Predictive policing holds potential to help police departments use scarce resources more efficiently, but the closed, proprietary products that dominate the commercial market have a mixed independent track record and carry real constitutional risks, according to a new report from the R Street Institute, published as part six of a seven-part series examining policing strategies in America. The report concludes that predictive policing belongs in the "adopt with caution" category, with transparent approaches showing stronger evidence than opaque commercial tools that have largely failed rigorous testing.
The report identifies two broad families of predictive policing: place-based prediction, which forecasts locations where crimes will occur, and person-based prediction, which forecasts individuals likely to offend or be victimized. A 2024 systematic review of place-based prediction screened 161 studies but found that only six met the strongest evidence standard—randomized, field-tested, and accounting for crime displacement. RAND's randomized evaluation in Shreveport, Louisiana, found no meaningful drop in property crime compared with areas that didn't use the predictive model. Chicago's Strategic Subject List, launched in 2012, assigned more than 280,000 residents a score from 0 to 500 reflecting their predicted likelihood of being a "party to violence," but RAND's evaluation found the list didn't reduce gun violence. A 2023 analysis of roughly 23,600 predictions from Geolitica (formerly PredPol, the most widely used predictive policing software in the country) for Plainfield, New Jersey, found they almost never matched reported crime. The Los Angeles Police Department's inspector general concluded there was insufficient data to show that PredPol reduced crime, and the department dropped it in 2020.
The report finds that the approaches with the strongest evidence—risk terrain modeling on the place side and focused deterrence on the person side—are also the most transparent. According to the analysis, evaluations across several cities, including Chicago, Colorado Springs, Glendale, Kansas City, and Newark, found that directing tailored interventions to risk terrain modeling-identified locations reduced crime. A systematic review of two dozen focused deterrence evaluations found that most programs produced clear reductions in violence, often substantial ones. The report states that "the weakest results come from opaque commercial algorithms and enforcement-only watchlists," while transparent forms succeed "precisely because they identify the environmental conditions driving risk, which lets agencies design targeted interventions rather than simply flood a hot spot with patrols."
The report explains that the central limitation of predictive systems is that they learn from historical police data, which records where police have looked rather than where crime objectively is. The authors warn that data produced during periods of flawed, racially skewed, or unlawful policing becomes "dirty data" that yields flawed predictions, creating a self-reinforcing cycle: an algorithm directs officers to a neighborhood, their presence produces more stops and arrests there, that new activity flows back into the model, and the location stays "hot" regardless of the level of underlying crime. A simulation showed that feeding Oakland's drug-arrest records into a PredPol-style algorithm would have concentrated patrols in predominantly Black neighborhoods, not necessarily because more crime occurred there, but because that's where past enforcement had been focused. The report also raises Fourth Amendment concerns, noting that legal scholars have warned algorithmic prediction can transform reasonable suspicion from a protection against unreasonable stops into a justification for them. The starkest example came from Pasco County, Florida, where the sheriff's office used an algorithm to flag residents—including minors—then dispatched deputies to make repeated, often nighttime, visits until targets moved away or sued. In 2024, the office settled a federal civil-rights suit for $105,000, acknowledging the program violated residents' First, Fourth, and Fourteenth Amendment rights.
The report recommends that agencies deploy predictive policing only with independent validation, transparency, bias auditing, and constitutional guardrails, favoring open, explainable models over proprietary ones. It concludes that agencies should keep a human in the loop, treating a prediction as a tool to inform an officer's judgment, never a replacement for it, and that "taxpayers should not be asked to fund these tools without independent evidence that they work." While standalone predictive policing has largely receded, the underlying impulse has migrated into broader AI-driven police technology, shifting from "predicting crime" to "data fusion" and "real-time intelligence"—but the civil liberties questions remain.