Boardrooms are now facing an informational asymmetry because management uses AI to produce and analyse information, reducing the board's oversight. In this thesis, an analysis was conducted to understand if the implementation of an agentic AI system can give back the independence to the board, while still maintaining compliance with the regulatory environment. Starting from the model of Ferreira and Li (2025) and combining the constraints of Article 2381 of the Italian Civil Code and the EU AI Act, three design requirements were derived: auditability, explainability, and human oversight. Then an AI agent called BoardMind was developed, running on a Streamlit interface, and evaluated under the three design requirements already mentioned. The test of the agent was conducted on Brunello Cucinelli S.p.A. The modules analysed were: financial analysis, what if scenario, stress testing, market intelligence, board deliberation simulation, and generative AI assistant. The evaluation shows that an AI agent operating under the constraints of the regulatory environment can provide boards with independent and auditable analysis, solving the problems identified in the literature.
I consigli di amministrazione si trovano oggi di fronte a un'asimmetria informativa, poiché il management utilizza l'IA per produrre e analizzare informazioni, riducendo la capacità di supervisione del consiglio. In questa tesi è stata condotta un'analisi per capire se l'implementazione di un sistema di IA agentica possa restituire l'indipendenza al consiglio, mantenendo allo stesso tempo la conformità con il quadro normativo di riferimento. Partendo dal modello di Ferreira e Li (2025) e combinando i vincoli dell'articolo 2381 del Codice Civile italiano e dell'EU AI Act, sono stati derivati tre requisiti progettuali: auditabilità, spiegabilità, e supervisione umana. Dunque è stato sviluppato un agente di IA chiamato BoardMind, che opera su un'interfaccia Streamlit, e valutato rispetto ai tre requisiti progettuali già menzionati. Il test dell'agente è stato condotto su Brunello Cucinelli S.p.A. I moduli analizzati sono stati: analisi finanziaria, scenario what if, stress test, market intelligence, simulazione di delibera del consiglio, e assistente IA generativo. La valutazione dimostra che un agente IA operante entro i vincoli del quadro normativo può fornire ai consigli di amministrazione un'analisi indipendente e verificabile, risolvendo i problemi citati nella letteratura.
AI in the boardroom
BRAGATO, ALESSANDRO
2025/2026
Abstract
Boardrooms are now facing an informational asymmetry because management uses AI to produce and analyse information, reducing the board's oversight. In this thesis, an analysis was conducted to understand if the implementation of an agentic AI system can give back the independence to the board, while still maintaining compliance with the regulatory environment. Starting from the model of Ferreira and Li (2025) and combining the constraints of Article 2381 of the Italian Civil Code and the EU AI Act, three design requirements were derived: auditability, explainability, and human oversight. Then an AI agent called BoardMind was developed, running on a Streamlit interface, and evaluated under the three design requirements already mentioned. The test of the agent was conducted on Brunello Cucinelli S.p.A. The modules analysed were: financial analysis, what if scenario, stress testing, market intelligence, board deliberation simulation, and generative AI assistant. The evaluation shows that an AI agent operating under the constraints of the regulatory environment can provide boards with independent and auditable analysis, solving the problems identified in the literature.| File | Dimensione | Formato | |
|---|---|---|---|
|
Bragato_Alessandro.pdf
embargo fino al 20/07/2027
Dimensione
1.93 MB
Formato
Adobe PDF
|
1.93 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/112754