² Department of Public Administration, University of Accra, Ghana
Abstract
As governments worldwide have rapidly adopted automated decision-making systems (ADMS) to streamline the distribution of social welfare benefits, concerns regarding systemic algorithmic bias have intensified. This study provides a comprehensive retrospective of ADMS implementations across five OECD nations between 2020 and 2025, evaluating their impact on marginalized demographic groups.
By analyzing public audit reports and conducting qualitative interviews with affected individuals, the research reveals that algorithmic efficiency often comes at the cost of equity. Training data heavily reliant on historical welfare disparities resulted in disproportionate benefit denials for low-income single-parent households and specific ethnic minorities. The paper concludes with actionable policy recommendations for establishing algorithmic auditing boards and maintaining mandatory human-in-the-loop appeal processes for all automated public sector decisions.