One guard watching 12 camera feeds catches maybe half of what happens. The math doesn't work.

1. The vigilance decrement: what happens after 20 minutes

The term "vigilance decrement" comes from research that began in the 1940s with Norman Mackworth's studies on radar operators during World War II. Mackworth found that operators tasked with watching a monotonous display for infrequent signals showed a sharp decline in detection accuracy after approximately 20 minutes. The finding has been replicated hundreds of times across different domains, and it applies directly to security camera monitoring.

Here is what happens cognitively when a guard sits down to watch a bank of camera feeds:

• Minutes 0 to 15: peak performance. The guard is alert, scanning feeds, and likely to catch any anomaly. Detection rates during this window are at their highest, typically above 80% for obvious security events.
• Minutes 15 to 30: the decline begins. Attention starts to wander. The brain, receiving mostly uneventful visual input, begins to reduce its allocation of cognitive resources to the monitoring task. Detection rates can drop to 50 to 60%.
• Minutes 30 to 60: sustained degradation. The guard may still appear to be watching, but the ability to detect subtle or brief events is significantly compromised. Studies show detection rates for infrequent events can fall below 50% during this window.
• Beyond 60 minutes: severe impairment. Without a break or task rotation, the guard is functionally less effective than a motion-activated recording system. The cognitive cost of maintaining sustained attention on low-event-rate displays becomes overwhelming.

This is not a failure of training, motivation, or discipline. It is a fundamental limitation of human cognition. The brain evolved to attend to dynamic, changing environments, not to stare at static video feeds waiting for something to happen. No amount of coffee, training protocols, or disciplinary threats can override this biological constraint.

2. Research on camera monitoring and sustained attention

Multiple studies have examined camera monitoring performance specifically in security contexts. The findings consistently paint the same picture.

A widely cited study by the Security Industry Authority (SIA) in the UK found that CCTV operators watching multiple screens detected only 45% of staged incidents over a one-hour period. When the number of screens exceeded 16, detection rates dropped further. The study recommended that operators monitor no more than 16 cameras, with mandatory breaks every 20 minutes, a standard that very few real-world security operations actually follow.

Research from the University of Surrey found that after just 22 minutes of CCTV monitoring, operators began to miss events that they would have caught in the first few minutes. The researchers noted that event frequency was a critical factor: when events are rare (which they typically are on security camera feeds), the vigilance decrement is more severe because the brain receives fewer "rewards" for maintaining attention.

A 2013 study published in the journal Applied Ergonomics measured performance across different camera-to-operator ratios. Key findings included:

• 4 cameras per operator:Detection rates averaged 78% over one hour. This is the "optimal" ratio, but it requires one guard for every four cameras, which is prohibitively expensive for most properties.
• 8 cameras per operator: Detection rates dropped to approximately 62%. The operator had to cycle attention across feeds, creating gaps in coverage.
• 12 or more cameras per operator: Detection rates fell below 50%. The operator could physically look at each feed but could not cognitively process all of them with enough depth to reliably catch anomalies.
• 16+ cameras per operator: At this point, the operator was essentially performing random spot-checks rather than continuous monitoring. Detection became heavily luck-dependent.

Most real-world security operations assign far more than 12 cameras per guard. A typical multifamily property with 20 to 40 cameras and one on-site guard is operating well beyond the cognitive threshold where reliable detection is possible. The guard is not the problem. The staffing model is.

3. Flipping the model: AI watches, humans respond

The traditional model asks humans to do what they are worst at (sustained attention on low-event-rate displays) and leaves them idle for what they are best at (judgment, de-escalation, physical response). The better model flips this entirely.

In an AI augmented monitoring workflow, the responsibilities split like this:

• AI handles continuous surveillance. Computer vision processes every frame from every camera simultaneously. It does not experience fatigue, boredom, or cognitive decline. It maintains the same detection performance at hour one and hour ten thousand. Whether the property has 8 cameras or 40, the AI analyzes all of them at once.
• AI classifies and prioritizes events. Not every motion trigger is a security event. AI systems trained on security scenarios distinguish between a resident walking to their car, a stray animal crossing the parking lot, and a person attempting to access a restricted area. This classification reduces false alerts and ensures human attention goes to genuine threats.
• Humans receive curated alerts. Instead of staring at screens, the human responder receives notifications only when the AI identifies a genuine security event. Each alert includes a screenshot from the relevant camera, the location, a timestamp, and a threat classification. The human can verify, escalate, or dismiss in seconds.
• Humans handle response. Once alerted, the human does what humans do best: make judgment calls, contact police, activate two-way audio, dispatch on-site personnel, or determine that the event is benign. This is high-value, intermittent work that plays to human cognitive strengths.

