Staffing a Real-Time Crime Center When Software Does the Watching

Most chiefs who hear a pitch for a real-time crime center ask the same question before any other, which is whether they will have to hire a room full of analysts to watch the screens, and for most of the history of these centers the honest answer was yes. That answer rested on an assumption that no longer holds, because it treated watching cameras as the job, and software can now watch every camera at once, all the time, which no team of people can do. It will still be wrong sometimes, which is why a person checks an alert before anyone is sent, but the watching itself no longer needs a room full of people. This article explains how staffing changes once the system does the watching, what still has to be done by a person and why, and how to work out how many people your center actually needs. If you are still deciding whether you need a center at all, start with our guide to what a real-time crime center is.

Why the old model needed so many people

The traditional real-time crime center was built around a video wall and a row of people watching it, and every staffing number that came out of that model followed from how little a person can watch well. Someone looking at a wall of feeds can give real attention to a handful of screens for a limited stretch, so covering more cameras meant more people, and covering more hours meant multiplying those people by shifts. A seat that has to be staffed around the clock needs 168 hours of coverage a week, which is 4.2 full-time people at a forty-hour schedule before leave, training and turnover, and close to five once those are counted.

Federal guidance written for that model reflects how many people it took to run, and an older planning guide from the Bureau of Justice Assistance put a round-the-clock center at about twelve full-time staff, and the largest centers in the country are staffed on that pattern, with New York's run by full-time investigators on three shifts, according to Police1. Most agencies looked at those numbers and concluded, reasonably, that a real-time crime center was something for other departments.

What the software takes off people's hands

Camera analytics changes the arithmetic because it takes over the work that consumed most of those seats, and it is worth being specific about which work that is, since the rest of the staffing plan depends on it.

The software watches every camera it is connected to, continuously, for the conditions it has been told to look for, such as a person in a closed park after midnight, a vehicle stopped on a highway shoulder or someone lingering by a closed storefront. When one of those conditions occurs it raises an alert with a snapshot or clip, the camera, the time and a severity, so nobody has to be looking at the right screen at the right moment. It can also decide where each alert goes, sending it into the systems people already use, such as a dispatch workflow, a supervisor's messages or a case file, depending on the type of alert and the time of day. On some platforms an automated step can run on each alert before a person sees it, pulling related records or applying conditions a simple camera rule cannot express, so that what reaches a person is more complete. And the system records all of it, keeping the footage, the alert and who handled it, which removes a large amount of manual logging.

None of that requires anyone to sit in front of a video wall, and it means the question of how many people it takes to watch four hundred cameras no longer has to be answered.

What still needs a person, and why

What remains is judgment and accountability, and it is worth stating plainly why each part stays with a person rather than treating it as a temporary limitation that better software will remove.

The first part that stays with a person is checking an alert before an officer is sent, because no detection system is right every time, and sending officers to a situation that turns out to be nothing carries real risk for the public, for the officers and for the agency, so a person should confirm any alert that will lead to a response. Cities have started writing this into policy, and Seattle's AI policy, updated in 2025, bars the use of AI for consequential decisions and requires that AI output be reviewed by a person, according to KNKX.

The second is deciding what to do once an alert is confirmed, which might mean asking dispatch to create a call and send a unit, watching and waiting, passing it to a detective, or deciding it needs nothing at all. The third is talking to officers during a live incident, which takes someone who knows what an officer on the way needs to hear and when to stay off the radio. The fourth is investigative follow-up after the fact, such as building a timeline or pulling footage for a case. The last is answering for all of those decisions later, in court, in response to a records request or in front of a council, which only a person can do.

Put together, these describe a different job from the one the old model staffed. The person in the room is no longer a watcher but a decision-maker who is interrupted only when the system has something worth deciding about.

Staffing to decisions rather than cameras

Once watching is handled by software, the number that drives headcount is not how many cameras you have but how many decisions per shift the system hands to people, and that number is one the agency controls through the rules it sets.

