The scoring system most commonly used to prioritize people experiencing homelessness for scarce housing and services—the Vulnerability Index - Service Prioritization Decision Assistance Tool (VI-SPDAT)—is widely seen as flawed. Using machine learning, we develop a simplified, data-driven algorithm that easily outperforms the VI-SPDAT. We then test the effectiveness of this new tool by conducting a randomized controlled trial comparing data-driven prioritization with the vulnerability ratings given by skilled assessors. Outcomes of interest include housing stability, returns to homelessness, creditworthiness and use of credit, criminal justice involvement, employment, and income. The RCT began in November 2022 and preliminary results are expected in winter 2023-24; thus far, several thousand individuals have been prioritized. A second phase of the study will allow clients to express preferences over placements. Results will inform the development of an optimal hybrid tool combining insights from data, experts, and clients. The objective is to reduce homelessness and improve housing stability by identifying the people at greatest risk of remaining homeless without housing assistance.