A Book of Accounts
Every remote video monitoring operator works a book of accounts, and every account comes with its own action plan, with its own call list, talk-down language, and rules for when police get called. The operator is expected to apply the right one, correctly and quickly, whether it's the first hour of the shift or the last.
That's a demanding job, and it isn't paid like one. Monitoring operators don't have their own federal job category. The Bureau of Labor Statistics folds them in with security guards, whose median wage was $18.29 an hour in May 2025. A 911 dispatcher, the job most operators could move to with the same skills, earned a median of $24.39 in 2024. That gap alone explains a lot of the empty chairs.
Weeks to Train, Days to Lose
Proficiency takes time. Monitoring center leaders told SDM in 2024 that their onboarding ranges from eight weeks to a program that runs 18 months, because learning a full book of accounts and the judgment calls behind them takes time on the console.
Centers plan around the churn. One executive told SDM his company hires 20 to 30 percent more staff than it needs just to cover attrition. Covering a single seat around the clock takes a little over four full-time people before anyone takes vacation, so every departure pulls on the schedule for weeks while a replacement comes up to speed.
When Events Stack Up
Monitoring is sequential work. An operator clears one event, then the next. That holds up on a quiet night. When several events land in the same few minutes, the queue sets the order. The first event gets a verified look and a talk-down. The last one gets reviewed after the subject has already left.
Most of what fills that queue isn't real. The Department of Justice's guide to false burglar alarms puts false alarms at 94 to 98 percent of the alarm calls police respond to. Every one still takes operator time to open, review, and close. And the human side has limits that decades of research keep confirming. In a 2015 study published in Applied Ergonomics, experienced CCTV operators caught about half of the target behaviors on their screens.
SOP adherence drifts under that load too. A veteran follows the escalation path from memory, while a newer operator scrolls the account notes as the event plays out. Both are doing the job as trained, and the site gets a different response depending on who picked it up.
Police Want Proof
The pressure doesn't stop at the monitoring center. A growing list of cities require verification before police respond to an alarm. Seattle joined them in October 2024 after logging about 13,000 alarm calls in 2023, fewer than 4 percent of them confirmed crimes. When Milwaukee made the same move, alarm responses fell from about 30,000 a year to roughly 800.
TMA's AVS-01 standard gives operators a consistent way to score each alarm from zero to four, so police get the context they need to prioritize a call. That's good for everyone, and it means every event needs a verified look before it goes out. Verification lands on the operator's desk.
Recorded, Never Watched
That brings it back to the cameras. Live monitoring pays well for dealers and integrators. One integrator told SDM this year that video monitoring runs $150 to $200 a month, against $30 to $60 for traditional alarm monitoring. Operator time is the biggest cost in live monitoring, and it's the reason most cameras go unmonitored. The camera goes up, and the footage waits in the VMS until someone needs it after the fact.
That choice has a measurable cost. A 2019 review of 80 CCTV studies, led by criminologist Eric Piza, found actively monitored cameras were linked to about a 15 percent reduction in crime. Passive, record-only systems showed no significant effect.
Detection scales with hardware. Live response still scales with operators, and operators are the hardest thing in the program to add.
Why Agentic AI Fits
Analytics solved the first problem years ago. AI detection can tell a person from a swaying branch and a vehicle from a shadow, and it sends far fewer nuisance events than motion detection ever did. But detection hands off to the same operator, who still has to verify, decide, talk down, call, and write it up.
Agentic AI is built for that handoff. It doesn't stop at raising an alert. It takes the goal of an SOP, carries out the steps in order, and adjusts to what happens on the scene. That lines up with every problem above. Events get worked in parallel, so a busy stretch doesn't become a queue. The SOP runs the same way on the first shift and the last, whoever is on the console. Each alarm reaches police already verified, with the scene described. And because the repeatable work no longer depends on a person watching, live coverage can reach every camera on the account.
The operator's job changes with it. Operators take the events that need judgment, with the verification, the talk-down, and the call history already done.
SARA at Work
RAD built SARA Agentic AI for that seat, and it's now standard on every RAD device. SARA takes on the repeatable work of the operator. When AI detection flags an event, SARA verifies it in real time, issues a live talk-down, and executes the account's SOP. It notifies up to three stakeholders at once by call, text, and email, and documents every step in RADSOC. It handles events in parallel, so a busy stretch doesn't turn into a queue, and the SOP runs the same way on every camera and every shift.
Operators work the verified events SARA escalates, with the scene and the actions already taken attached. For dealers and integrators, a camera can move from record-only to monitored without adding a seat.
"Traditional workflows built around manually reviewing alerts, coordinating escalation, and managing every operational event individually simply do not scale efficiently anymore, particularly as surveillance infrastructure continues expanding."
Steve Reinharz, CEO and CTO, AITX and RAD
Start With the Busiest Console
Pick the console with the most accounts on it. Pull 90 days of event logs and measure four numbers: events per hour at the peak, time from alert to first action, the share of events that turned out to be nuisance alarms, and how often an SOP step was skipped or late. Then count how many of the cameras on those accounts are record-only. Those numbers are the baseline for a pilot and the scorecard at 30, 60, and 90 days.
David Marsh
Vice President of Marketing
Robotic Assistance Devices
linkedin.com/in/davidmarsh
To learn how SARA Agentic AI keeps every camera on the same SOP as your program grows, visit radsecurity.com/sara.
