The Netherlands' Crime Anticipation System — A Decade of Hotspot Predictions With No Proven Crime Reduction

The Crime Anticipation System (CAS) was developed for the Amsterdam police starting in 2013 and officially launched there in 2015, generating predicted crime "hotspots" across 125-by-125-metre grid cells in four-hour windows, covering burglary, car and bicycle theft, and street nuisance. In May 2017 the national police announced a rollout across more than half of the Netherlands' 167 police departments.

An internal Police Academy evaluation of the 2015 pilot found CAS correctly predicted about 13.5% of burglary hotspots — described at the time as five times better than random patrols — but also concluded there was "no evidence" that predictive policing measurably reduced crime, and that departments using CAS did not see significantly lower crime rates than those that did not. A 2022 review by the Netherlands Court of Audit found CAS scored "mid to high risk" on every criterion it tested, including insufficient control over biased outputs and the absence of any privacy impact analysis.

A final internal assessment (the September 2025 Kraai Report) found accuracy had fallen to roughly 1 correctly predicted incident in 50 in Amsterdam — about 11.5% — a sharp decline from the figures cited at launch. Police discontinued CAS in late 2025, and the force's Algorithm Board formally halted it in January 2026, with police stating they could not prove the system had reduced burglaries.

CAS illustrates a nationally scaled predictive-policing system that ran for roughly a decade in a majority of Dutch police regions before its own operators and independent government auditors concluded its crime-reduction claims were unsupported.

Read the full analysis: https://nltimes.nl/2026/02/27/dutch-police-discontinue-controversial-crime-predicting-algorithm-cas