As demands for greater policing accountability and outcomes continue to increase, police departments across the nation are embracing predictive analytics to not only increase efficiency in operations, but also improve crime prevention and response. Yet with all the potential, most leaders struggle with critical start-up questions such as: How do we start? Where should the methods be applied? What is the impact on the community? How do we test and scale the initiatives?

During The 2018 Public Safety Summit: Leadership in Turbulent Times, Evan Levine from the New York Police Department, Jonathan Lewin from the Chicago Police Department, and Sean Malinowski from the Los Angeles Police Department shared lessons learned on implementing emerging data science and “predictive policing” into public safety strategies. The following synthesizes some of the most important leadership lessons learned.

1. Forge a diverse team to lead the effort.

   Civilian analysts can help train police chiefs on the daily practice of analyzing data from cameras, maps, and other technologies, and incorporating it into resource deployment. For example, in New York City, Chicago, and Los Angeles, analysts helped create maps, interpret data, and suggest inputs. However, once trained, policing leaders led daily discussions interpreting the data and informing analysts on how to refine algorithms. Police leaders need to understand inputs on algorithms, and have the freedom to reject illogical suggestions. Hiring analysts internally (with proper training) can also be a good idea, for police priorities and understandings of problems may differ from the ideas of a civilian analyst or researcher.

2. Build algorithms that are tailored to community challenges.

   In Chicago, an algorithm used to identify people most at risk for future violence relies on only 11 variables based on criminal arrest records. Down from an initial list of 55 variables, this model is agile, and does not include race, gender, or ethnicity in the equation. Panelists also mentioned using type of shooting, place of crime, or seasonal data as significant variables in their algorithms. There is no “one size fits all” algorithm, an important consideration for any department looking to hire a vendor. Questions a policing leader should ask a vendor include how the algorithm would work in their area, and if vendors will run the department data against the proposed algorithms to test reliability.

3. Engage stakeholders and communities early in the process.

   Panelists lamented that they did not garner enough public support before launching predictive policing efforts. As a result, some citizens were confused about the purpose, and vulnerable to media messages that stoked fear. Sharing plans and gaining feedback from police conduct review boards and the public can ease fears and implementation. It is important to be able to detail what the analytics are looking for, such as patterns and repeat offenses, and not that it is guessing when someone might be involved in a crime. Naming predictive policing as a crime reduction tool or an intervention tool, rather than an enforcement tool, sends a clear message around purpose and can garner more public support.