AI workforce of the future - Enhancing qualitative organizational data analysis through integrated GraphRAG and multi-agent AI systems with comparison to baseline AI

Abstract

The exponential growth of unstructured data presents a critical paradox for organizations and researchers: while the volume of rich qualitative information, such as interview transcripts and internal reports, is expanding, the capacity to analyse it using rigorous, interpretive methodologies remains con-strained by human cognitive limits. Traditional qualitative data analysis (QDA) is inherently unscalable, while standard Artificial Intelligence (AI) solutions, including Retrieval-Augmented Generation (RAG), often fail to capture the structural nuance and global context required for deep inquiry. This thesis addresses this gap through a Design Science Research Method (DSRM) study. The research designs, implements, and evaluates a novel artifact: a web-based multi-agent system empowered by Graph Retrieval-Augmented Generation (GraphRAG). This system structures unstructured text into knowledge graphs to support complex reasoning and employs a staged autonomy model to balance AI efficiency with human interpretive control. Evaluated through quantitative analytics and semi-structured interviews with organizational practitioners, the study demonstrates that while higher levels of agentic autonomy increase analytical depth, they introduce significant coordination costs. The findings suggest that "Human-in-the-Loop" (HITL) mechanisms are not merely safeguards but essential components for organizational trust. This research contributes to the intelligent agentic systems of the future by offering design principles for AI systems that scale qualitative and quantitative data analysis.