Skip to content
Camzify
Beyond the checklist

AI risk detection on patrol

A checklist can only ask what you thought to ask. During an automated patrol round, Camzify assesses each camera for safety and security risks in its own right and raises a critical alert for what it finds — whether or not any checklist item covered it.

Book a demo Automated scheduling
A gate camera: guard at the gatehouse, vehicles passing the barrierCAM 01 · Main gate
A loading dock camera: two trucks at the bays, pallets on the apronCAM 04 · Loading dock
A parking lot camera: rows of parked cars with a person walking between themCAM 02 · Parking lot
A server room camera: rows of racks with a technician walking the aisleCAM 07 · Server room

The gap a checklist leaves

Checklists are written from experience: the gate that gets left open, the dock door that gets propped, the corridor that collects boxes. They are good at the things that have gone wrong before, which is exactly why they are worth having.

What they cannot do is anticipate. The condition that causes the next incident is usually not the one somebody thought to write down — and a round that only answers the questions on the list will walk past everything else without recording that it saw anything at all.

That gap is what risk detection covers. The same stop that answers “is the dock door secured” is also assessed for what is actually in frame, so the pallets stacked against the fire exit behind it are flagged too.

It is assessed at every stop, not only when something is wrong

Every camera on an automated round gets two entries in the report: possible safety risks and possible security risks. They are filled in whether or not anything is wrong. A gym might read “possible tripping hazards due to equipment left out on the floor”; a pathway, “wet surfaces might cause slipping”; most stops, “none apparent”.

Recording the negatives is what makes the positives worth reading. A system that only speaks up when it has something to say gives you no way to tell the difference between a quiet night and a system that stopped looking — and it leaves no record that a hazard was absent at the time somebody later says it was there.

Each stop also records what the camera actually saw: a plain-language description of the scene, the number of people present, and the objects detected in view. That is the context an assessment is made against, kept alongside it.

Why the timing is the whole point

Most site risks are not instantaneous. A fire door wedged open at the start of a shift is wedged open for hours. Pallets in front of an exit stay there until someone moves them. A bag left in a lobby sits there until it is noticed. Each is a risk for a stretch of time before it is anything worse, and during that stretch it is a five-minute fix.

A scheduled round that looks at every camera every couple of hours lands inside that window. That is the entire claim — not that the system foresees events, but that it reaches a hazardous condition while it is still just a condition, and tells a named person who can deal with it.

What this is not: it is not prediction, and we do not claim it is. The system reports conditions visible on camera now. Anyone selling you software that forecasts incidents is describing something that does not exist, and our position on claims like that is on the record.

How it fits with everything else

The checklist proves the control

Each item is answered and recorded against a snapshot, which is what an auditor or insurer asks for. It covers what you decided to verify.

Patrol checklists →

Risk detection covers the rest

The same stop is assessed for hazards and security risks in frame, raising a critical alert for anything found that the list did not ask about.

AI detection models →

Scene observation adds context

A stop can be judged on a few seconds of live video rather than a single frame, which is what separates somebody passing through from somebody staying.

Automated scheduling →

The alert reaches a person

Critical notifications go to the guard configured for that camera on email, SMS, WhatsApp or push, and expect an acknowledgment.

Guard notifications →

Where it matters most

Sites where a hazardous condition can persist unseen for hours get the most from it: a warehouse after the shift ends, a construction site overnight, a remote or unmanned site where nobody walks past at all. Anywhere the answer to “how long before someone would notice” is measured in hours, a scheduled round that looks properly is worth more than another camera.

FAQ

Frequently asked questions

It is the system flagging a hazardous or insecure condition while it is still a condition, rather than recording the incident that follows. A fire exit blocked by pallets, a security door propped open, an unattended bag in a public area and smoke in a plant room are all risks for a period of time before anything happens, and that period is when they are cheap to deal with. Camzify raises these as critical notifications during automated patrol rounds, alongside the checklist results for the same cameras.

A checklist answers the questions somebody wrote down. Risk detection answers the question nobody thought to write down. Both run on the same round: each camera is checked against its checklist, and the same stop is also assessed for risks in its own right, so a blocked exit is flagged whether or not "exit clear" was ever added as an item. The two are complementary, the checklist is what proves a specific control was verified, and risk detection is what covers the gap between controls.

No, and be wary of any vendor claiming it does. What it does is narrower and more useful: it observes conditions that are present now and would take time to become an incident, and tells someone while there is still time to act. A propped door is not a prediction, it is a fact about the site right now, and the value is that a person hears about it tonight rather than reading about it in an incident report next week.

The categories align with the detection models the platform runs, fire and smoke, abandoned or unattended objects, people in restricted areas, PPE not being worn where it is required, aggression, obstruction and camera tampering among others. See the AI features index for the full set. What a given round flags depends on which features are active on that camera.

In the same notifications queue as detections from continuous monitoring, marked critical, carrying the snapshot from the camera and expecting an acknowledgment. They are not buried in the patrol report, the report records the round, while a risk that needs somebody now goes out as a notification on the channels that guard is configured for: email, SMS, WhatsApp or push.

No. Continuous monitoring watches a camera all the time and is what catches an event as it happens. A patrol round is a scheduled sweep that reaches every camera in the sequence in order and leaves a record that it did. Risk detection on a round is the second of those doing some of the work of the first: a deliberate look at every stop, on schedule, whether or not anything triggered.

Ready to patrol your site 24/7?

Book a 15-minute demo and see a live patrol run on your own cameras.

Book a DemoCalculate Your Savings

This site is being updated

We are rebuilding pages as you read this, so an image, a link or a section may look unfinished for a while. The product itself is unaffected. If something important is broken, tell us at the contact page and we will fix it.