Fabric data agents: Adoption and development of Fabric data agents in organizational data analytics

Abstract

Data-driven decision making has become a central capability for modern organizations, yet access to data insights is often constrained by technical barriers, specialized tools, and limited user expertise. Advances in artificial intelligence, particularly large language models and agent-based systems, have created new opportunities for conversational data analytics, allowing users to interact with organizational data using natural language. Microsoft Fabric data agents enable users to build their own conversational data querying systems by using generative AI. However, as a recently introduced technology, there is limited research on their practical adoption, usability, and real-world value in organizational settings. This thesis addresses this research gap through a constructive study focusing on the adoption, implementation, and evaluation of a Microsoft Fabric data agent within an existing organizational data platform. The study examines the technical and organizational requirements of deploying a data agent, the limitations of the solution, as well as its analytical capabilities compared to traditional business intelligence workflows. The research is conducted in a pre-existing organizational data platform with established data engineering practices, enabling the agent to be integrated into a realistic enterprise analytics environment. The implementation demonstrates that adopting a Fabric data agent is feasible and efficient when supported by a well-structured data architecture and mature data platform. The agent’s adoption required relatively low amount of technical effort, provided that suitable data sources were available. With well-defined source data and curated example queries and instructions, the agent was able to provide accurate and relevant responses to natural language questions. The agent showed promising performance in exploratory data analysis and insight retrieval, allowing users to obtain answers through conversational interaction. The integration with other applications and channels further enhanced conversational analytics capabilities by enabling the most common data visualizations in the output. The results indicate that while Fabric data agents cannot fully replace traditional business intelligence tools in all use cases due to limitations in governance features, visualization capabilities and output reliability, they can effectively complement existing analytics solutions. The thesis concludes that Fabric data agents can provide non-technical users with wider access to data analysis through natural language conversations. The scope of the evaluation case in this study was limited. To get a more comprehensive picture of the capabilities of agent-based analytics solutions, further research and testing across different organizational contexts and use cases is required.