In an unprecedented information technology (IT) revolution in the public service sector, an increasing number of police departments use advanced statistical methods to improve their productivity in fighting crimes. Since 2007 the police department of Milan has been using a predictive policy software that is unique, as it not only produces aggregate crime predictions but also individual ones.

This paper uses detailed information on individual crime incidents, coupled with offender-level identifiers produced by the software, to show that criminals follow habits, and that such habits make their future actions predictable. Using quasi-random assignment of crimes to two police forces that differ in the availability of this predictive policing software, this study shows that the adoption of this very advance yet inexpensive IT innovation doubles the productivity of policing.