Edge AI devices like Cyrano connect directly to existing DVR/NVR systems via HDMI and process video on-site. This means no cloud upload of footage, lower latency, and compatibility with camera systems already installed on the property. The AI layer sits on top of whatever hardware you already have.

This model does not eliminate the need for human involvement. It eliminates the need for humans to perform the one task they consistently fail at: hour after hour of unbroken visual monitoring. The guard who was sitting in front of 16 monitors, catching perhaps 45% of events, becomes a responder who is notified of 95%+ of events and can act on each one with full context.

4. Staffing approach comparison: costs and detection rates

The financial case for AI augmented monitoring becomes clear when you compare the real numbers across different staffing approaches. The table below uses typical costs for a 20-camera multifamily property requiring 24/7 coverage.

A few key observations from this comparison:

• A full-time guard costs 40 to 75 times more than AI monitoring per month and delivers a lower detection rate. The guard provides physical presence, which has deterrence and response value, but the monitoring function itself is inferior to AI.
• The $200/month AI monitoring price point makes it accessible to properties that could never justify $3,000+ per month for a night guard. A 50-unit apartment complex paying $200/month is spending $4 per unit, a negligible cost that can be absorbed in operating budgets or passed through as a minor amenity fee.
• The hybrid approach (AI plus part-time guard) delivers the best outcomes. AI handles the cognitive task of sustained monitoring while the guard provides the physical deterrent and response capability during high-risk hours. This combination costs roughly what a single night-shift guard would cost, but with dramatically better detection coverage.
• Cyrano's pricing structure of $450 one-time hardware plus $200/month starting in month two means the first month is essentially a hardware investment with the ongoing monitoring cost lower than almost any alternative staffing approach.

The question is not whether to use technology or humans. The question is what each should be doing. Asking humans to perform continuous visual surveillance is asking them to do something their brains are not designed for. Asking AI to detect events and surface them to humans plays to each party's strengths.

5. Implementing an AI augmented monitoring workflow

If you are currently relying on guards to watch camera feeds, or if you have cameras that nobody watches at all, here is how to transition to an AI augmented model:

• Assess your current camera infrastructure. AI monitoring works with existing cameras and DVR/NVR systems. You do not need to replace hardware. Verify that cameras are operational, properly aimed, and recording at sufficient resolution. Edge AI devices connect via HDMI to your existing recorder and begin analyzing feeds immediately.
• Redefine your guard's role.If you have on-site security personnel, shift their job description from "monitor cameras" to "respond to AI alerts and maintain physical presence." This is a better job for the guard (more engaging, more varied, less cognitively draining) and a better use of your budget. Guards who respond to specific alerts are more effective than guards who stare at monitors.
• Establish alert routing and escalation protocols. Define who receives alerts during business hours, after hours, and on weekends. Most AI systems allow multiple recipients and escalation chains, so if the primary contact does not acknowledge an alert within a set time, the next person in the chain is notified.
• Calibrate detection zones and sensitivity. During the first 30 days, work with the AI system to tune detection zones. High-traffic areas like main entrances may need different sensitivity than restricted areas like mechanical rooms. Proper calibration reduces false positives and builds trust in the alert system.
• Consider the hybrid staffing model. For properties where physical security presence matters (Class C multifamily, construction sites, properties with history of on-site crime), pair the AI system with a part-time guard covering high-risk hours. The AI covers the 24/7 detection, and the guard provides the human element during the windows when incidents are most likely to require physical intervention.
• Track metrics from day one. Record every alert, every verified incident, and every response time. After 90 days, you will have clear data showing how the AI augmented model compares to your previous approach. Most properties see a significant increase in detected incidents during the first month, not because crime increased, but because they were previously missing events.

The shift from human-centric monitoring to AI augmented monitoring is not about removing humans from security. It is about placing humans where they add the most value. Watching screens for hours is not that place. Making decisions, responding to threats, and maintaining a visible security presence are. When the AI handles the cognitive load of continuous surveillance, the human element of your security operation becomes more effective, not less.