In 1990, New York City recorded 2,245 murders. Crack cocaine had gutted entire neighborhoods. The subway was widely understood to be unsafe after dark. The prevailing theory in criminology held that crime was primarily a function of poverty and social conditions — structural forces beyond the reach of anything a police department could do.

What that theory missed was that the NYPD also had a data problem.

Crime information in New York traveled upward through the department on paper — patrol officers filed reports, sergeants compiled summaries, precinct commanders reviewed aggregated numbers that arrived days or weeks after the underlying events. By the time anything reached a level of authority capable of directing resources, the pattern had already moved. Precincts were deploying officers to where crime had been, not where it was going. The data existed. It was just too slow to be useful.

The crime was already there. The data was already there. The gap was in the distance between the signal and the person authorized to act on it.

What CompStat actually was

When William Bratton became Police Commissioner in January 1994, he asked a question that sounds obvious in retrospect: why is the department making resource decisions based on information that's three weeks old?

The answer was that no one had built a system to move the information faster. Bratton and his deputy Jack Maple did.

CompStat — short for computer statistics — was, at its core, a decision environment. Twice a week, every precinct commander in the city sat in a room in front of a large digital map showing crime patterns updated in real time. Where were robberies clustering? Which blocks had seen spikes? How were patterns moving relative to last week, last month, last year?

The commanders didn't receive a report. They received a question: what are you doing about it?

A commander who said they'd been concentrating patrol on the south end of the precinct could be shown, right there on the map, that crime had migrated north. They couldn't defer to outdated data. They couldn't hide behind the lag between events and information. The map was current. The accountability was immediate. Adjustments could happen in days rather than quarters.

The mechanism

What CompStat created wasn't just faster information — it created a format that forced decision-makers to be present with that information on a regular, structured schedule. The twice-weekly meeting was not optional. Commanders could not send a deputy. The senior leadership of the department was in the room. The implicit message was that current information about your precinct was your responsibility, and that being caught without an answer to an obvious question about a pattern in your own data was a career event.

This sounds harsh, and it was. But the effect was that information stopped aging in inboxes. Commanders had strong incentives to know their numbers before the meeting — which meant they were engaging with real-time crime data on a continuous basis rather than receiving a quarterly briefing. The decision environment made ignoring the data functionally impossible.

Resources followed patterns instead of chasing them. Officers were deployed to where analysis showed crime was moving, not where it had already been. Response time to emerging patterns dropped from weeks to days.

The numbers

Between 1994 and 2000, murders in New York City fell by more than 70 percent. Robbery dropped 67 percent. Felony assault fell by half. These were not marginal improvements — they were transformations of what life in the city felt like for millions of people.

The NYPD had not hired more officers. It had not received additional funding. It had not solved poverty or rebuilt neighborhoods or addressed the structural conditions that criminologists had identified as the root causes of crime. What it had done was change the speed and structure of information flow — and tie that information directly to the people with the authority and incentive to act on it.

CompStat was subsequently adopted by police departments in Baltimore, Los Angeles, Boston, and eventually hundreds of cities worldwide. In every case, the model was the same: surface the signal, eliminate the delay between data and decision, and create an organizational structure that makes acting on the information both possible and expected.

The data was already there in 1990. The murders were already being recorded. The patterns existed in the paper reports stacked in precinct offices across the five boroughs. The difference wasn't more information. It was a decision environment built around getting the right information to the right manager at the right time — while there was still something that could be done about it.