Every rule produces some volume of alerts, and every alert that needs a person's judgment takes some minutes of that person's time, so the planning exercise is to estimate alerts per shift by rule and multiply by the time each one takes to check and act on. A rule watching a closed park overnight might produce a few alerts a week, while a rule on a busy intersection might produce dozens a day, and a rule that produces more alerts than anyone can read properly is a rule that needs to be narrowed or dropped rather than a reason to hire. The National Institute of Justice noted in a 2025 technology brief that there is little research on optimal staffing and recommended starting from an analysis of calls for service by hour and day, and alert volume by hour is the same kind of analysis applied to the camera side.

A simple worksheet is enough to start with, and it should be revisited every time cameras, rules or staffing change.

Adding up the minutes by hour shows where the decisions cluster, and that is what tells you whether you need a dedicated person at all.

Event days and bad nights are the exception to the idea that the agency controls its alert volume, because a stadium crowd, a festival or a storm can push the number of alerts well past what the usual rules and people were sized for. Plan for them separately: a lighter rule set for event days that holds back low-priority alerts, extra staff or borrowed analysts scheduled in advance, and a written agreement on which alerts still go straight to dispatch while the room is overloaded.

Three staffing shapes that follow

Most agencies land in one of three shapes once they staff to decisions, and the first is where most small and mid-size agencies can stay.

Where alerts go is configurable to each city's own workflow, by alert type and by hour, so the same system can send loitering alerts to a patrol supervisor's phone, fire and smoke to dispatch and everything else to a morning review queue. In the first shape that work lands on people who already have jobs, so whoever receives the alerts takes it on, and the routing should be agreed with them in writing, and with the 911 director if any alerts go to dispatch, since many emergency communications centers are short of staff themselves, as our article on getting camera alerts to 911 dispatch explains. Agencies that want the third shape without building it can join a center someone else runs, which our guide to real-time crime centers for small agencies explains.

Who the decision-makers should be

Because the job is now judgment rather than watching, the question of who fills it shifts toward people who know how officers work and how to weigh an ambiguous situation quickly.

The NIJ brief sets out the tradeoffs between sworn, civilian and mixed staffing and concludes that a mix is likely to work best. Sworn staff bring field experience and know what an officer on the way needs to hear, but they cost more and rotate out. Civilian analysts, often recruited from dispatch or crime analysis, bring fluency with the dispatch system and radio, cost less and tend to stay longer, and the brief cites Elk Grove, California, which staffs its center entirely with civilians and gives lower turnover as one reason. For a small room, a mostly civilian team with at least one person who has worked the street is a sensible starting point, and job classification is worth settling with HR and the unions before the position is posted.

Whatever mix an agency chooses, two things about keeping those people apply to every room. APCO International's Project RETAINS found that supportive supervision, support from co-workers and feeling recognized are the strongest predictors of whether emergency communications staff stay, and the same labor pool fills real-time crime centers. And the people making these decisions see violent events, often live and then again during review, which the Office for Victims of Crime describes as a source of vicarious trauma for anyone repeatedly exposed to violence, so access to peer support and a debrief after serious incidents belong in the plan from the start.

How VIDIZMO approaches it

AI Live Insight is built to do the watching so that people only handle the decisions, and several of its features exist to keep the number of those decisions in line with the people on duty. Rules are set per camera, with zones for loitering, intrusion, crowd count or direction of travel and a severity and confidence threshold for each detection class, so a closed park at night can be tuned differently from a busy transit stop and the alert volume stays within what the agency can handle. Instead of an operator cycling through hundreds of camera feeds, the operator dashboard brings forward the cameras that have active alerts, alongside live counts of open alerts by severity, so the person on duty looks only at the few cameras where something is happening. Alerts also reach people who are not watching the dashboard, through email notifications set per camera that carry a snapshot of the moment, and through webhooks to external systems, so a detection can land in a dispatch platform, a messaging tool or an SMS service that texts the sergeant on shift. With AI Intelligence Hub, an automated workflow can also run on a detection first, gathering related records before a person looks at it; our guide to real-time crime centers explains how the VIDIZMO products fit together.

Every alert then moves from New to Acknowledged to Resolved, with a note required when it is resolved or marked a false positive and each action recorded against the person who took it, so the agency has a record of who decided what, and a dispatcher or analyst handing over at shift change leaves a clear picture of what is still open.

To see how that works on cameras like yours, see how AI Live Insight works on the cameras you already have.