Somewhere in the last two years, someone asked your department whether it should have a real-time crime center. Maybe a council member saw one on the news, maybe a neighboring sheriff opened one, or maybe a vendor left a brochure with a photograph of a dark room and a wall of screens. The question sounds like a technology question, and most of the material written about it treats it as one, because most of that material is written by companies that sell the screens, but in practice a real-time crime center is mostly a decision about people, records and public trust, with technology underneath, and the agencies that run one well tend to be the ones that worked out the people and records questions before they bought anything.
What follows covers what a real-time crime center (RTCC) is and what it is not, the parts it runs on and which of those you probably own already, who staffs it, how an alert reaches an officer, what it costs to keep running after the first year, and what your council will ask before it votes. It is written for a police chief or sheriff at a small or mid-size agency, and for the city manager and IT director who will be in the room when the decision is made.
Why so many agencies are asking now
Real-time crime centers started in the largest cities, with New York opening the first widely known one in 2005, and for about a decade the model stayed with departments big enough to staff a room around the clock. A criminologist writing in The Conversation in 2024, drawing on the Electronic Frontier Foundation's Atlas of Surveillance, counted about 150 agencies with one and put the growth at 148% over four years, while noting that fewer than 1% of US police agencies had one at the time. So the growth is real but the base is still small, which is why so many agencies are looking at the idea for the first time rather than replacing something they already have.
Most of that growth comes from two pressures, and the first is staffing: departments of every size have struggled to hire and keep officers since 2020, and a center that lets fewer people cover more ground is an obvious answer to a chief who cannot fill patrol shifts. The second is that cities already own far more cameras than anyone watches. Traffic cameras, cameras on city buildings and parks, transit and school cameras were bought one project at a time, and almost none of them has a person looking at it when something happens in front of it. A real-time crime center promises to turn that footage into something an officer can use while an incident is still unfolding.
Federal pandemic recovery money paid for a lot of the recent wave. Spokane County, Washington, built its center for about $4 million, nearly all of it American Rescue Plan Act money, and staffs it with three analysts, according to a 2024 report in Police1. That funding source has to be spent by the end of 2026, which matters for anyone planning a center today, and the cost section below comes back to it.
What a real-time crime center is, and what it is not
A real-time crime center is a function inside a police department or sheriff's office that uses live information in two directions. It supports officers already responding to a call, by putting better information in front of them before they arrive: what the person looks like, which way they went, what they are driving, whether there is a weapon. It can also start a response that would otherwise never happen, because when a camera or sensor picks up something nobody has phoned in, such as a break-in at a closed business or a fight outside a bar at 2 a.m., the alert can be checked and turned into a call for service with an officer sent to intervene. After an incident, the same people and systems help investigators reconstruct what happened.
The word "center" suggests a room, and in large cities it usually is one. But the function scales down further than the name implies, and it helps to think of three sizes.
| Size |
What it looks like |
Who does the work |
| A capability without a room |
Software watches cameras and sends alerts to people who have other jobs |
A dispatcher, a patrol sergeant or a detective, as alerts arrive |
| A staffed room |
A dedicated space with analysts on some or all shifts |
Two to five analysts, often civilians, plus a supervisor |
| A multi-shift operation |
Around-the-clock coverage, several analysts per shift, investigative follow-up |
A team of ten or more, with its own management |
A small agency usually starts in the first row and often stays there, and that is a reasonable place to be rather than a lesser version of the idea. An agency of forty officers with no room and no dedicated analyst can still have cameras that are watched by software, alerts that reach whoever is on duty, and footage that is easy to find afterward, and for many agencies that is the whole of what they need. Some small agencies get the second or third row without building it, by joining a center that a county or a larger neighboring city already runs, and our guide to real-time crime centers for small agencies covers how to choose between running your own and sharing one, while our sample plan for a town of 20,000 shows what the first row could look like in practice.
