Abstract
Police departments across the US are rapidly adopting AI-assisted report-writing, yet its cost-benefit case remaines largely unexamined, and the field-trial evidence can be read as either favorable or unfavorable. This paper presents a probabilistic cost-benefit analysis of one such deployment at the Manchester Police Department, varying how the tool's cost is attributed and how much of the measured productivity gain is credited to the AI rather than to organizational learning. Under per-officer subscription pricing, the tool pays for itself. For a medium department, the average net present value is $2.3 million under a favorable, frontier-based reading of the productivity evidence and $0.8 million under a conservative reading anchored to the null result of the department's randomized controlled trial, with a positive return likely in both. Attributing a bundled vendor contract to the program instead yields a negative return. Cost attribution, therefore, determines the sign of the result, while the extent to which the gain is credited to the tool governs how confident a department can be in adopting this technology.