and collaboration: not replacing the analyst but enabling them to explore correlations that they would not have the time or tools to seek out alone.
On the operational front, LUCA generates about twenty statistical insights each week on the eight games of the round, from which dozens of content items are derived and distributed across the league's channels, clubs, or provided to journalists. The queries occur in natural language, eliminating the need to write code. LBA commentator Mario Castelli, during the presentation, demonstrated how a statistic generated by the system can enter the live game narrative as soon as a basket is made.
From use case to replicable model "Our goal is to offer fans exactly what they are looking for, and also what they don't yet know they want to discover," explained Massimo Cortinovis, Head of Digital at LBA, summarizing the editorial promise of the project. The project aims for continuous editorial production, distributed across the website and social media, with future expansions on the LBATV platform and paid channels, potentially generating commercial benefits for sponsors.
Stefano Deantoni, Marketing Director of Infront Italy, pointed out the scalable trajectory: "The Lega Basket Serie A is the first partner to have believed in and invested concretely in this vision." For the sports marketing company, the model is exportable to other leagues and federations in Italy and abroad.
Microsoft Italia presents the project as a showcase for the AI L.A.B. program, the acceleration initiative that has been guiding Italian companies in the adoption of generative and agentic AI for three years. "In sports, data is essential, but it often proves to be little accessible," noted Annamaria Bottero, Global Partner Solutions Lead at Microsoft Italia, reminding everyone of the role of the Italian channel partners, over 14,000, in implementing technology into business processes.
The process of seeking what is unknown within a structured database, validating it with a second agent, and transforming it into content is in fact transferable outside of sports. Grandini mentioned pharmaceutical research, where similar approaches are compressing the timeframes for testing new molecules. Economically, the investment is within reach even for small and medium enterprises: a total journey of six months, with the AI development phase focused in about two months once the data is prepared.
LUCA is being released in beta, aiming to be fully operational in the next season. The field data collection remains human, currently conducted by three individuals in venues, and the generation time for a single insight can reach twenty minutes: hence, the system operates in the background during the week and not yet live during the games.