Pietro Calabrese

Session 3: Function Point Analysis in the Age of AI: Rethinking Boundaries, Complexity and Effort
Time: 11:00 am – 11:30 am

Session Abstract:
Function Point Analysis assumes that elementary processes are repeatable, that logical data is user-recognizable, and that functional size is independent of the delivery technology. AI-based systems appear to violate all three. This presentation argues that much of the difficulty stems from boundary placement rather than from the technology itself.

The first part addresses the sizing of AI-based systems, weighing three strategies. Drawing a separate boundary around the AI component makes the inference service an identifiable elementary process, an EO, given its derived, algorithmic processing crossing the boundary and its knowledge base an EIF from the host, an ILF within. FPA counts unique elementary processes, not executions, so conversational volume does not inflate the count. The second strategy allocates AI-specific complexity to SNAP: Data Entry Validations for guardrails, Logical and Mathematical Operations for embedding and ranking. The third, more radical, adds a complementary dimension for model and uncertainty complexity.

One consequence of the boundary decision deserves attention: prompt templates, guardrail rules and retrieval configurations maintained by business users may qualify as user-maintained logical data, with the transactions maintaining them work now dismissed as invisible, yet countable under the unmodified standard.

The second part addresses software developed with AI. Technology independence implies size remains unchanged; the effect therefore concentrates in hours per function point, the term governing FP-based contracts and benchmark repositories. Output-based pricing must then decide who captures the productivity gain.

AI does not make functional sizing obsolete; it separates what is built, how it is built, and how complex it is to build.

Speaker Bio:
Pietro Calabrese is a PhD Candidate in Information Engineering at the University of Pisa, where he is a member of the Machine Learning and Process Intelligence (MLPI) research group at the Department of Information Engineering.

He holds a Bachelor’s Degree in Computer Engineering and a Master’s Degree in Artificial Intelligence and Data Engineering, both from the University of Pisa, with a master’s thesis on counterfactual-based feature importance measures. His research centres on Explainable AI, aiming to improve the fidelity and comprehensibility of machine learning models — a concern closely related to the questions of determinism and measurability raised in this presentation.

Alongside his academic work, he has practical experience as an IT Specialist and Database Administrator, implementing scalable cloud computing architectures and developing optimized software solutions. His applied projects span anomaly detection in industrial environments, IoT management for energy efficiency, and the classification of medical images.

This combination of perspectives informs his interest in software measurement. Having built AI systems in production, he approaches functional sizing from the side of what such systems actually contain, rather than from the standard outward, and is particularly interested in how established measurement frameworks can absorb probabilistic components without losing comparability across portfolios.