Diego Rocha

Session 6: Sizing an AI Legal Assistant through the Function Point Analysis Lens
Session eligible for certification renewal activity
From multi-agent legal workflows to measurable Transactional and Data Functions
Time: 1:30 pm – 2:00 pm BRT – (ET time 12:30 pm – 1:00 pm)

Session Abstract
Artificial Intelligence is rapidly reshaping legal work through specialized agents that can triage case files, draft initial petitions and defenses, compare pleadings, support appeals, retrieve jurisprudence, and coordinate retrieval-augmented generation (RAG) over trusted legal knowledge. As these solutions evolve from generic chatbots into orchestrated multi-agent systems, a practical measurement question emerges: how should the functional size of an AI agent be identified and expressed?

This presentation answers that question through the Function Point Analysis (FPA) lens, using a single concrete scenario: an AI assistant deployed in a court chamber. The assistant is delivered behind one uniform conversational interface, yet it performs six activities that users request separately — case triage, conflict-and-duplicate detection, jurisprudence recommendation, draft ruling generation, appeal admissibility analysis, and plain-language rewriting of published decisions for citizens.

The measurement question is deceptively simple: is that one transaction, or several? A frequent shortcut treats the whole agent as a single transaction, on the grounds that its primary intent is “to assist the chamber”. That shortcut confuses the purpose of the application with the primary intent of an elementary process. Applied against the standard uniqueness criteria — data element types, referenced files and processing logic — the six activities prove to be distinct elementary processes: differing on a single criterion would already be sufficient, and these differ on all three.

The session sizes the scenario end to end, reaching 77 function points across four data functions and twelve transactional functions, and contrasts it with the 53 function points produced by the “one agent, one transaction” shortcut — a 31% difference that grows with the capability of the agent. The opposite error is treated with equal rigor: not every prompt is an elementary process, and an agent that internally chains several model calls and tool invocations to produce one deliverable is still performing one elementary process.
Join us and turn AI-agent complexity into measurable functional clarity.

Speaker Bio:
Diego Rocha is a software measurement and project management professional with extensive experience applying Function Point Analysis to software contracts and delivery environments. He has worked with function points for more than 15 years and is an active member of IFPUG’s Functional Software Sizing Committee (FSSC).

He is a Certified Function Point Specialist (CFPS) and a COSMIC Software Functional Size Measurer (CCFL), with additional background in requirements management, software estimation, data analysis, and project leadership. Diego has also shared his expertise with the IFPUG community as a speaker on Elementary Processes and User Stories.

His experience combines practical measurement, contracting, governance, and the application of functional sizing concepts to evolving software delivery contexts.