An Iterative Framework for AI Adoption in the Public Sector: Insights from a Swiss Case Study

Abstract

In the step from prototype to the productive, the sustainable use of Artificial Intelligence (AI) applications remains a significant challenge. This creates a demand for frameworks that guide organisations and help to bridge this gap in specific domains. This study validates the Adoption Framework for AI in the Public Sector (AFAIPS) by examining its application in a Swiss AI chatbot project for healthcare cost verification, addressing how an iterative framework can be used to navigate public sector constraints. Using a detailed single case study within an Elaborated Action Design Research (EADR) setting, this research recorded a 34-week AI chatbot implementation over eleven iterations, drawing evidence from project documentation and user interviews. The findings show that AFAIPS effectively guided the project through its adoption phases, enabling user-driven development and flexible adjustments. Iterative prototyping and continuous feedback were essential for handling public sector limitations and technical uncertainties. Key results included organisational learning, such as the development of "AI-ready" documentation, and strategic changes like planning for external Machine Learning Operations (MLOps) support to address internal capability gaps. This study provides empirical validation of AFAIPS in public healthcare administration, offering new insights into how an integrated framework — combining the Diffusion of Innovations theory (DOI) with Design Thinking (DT), Lean Startup (LS), and MLOps connects theory with the practicalities of government AI implementation, to improve public value. For public managers and policymakers, this research offers a validated roadmap and practical strategies for stakeholder engagement, iterative development, and agile management to guide digital transformation and AI governance.