It is also worth being clear about what a real-time crime center is not, because cities do end up paying for the wrong thing. It is not a 911 center, although the two work closely together. The 911 center answers calls and is the part of the agency with the authority to send units, and a real-time crime center feeds it: sometimes with information for a call already under way, and sometimes with an incident of its own, when a camera alert shows something happening that no one has reported and the center asks dispatch to send someone. Nor is it an emergency operations center, the room that activates during floods, storms and large events to coordinate every city department. The three rooms sometimes share walls, screens and even staff, and cities regularly fund one while believing they are buying another. Our comparison of RTCCs, EOCs and situational awareness centers walks through how to tell which one is actually being proposed.
What a real-time crime center runs on
Every real-time crime center, whatever its size, draws on the same kinds of sources. The useful exercise for a department starting out is to list which of these it already has, because most agencies own more than they think.
Cameras are the first thing to count, starting with the ones the city owns outright on buildings, parks and streets. Then there are traffic cameras, which in most cities belong to public works or the transportation department rather than to the police, so connecting them means an agreement between departments before it means any technology, and it also means the same cameras and platform can serve traffic safety, public works and code enforcement as well as the police, which is often what persuades a council. There are cameras at schools and on transit. And there are private cameras, at businesses, apartment complexes and places of worship, whose owners may agree to share them. A private camera can join the same network as the city's own when its stream can be reached over the standard RTSP protocol, either because the camera publishes a stream the system can pull or because the owner's recorder pushes one out. Programs that bring private cameras in are a governance project as much as a technical one, and our article on community camera registries explains what a department takes on when it accepts them.
Beyond cameras, a center typically draws on license plate readers, gunshot detection sensors where a city has them, records from the computer-aided dispatch (CAD) system about what calls are open and where units are, and the records management system (RMS) for history on people, vehicles and places.
Most cities already record their cameras in a video management system (VMS), the software that stores the footage and lets people play it back, and often they have more than one. Adding analytics does not have to mean replacing it, because a platform that fetches live streams directly from cameras, or from any recorder that republishes them over the standard RTSP or ONVIF protocols, can add the part the VMS does not do, which is noticing when something in the picture matters. Some analytics platforms also record and play back footage themselves and can serve as the VMS, which is worth knowing when an old recorder is due for replacement anyway. Our engineering guide on analytics that work with any camera goes further into how that connection works.
That noticing is where the arithmetic of a modern center has changed. A person can watch a handful of screens with real attention for a limited time. Software can watch every stream continuously for the conditions it has been told to look for, such as a person in a closed park after midnight, a vehicle stopped where vehicles should not stop, or a crowd forming faster than usual outside a venue, and raise an alert only when one of them occurs. The analyst's job shifts from watching screens to deciding what to do about alerts, which is a much better use of a scarce person. It is not a reason to believe the software will be right every time, and no honest vendor will tell you it is. Ask any vendor to show you how its alerts perform on your own cameras rather than in a demonstration, ideally in a structured pilot that measures how many alerts each rule produces and how many turn out to be real, which our guide to running a video analytics proof of concept walks through, and plan for someone to review what the system flags.
Who staffs it
Staffing used to be the largest cost of a real-time crime center, because the job was watching screens, and a seat covered around the clock takes close to five people once leave, training and turnover are counted, so most agencies concluded that a center was something for bigger departments.
Once software watches the cameras, the job in the room changes from watching to deciding, and the number of people it takes changes with it. The system watches every connected camera for the conditions it has been told to look for, raises alerts with a clip and a location, sends each one where the agency has decided it should go and keeps the record. What stays with people is judgment: checking an alert before an officer is sent, since no detection system is right every time, deciding what to do about it, talking to officers during a live incident, following up for investigators and answering for those decisions afterward, which is also what a growing number of city AI policies require.
That means headcount follows the number of alerts that need a decision rather than the number of cameras, and the agency shapes that number through the rules it sets. Where alerts go is configurable to each city's own workflow, by alert type and by hour, whether that is an analyst, a patrol supervisor, dispatch or a messaging or SMS service, and for many small and mid-size agencies the volume is low enough that no dedicated analyst is needed. Whoever receives the alerts takes on that work, though, so the routing should be agreed with them, and with the 911 director if any alerts go to dispatch. Our article on staffing a real-time crime center when software does the watching shows how to work out the number from your own alert volume, how to plan for event days when volume spikes, and who to hire.
