Boosting police productivity with AI? An assessment of productivity and efficiency gains in report writing
Erik Alda, Felipe Salazar-TobarPurpose
Writing police reports takes a considerable amount of police officers' time. We investigate the impact of artificial intelligence (AI) report-writing technology on police productivity and efficiency.
Design/methodology/approach
We use data from a randomized field experiment from the Manchester Police Department. We use a two-stage approach to investigate changes in productivity. We first calculate the Global Malmquist Productivity Index (GMPI), which allows us to decompose changes in total factor productivity into changes in technical efficiency and changes in the best practice frontier. In the second stage, we use each GMPI component as the outcome in a difference-in-differences model.
Findings
We find that the difference-in-differences coefficient for the treatment condition is positive but statistically insignificant, consistent with Adams et al. (2026). However, the GMPI shows that officers in both the treatment and control groups experienced large increases in productivity (TFPCH 1.476 vs 1.424), which were almost entirely driven by changes in technical efficiency (catching up to the best-practice frontier; TECH 1.636 vs 1.426), not changes in the best-practice frontier itself (technological change; BPC 1.065 vs 1.064). Despite the null results, back-of-the-envelope calculations suggest that efficiency gains translate into roughly 1.68 h saved per officer per week and about $642,000 in annual labor value for the department.
Research limitations/implications
Our data come from a medium-sized police department. Different spillover dynamics may occur in larger and more segmented organizations. Additionally, while the GMPI lets us examine changes in labor productivity, it does not provide information on changes in police report quality, impacts on criminal justice outcomes or impacts on officer well-being. Replicating this study in a variety of settings will help assess the generalizability of our results.
Practical implications
Our results suggest multiple potential mechanisms through which AI can increase organizational productivity, consistent with our finding that productivity increases were driven by improvements in technical efficiency, not changes in the productivity frontier. We discuss these mechanisms and show how the GMPI can be used by managers to investigate which mechanisms are most relevant in their context.
Originality/value
This is one of the first studies to use a frontier approach to investigate the impact of AI in an organizational setting.