Developing Predictive Risk Models: Lessons from a Sandbox Experiment in Child Welfare Services

Societies are facing a crisis of credibility, authority, and trust in experts (Heimstädt et al., 2024) as algorithmic decision-making technologies increasingly permeate public services. Technologies offer benefits but also introduce risks, aligning with Beck’s "risk society" paradigm (1992). Their integration complicates the inherently non-linear nature of decision-making, raising critical questions about their impact on organizational practices and roles, particularly in social work (Glaser et al., 2024). This study presents findings from a sandbox experiment exploring predictive risk modeling for identifying child maltreatment in child welfare services. The simulated environment combined diverse methods to examine how social workers interpret and engage with predictive risk modeling, highlighting implications for practice, organizational change, and policy development. These insights contribute to debates on human-machine augmentation in public administration and its influence on professional discretion and decision-making.