The use of artificial intelligence (AI) and natural language processing (NLP) in the public sector is expanding rapidly, creating new opportunities to improve the accessibility and efficiency of government services. However, designing AI chatbots for government services requires attention not only to technical requirements but also to the distinctive characteristics of the public sector, including trustworthiness, transparency, security, privacy, and accountability. Accordingly, this study aimed to design and qualitatively evaluate a conceptual architecture for AI chatbots supporting the effective delivery of government services, with a focus on identifying key design requirements, design elements, an appropriate architectural approach, and evaluation criteria tailored to the public sector, particularly in the provision of legal services in the cooperative sector.
The study adopted the Design Science Research (DSR) paradigm and employed a qualitative, multimethod approach comprising a systematic literature review, Abductive analysis, exploratory interviews, and qualitative evaluation through a focus group. Using the PRISMA protocol, 112 articles were reviewed and analyzed. Synthesizing the review findings with deductive analysis resulted in a comprehensive framework of functional and non-functional design requirements. Functional requirements included natural language understanding, accurate response generation, dialogue management and personalization, interoperability with other systems, learning and feedback capabilities, human operator handover, authentication, and access control. Non-functional requirements included reliability, availability, security, privacy, transparency, maintainability, flexibility, efficiency, and usability.
Furthermore, interviews with 10 AI experts and enterprise chatbot designers identified key design elements, including front-end and back-end components, user interface design, natural language processing components, semantic search, dialogue management, knowledge bases, application programming interfaces (APIs), real-time monitoring, event logging and analysis, response generation, and security and privacy mechanisms. Based on the findings, Retrieval-Augmented Generation (RAG) was identified as a proposed architectural approach to support responses grounded in authoritative government documents. Subsequently, exploratory interviews with 15 AI chatbot experts and public-sector stakeholders informed the identification of evaluation criteria across three dimensions: system quality, information quality, and service quality. The proposed conceptual architecture was then qualitatively evaluated by a focus group comprising 17 AI experts and public-sector stakeholders in terms of its suitability for legal service delivery in the cooperative sector and its alignment with these three quality dimensions. The evaluation findings indicated that participants perceived the proposed architecture as conceptually suitable for the intended context. This study provides a conceptual framework for designing government AI chatbots that can serve as a basis for the subsequent implementation and empirical evaluation of AI-based systems for public service delivery.