How an alert reaches an officer
In most centers the difficult part is not spotting something but getting it to the officer who can act on it, quickly and without adding a step that slows the response down.
This matters most for incidents that start with the camera rather than with a phone call. When the software raises an alert on something nobody has reported, someone has to look at it, decide whether it is real, and if it is, get a call for service created so an officer can go and intervene. In a staffed center that is the analyst's decision. In an agency without a room, the alert can go to the dispatcher or a patrol supervisor, who makes the same judgment. Either way, the rule that an alert is checked by a person before a unit is sent is what stops a camera program from sending officers to false alarms, and it is worth writing down which kinds of alert need that check and who does it.
In most agencies the path runs from the analyst to the dispatcher, who relays it to the unit by radio, or from the analyst directly to officers through the mobile data terminals and phones they already carry. Each extra person in the chain adds time and a chance of the detail being lost, so a center built around a new screen that nobody outside the room looks at can produce good information that still reaches the officer too late.
It works better when alerts go into the systems people already use rather than onto a new screen. Modern analytics platforms can send each detection out as a webhook, a message another system can receive the moment the alert is raised, so an alert can land in a messaging tool, a case system or a dispatch workflow without anyone retyping it. Some platforms can also run automated steps on a detection before it goes anywhere, gathering related records or applying conditions a simple camera rule cannot express, so that what reaches a person is more complete.
Integration with CAD needs particular care, because the CAD vendor controls its own interface. Many analytics platforms can send each detection to a CAD system by webhook out of the box, and some can connect to any system with a REST API by configuring its endpoints rather than waiting for custom development, but the CAD side still has to accept the message, and some CAD vendors quote for that separately and schedule it on their own timeline. Ask any analytics vendor what their side of the connection involves, and ask your CAD vendor what it will take on theirs. Our guide to getting camera alerts to 911 dispatch covers the options in more depth, including what to do when the dispatch center is itself short-staffed.
What it costs to keep running
Most real-time crime centers are opened with one-time money and then have to run on recurring money, and programs that close usually do so because nobody planned for the second part. A grant or a capital appropriation builds the room, buys the hardware and pays for the first term of software. Two or three years later the software renews, the servers need replacing, the analysts are on the payroll, and the general fund was never asked to carry any of it.
The recurring lines are worth going through one by one, because several of them surprise people the first time they see them. Software is usually priced per camera, per stream or per type of analytic, and the unit you are charged on decides whether doubling your camera count doubles your bill. Computing hardware for live analytics has a replacement cycle of a few years. Storage grows with the number of alerts and clips the system keeps, not just with the number of cameras. Integration with CAD and other systems is often another vendor's invoice, and staffing, which the previous section showed depends on how many alerts need a person's decision, can range from a share of existing dispatchers' time to a full team. There is also a cost that lands outside the police budget entirely, which is the records work: every alert and clip can become the subject of a public records request, and the review and redaction time falls on whoever handles those requests.
Funding has also become harder to plan around, because federal public safety grant priorities shifted in 2025 and pandemic recovery money, which paid for many recent centers, has to be spent by December 31, 2026. A center planned today should be planned on the assumption that the general fund will carry its recurring costs from year three onward, and anyone offering you a grant should be asked, in writing, which of those lines it will still pay for at renewal. Our guide to budgeting a real-time crime center past the grant period builds that five-year model line by line.
What your council will ask
In a growing number of cities, a real-time crime center cannot be bought without a public process. According to the ACLU, 26 jurisdictions covering nearly 18 million people have adopted community control over police surveillance ordinances, which typically require council approval, a published use policy and regular reporting before a department can acquire this kind of technology. Even where no ordinance exists, councils have become much more likely to ask detailed questions, and several have paused or cancelled camera and license plate reader programs over the answers.
The questions are predictable, which means they can be answered before the meeting rather than during it. Who can see the footage, and is every viewing recorded? How long is footage kept, and what happens to it after that? Is it shared with other agencies, and on what terms? Can the vendor use it to train its own products? Is facial recognition switched on, and if so, under what rules? Where does the data physically live, and what happens to it if the city stops paying? And how will the department show, a year from now, that the center did what it promised?
