ShotSpotter in Chicago

ShotSpotter uses networks of acoustic sensors and software to detect sounds that may be gunshots locate them, and alert police. Chicago used the system for several years. A three-year deal signed in 2018 was worth $33 million, and the city was spending about $10 million a year on the technology.¹

The company says its system identifies gunshots with 97% accuracy.¹ Chicago’s experience shows why that figure, even if accepted, does not settle the question a city actually needs to answer.

What the Inspector General found

In August 2021, Chicago’s Office of Inspector General (OIG) published an analysis of how police responded to ShotSpotter alerts. It examined more than 50,000 alerts over a 17-month period.³

Only 9.1% of the 50,176 alerts were linked to evidence of a gun-related criminal offence.² About 2% of alerts could be directly tied to a stop and search recorded by an officer.³ The OIG concluded that “CPD responses to ShotSpotter alerts rarely produce evidence of a gun-related crime.”¹

The report also raised a less obvious effect. Officers sometimes cited the general frequency of alerts in an area, rather than a specific alert, as part of their reason for stopping someone. As the deputy inspector general, Deborah Witzburg, put it, “a CPD member’s impression of the frequency of alerts in an area changes their policing behavior.”²

The decision to stop

In February 2024, Chicago’s mayor, Brandon Johnson, who had previously called the technology “a failure”, announced that the city would not renew its contract. Chicago stopped using the system on 22 September 2024. By then, the company had renamed itself SoundThinking.¹

Why it is contested

This is not a simple failure. Supporters argue that the system detects shootings that are never reported by phone, gets officers to the scene faster, and may help victims receive care sooner. Critics argue that it sends armed officers into communities expecting gunfire, rarely produces evidence, and shapes where and how police act. Both can be partly true.

What the Chicago data shows clearly is a gap between two questions. The supplier’s question is whether the sensors correctly detect gunfire. The city’s question is whether deploying them reduces violence and builds public safety at an acceptable cost, financial and social. A system can score well on the first and still fail the second.

Lessons for leaders

1. Measure the outcome you bought, not the metric you were sold. Technical accuracy is an input. The value lies in what happens next: arrests, evidence, lives saved, trust built or lost.
2. Look for behavioural side-effects. Systems change how people behave. Here, the presence of alerts appears to have changed how officers judged whole areas, not just individual incidents.
3. Build independent evaluation into the contract. Chicago’s clearest evidence came from its own Inspector General, years into the deployment. Agree from the start how impact will be measured, by whom, and when the decision to renew will be taken.

Questions to ask in your organisation

1. For our AI systems, do we track the outcomes we care about, or only the accuracy figures the supplier reports?
2. How might this system change the behaviour of the people who use it, and of the people it affects?
3. Who independently evaluates whether our AI investments deliver, and do their findings reach the people who decide on renewal?