Optimizing Semantic Architecture for Agentic Business Intelligence and Conversational People Analytics in Microsoft Power BI

Abstract

Human resource teams often work with information spread across recruitment, employee information, payroll, and enterprise systems. Bringing these sources together is necessary for useful people analytics, but it creates challenges for natural-language reporting when semantic relationships and measures are unclear. This study examines how the structure of a Microsoft Power BI semantic model affects conversational people analytics. A Star Schema was designed with clear relationships, organized metadata, and DAX measures for important workforce calculations. The design was compared with a flat model using the same workforce data and 50 conversational prompts. The reported results show that the flat model successfully compiled 64% of prompts, while the proposed model achieved 100%. Semantic determinism improved from 42% to 98%, and average processing time decreased from 1,480 ms to 115 ms. The paper also discusses Dynamic Row-Level Security, data minimization, privacy, and responsible use of people analytics. The findings highlight the importance of explicit relationships, governed measures, and clear metadata for reliable conversational business intelligence.