Stuck in the Pilot Phase: Challenges in Using Machine Learning in Public Welfare Services

Although artificial intelligence (AI) is heralded as a transformative force, the challenges of integrating the technology into public welfare services remain underexplored. Many AI projects hold great potential but stall before full implementation, failing to progress beyond the pilot phase. In response, we have examined machine learning (ML), a subset of AI techniques, in the context of eldercare, a key public welfare service, where an aging population and labor shortages have intensified the demand for technological solutions. Drawing on a qualitative study in an eldercare department in a Danish municipality and guided by a sociotechnical framework, we investigated two projects that applied ML to optimize work planning and client rehabilitation. Despite high expectations, funding, and strong strategic support, neither project progressed beyond the pilot phase into daily operations, making them compelling examples for understanding the challenges in using ML in public welfare services. Our analysis reveals eight challenges, shaped by the interplay among technology, people, structure, tasks, regulation, and ethics. By implication, we offer new insights into the multifaceted and sociotechnical challenges of using ML in public welfare services.