A department that walks into the meeting with written answers, and with a system that can prove them, is in a much stronger position than one that promises to work them out later.
What your IT director will ask
The IT director will be the one who has to run the system, secure it and pass it through a security review, and their questions are different from the council's. The first is where the processing happens and what crosses the network. When the analytics run on servers on the city's own network, close to the cameras, the full video streams stay local and what leaves that network are the alerts and clips sent to the systems the agency chooses, which keeps bandwidth down and answers the council's question about where footage goes. Sizing depends on how many cameras are analyzed, how many frames per second on each and how many detection models run on each stream, and as a reference point VIDIZMO's live analytics handle from about 32 to more than 100 cameras on one RTX 5090 graphics card, depending on those factors.
The second set of questions is about security, starting with the fact that criminal justice information falls under the FBI's CJIS Security Policy, and the agency holds that compliance obligation, so ask how a vendor's controls support it and where each part of the system runs; our explainer on what CJIS actually requires when AI touches criminal justice data sets out the questions. Ask how outgoing alerts are authenticated when they are sent to other systems, and how streams from private cameras enter the city's network, since a business owner's recorder pushing video into a city system is a new connection that needs the same scrutiny as any other. Our article on on-premises versus cloud processing for real-time video covers the deployment choices in more depth.
What to ask any vendor
Every vendor in this market will demonstrate detection, but fewer can show you what happens after a detection, and that is where a program either holds up under a records request, a court challenge or a council review, or does not. These are the questions worth putting to every vendor on your list, in writing:
| Question |
Why it matters |
| Is every access to footage logged, with the person, the time and a stated reason? |
This is what you will be asked to produce when a viewing is challenged |
| Does automated activity, such as an AI process reading a file, appear in the same trail as people? |
Otherwise part of the record is missing |
| Does the custody record run from the moment footage is ingested to the moment it is deleted? |
A record that covers only the middle is hard to defend |
| Can a retention schedule run automatically, and does a legal hold reliably override it? |
Deleting footage under a hold is a serious problem |
| Can you export everything, footage and records, without raising a support ticket? |
You will need to, for requests and for leaving |
| Where does the system physically run, and who can reach the data? |
Councils ask, and some ordinances require an answer |
| How do alerts reach the systems we already use? |
Information on a screen nobody outside the room looks at arrives too late |
None of these questions is about how well a system spots something, yet together they decide how much staff time the center consumes in years two through five, and how the department fares the first time something goes wrong.
The question about where the system runs deserves a little more explanation, because a council that asks whether any footage leaves the city is asking about deployment. Systems that run on infrastructure the city or its county controls, including fully disconnected from the internet, answer that question differently from systems that send footage to a vendor's cloud. Our article on on-premises versus cloud processing for real-time video explains the tradeoffs, including why analysis close to the cameras also helps with speed.
How VIDIZMO approaches it
VIDIZMO builds the layer underneath a real-time crime center rather than the room itself, and it comes as a platform with separately licensed parts, which is worth explaining once since the other articles in this series refer to them. VIDIZMO Nexus is the platform every deployment is built on: it records and plays back live footage, so it can serve as the agency's video management system, and it keeps footage with chain of custody, retention schedules and legal hold. AI Live Insight runs on Nexus and watches the cameras, fetching streams directly from cameras or any RTSP or ONVIF endpoint, including private cameras whose owners share them, and raises alerts on the conditions the department defines. AI Intelligence Hub adds automated follow-up, running a workflow on an alert before a person sees it and connecting to other systems through their APIs, and Redactor handles redaction before footage is released.
In practice, the operator dashboard brings forward the cameras that have active alerts, so nobody has to cycle through every feed to find the one that matters, and alerts reach people by email and by webhook into the systems the department already uses, such as a dispatch platform, a messaging tool or an SMS service. It runs on infrastructure the city or its regional host controls, including fully disconnected from the internet, and it does not lead with facial recognition or include predictive scoring. Several agencies can share one deployment while each keeps its own footage and camera access, which suits a county running a center for the towns around it, and for a smaller agency without a room, the same layer can be the real-time crime center.
If you want to see how that works on cameras like yours, learn how AI adds intelligence to your existing cameras.