Database Chatbot Success: User-Centered Design With LLMs

Natural Language Data Query Solutions present a significant opportunity for business success. They offer unprecedented responsiveness to decision-makers by naturalizing and empowering their interaction with data. However, these solutions face barriers to acceptance among business professionals. The reluctance is due to the system-centered design approach followed. In contrast to user-centered design, this approach does not take into account users’ requirements, e.g. system explainability, awareness of the business context and user profile, or data security. Based on a User Study, this paper identifies, through 19 interviews with business professionals and IT experts, the user-centric requirements that drive the adoption of such solutions in the decision-making process. We then propose a generic framework, B2Data, which is aligned with these requirements. B2Data leverages the capabilities of Large Language Models. Finally, the validation of this architecture through technical implementation on medical data and 8 user tests demonstrates both the feasibility and acceptance of B2Data.