Leveraging LLMs for Conversational Data Access: A Human-Centred Perspective

The study reported in this paper advocates the utilization of LLMs to enhance conversational data exploration. Our approach harnesses LLMs not only for crafting SQL queries but also for generating visualizations, data summaries and explanations that can enrich the conversational user experience. Building upon user studies involving a total 32 domain experts—middle managers and IT project managers in medium to large IT and manufacturing companies—we approached the problem with a human-centered perspective, and designed a Web platform for querying data using natural language. This paper illustrates the platform’s design, emphasizing the integration of LLMs within the data access pipeline, and the user-based evaluation we carried out both from a quantitative and a qualitative perspective. Furthermore, it explores the implications of adopting the conversational interface on the efficiency and satisfaction of users seeking to retrieve and analyze data through self-service platforms catering to situational requirements.