Abstract

Public institutions, such as employment offices, face an ongoing tension between efficiently allocating limited caseworker capacity and accurately addressing the needs of the individuals they serve. In response to these pressures, algorithmic tools and artificial intelligence have generated significant interest as potential solutions to support and improve decision-making practices while operating in resource-constrained environments. For example, predictive models can identify jobseekers who have a low risk of prolonged unemployment, enabling caseworkers to better allocate their attention and resource where they are needed the most. However, integrating predictive systems into public administration introduces substantial methodological and practical challenges. We need to ensure that tools not only operate reliably but also effectively complement the expertise and decision-making capacity of human caseworkers. This research project investigates how algorithmic decision-support tools can improve caseworker decision-making and resource allocation within the German public employment services. Our analysis will center on the use of the Integrated Labour Market Biographies provided by the Institute for Employment Research (IAB). Using this data, we will construct several datasets that make use of employment histories and profiles of jobseekers to predict various unemployment outcomes. Based on these predictions, we will develop a triaging system that automatically identifies clear low-risk cases, while deferring uncertain cases to human decision-makers. Additionally, we will use historical profiling data from caseworkers to evaluate the effectiveness of such a hybrid human-algorithmic approach. We aim to establish robust guarantees for these systems that extend beyond predictive accuracy alone, such as stability, fairness, and overall efficiency in resource allocation. Furthermore, we will explore how to adapt themodel to diverse local institutional contexts by analyzing regional variation in decision-making and employment outcomes. Ultimately, this research represents a meaningful step toward responsibly integrating artificial intelligence into public employment services and directly contributes actionable insights to both policymakers and the broader field of scientific labor market and occupational research.