AI-Powered Video Analytics Cuts Hospital Security Gaps
Montana's largest independent healthcare system — and a security operation that couldn't keep up.
The client is the largest independent healthcare system in Montana, serving not only the state but also the regions of Wyoming and the Western Dakotas. Their flagship facility is a substantial 336-bed hospital — a complex, high-traffic environment where security isn't a background function. It's a patient safety obligation.
To protect patients, staff, and visitors, the hospital had invested in a network of over 300 Cisco Meraki MV cameras deployed across the premises. The infrastructure was there. The problem was what happened next: someone had to watch them.
Manual monitoring at that scale is not a security strategy. It's a liability.
300 cameras. Dozens of screens. And human attention that runs out.
The hospital's security team was doing what most security teams do: monitoring camera feeds manually, responding to incidents after they happened, and relying on human vigilance to catch threats in real time. In a busy 336-bed hospital serving three states, that's an impossible ask.
The numbers didn't add up. 300+ cameras. One screen at a time. Human attention that degrades after minutes, not hours. By the time an anomaly was spotted, the window to prevent it had already closed. This wasn't a staffing problem — it was a fundamentally broken operating model for security at scale.
The existing camera network was an asset that was being massively underused. It had the potential to do far more than record — it could detect, alert, and prevent. But only if the right intelligence was put behind it.
The challenge wasn't technical. It was an operating model problem: how do you turn 300 passive cameras into an active, always-on intelligence layer — without rebuilding the physical infrastructure from scratch?
We didn't start with the AI model. We started with the hospital.
OnStak's approach was to unlock the value already sitting in the client's infrastructure — not replace it. The 300 Cisco Meraki cameras were an asset. The Meraki Cloud was already there. AWS GPU infrastructure could power real-time inference at scale. The job was to wire it all together into something the security team could actually use.
The result was a unified Video Analytics (VA) platform: an AI-powered intelligence layer deployed on AWS that integrates every camera, processes every frame, and surfaces only what matters — in real time, before incidents escalate.
No rip-and-replace. Every Cisco Meraki MV camera already in place was retained and activated. The VA platform extended the existing estate — it didn't replace it. Capital already spent became capital that worked.
Six AI models. One platform. Every threat scenario covered.
The VA platform doesn't watch everything and alert on nothing. It watches everything and alerts on what matters — powered by six specialised AI detection models, each tuned for the specific threat scenarios a busy hospital actually faces.
What makes this platform different from standard CCTV.
Most hospital security systems record. This one reasons. OnStak's VA platform ships five capabilities that conventional systems don't have:
Event-Driven Alerts — not human-triggered reviews
The system detects and surfaces anomalies automatically. Security staff receive prioritised alerts with context — not a wall of camera feeds to monitor. Response time collapses from minutes to seconds.
Forensic Search Technology
When an incident occurs, security teams can search across the entire 300-camera archive by event type, time, location, or detection trigger — in seconds. What used to take hours of manual footage review now takes moments.
Predictive Policing — spotting precursors before incidents escalate
The platform identifies behavioural patterns that precede security incidents — not just the incidents themselves. It's the difference between responding to a fall and preventing one.
Linked Analytics — events connected across cameras and time
Events don't happen in isolation. Linked Analytics connects related events across multiple cameras and time windows — giving security teams a complete picture of an incident's development.
Real-Time Event Processing — no batch delays, no sampling
Every frame from every camera is processed continuously and in real time. Not sampled every few seconds. Not queued for batch processing. If something happens, the platform knows immediately.
"The camera network was already there. 300 lenses watching the hospital, recording everything. The question wasn't whether to invest in more infrastructure — it was whether to invest in the intelligence to make what was already there actually work."
OnStak Gen AI Practice — Engagement LeadFrom reactive to proactive. From recording to reasoning.
The shift from manual monitoring to AI-powered video analytics isn't just an operational improvement — it's a fundamentally different security posture. The hospital moved from a system that records incidents to one that prevents them.
Built on what was already there. Extended by what wasn't.
Every technology decision served the outcome. The Cisco Meraki infrastructure was already in place — OnStak extended it, not replaced it. AWS provided the GPU compute needed for real-time inferencing at scale.