Over the past 25 years, even as New York City’s overall homicide rate declined by 85%, the share of domestic violence-related homicides more than tripled. In 2016, the Crime Lab partnered with the New York Police Department (NYPD) to develop a machine-learning model that predicts which domestic violence survivors are most likely to experience another domestic violence incident. The prediction model relies on machine learning to process over ten years of NYPD’s domestic violence incident data, including statements made by survivors and responding officers. These predictions were used to prioritize the NYPD’s outreach and response efforts and to improve the allocation of the city’s limited resources for domestic violence survivors. The model was successful in accurately identifying individuals at high risk of re-victimization, reinforcing the importance of machine learning as a tool for public service delivery and for answering research questions about the data needed to predict domestic violence incidents.