Machine learning is increasingly used in government programs to identify and support the most vulnerable individuals, prioritizing assistance for those at greatest risk over optimizing aggregate outcomes. This paper examines the welfare impacts of prediction in equity-driven contexts, and how they compare to other policy levers, such as expanding bureaucratic capacity. Through mathematical models and a real-world case study on long-term unemployment amongst German residents, we develop a comprehensive understanding of the relative effectiveness of prediction in surfacing the worst-off. Our findings provide clear analytical frameworks and practical, data-driven tools that empower policymakers to make principled decisions when designing these systems.

Governments are increasingly using machine learning to help people in need, for example, by predicting who is most at risk of long-term unemployment and offering them early support. But how effective are these prediction tools compared to other options, like hiring more caseworkers?In our research, we explore how well prediction-based systems actually identify and support the worst-off individuals, those who are struggling the most, rather than just improving overall program performance. We combine mathematical modeling with a real-world case study in Germany to evaluate when and how prediction can make the biggest difference.Our work shows that while machine learning can potentially help target support more effectively, its benefits depend on how it’s used and what tradeoffs are involved. In some cases, such as when resources are very limited or the prediction is already decently accurate, expanding institutional capacity may matter more. We provide clear guidelines and tools that help policymakers think through these choices and design more equitable support systems.