The camera has become the operating system of the physical world

For more than thirty years, commercial organisations have quietly built one of the largest sensor networks in history. Retailers installed cameras to reduce shrinkage. Shopping centres, airports, universities and municipalities followed to improve security and safety. Today, millions of CCTV cameras watch the spaces where people work, shop, travel and gather: An enormous investment that, for most organisations, still serves a single purpose: recording what happened. That's the job security systems were built for: detecting incidents and raising alarms. Anyone who wanted to measure footfall, occupancy or customer movement had to invest in separate, dedicated sensors, because conventional CCTV couldn't deliver reliable operational data.

That is changing. Advances in artificial intelligence, computer vision and edge computing mean the cameras organisations already own can now do what only dedicated sensors once could. This is where PFM comes in: turning footfall data into actionable insights, so the same infrastructure that protects a location can help improve customer experience, optimise staffing and manage buildings more efficiently.

In other words, the camera is evolving from a monitoring device into a business intelligence sensor. And that changes everything. We spoke to Michel l' Amie, our Chief Product and Technology Officer (CPTO), about the ins and outs of footfall intelligence generated by existing CCTV infrastructures.

Every building is already full of untapped intelligence

For decades, organisations have approached operational technology by adding more infrastructure: One device per desired use case. This approach wasn't wrong. In fact, dedicated sensors remain the most accurate solution for many environments, particularly where contractual reporting or billing depends on precision. At PFM, we continue to recommend technologies such as our 3d- sensors whenever customers require accuracy levels exceeding 99%, because the right technology depends on the problem you're solving.

What has changed is the assumption that every new insight requires another piece of hardware. Take a modern shopping centre. It may already have 150 cameras covering entrances, corridors, food courts, car parks and public spaces. A retailer might have cameras above every entrance, checkout and sales floor. Airports routinely operate hundreds, sometimes thousands, of cameras across terminals. Viewed differently, those organisations are already sitting on one of the richest networks of sensors imaginable.

The cameras were never the limitation: Our ability to interpret what they were seeing was.

"Most of our customers already have CCTV cameras deployed," explains Michel l' Amie, CPTO at PFM Intelligence. "Those cameras are running 24 hours a day, but they generate no business intelligence whatsoever. The problem we're solving is not a lack of cameras. It's a lack of intelligence from cameras that are already there. The cameras were already paid for. We just unlock them."

AI video analytics has reached a turning point

Many people assume that camera-based people counting is a recent idea. But that isn't true. The industry has experimented with CCTV analytics for years. The problem was that earlier generations of technology relied on techniques that struggled in real-world environments. Changes in lighting, shadows, reflections and crowded scenes could all reduce reliability, making dedicated people counting sensors the preferred option for organisations that needed dependable data. Artificial intelligence has changed that equation.

Modern neural networks don't simply compare pixels between video frames. They recognise objects, distinguish between people and other moving objects, understand spatial relationships and continuously improve their ability to interpret complex environments.That leap is what makes today's AI video analytics fundamentally different from the systems many organisations dismissed a decade ago.

"In the past, trying to extract reliable people counting data from a standard security camera was genuinely troublesome," says l'Amie. "Machine learning models, neural networks and AI running on edge devices have advanced rapidly in the last few years. The accuracy and reliability of camera-based object detection is now in a completely different league."

The timing matters: Commercial organisations face growing pressure to improve efficiency, enhance customer experience and make evidence-based decisions while controlling costs. Replacing an entire CCTV estate with new hardware is rarely attractive. Unlocking intelligence from infrastructure that already exists is.

That is why AI video analytics is rapidly becoming one of the most compelling developments in the movement intelligence market.

Beyond CCTV analytics: understanding how places actually work

The first question many organisations ask is whether AI video analytics can deliver accurate people counting using their existing CCTV. The answer is: It can. (With the side-note that adding a dedicated people counting sensor allows you to boost accuracy even more) But stopping at counting misses the larger opportunity. Counting tells you how many people entered your location; understanding their behaviour tells you why it performs the way it does, and what to change.

Once existing cameras become intelligent sensors, the insights turn directly into action. Retail video analytics reveal where visitors dwell longest and which entrances perform best, so merchandising and campaigns go where they'll actually be seen. Queue analytics show where lines start building, so managers can open a till or reposition staff before customer satisfaction takes the hit. Heatmaps and customer journey analytics expose how people really move through a store or shopping centre. This is Evidence to redesign layouts, place tenants and price retail space on footfall rather than gut feeling. And occupancy monitoring allows facilities teams to adjust cleaning and staffing schedules to how the building is actually used.

Week over week, season over season, these insights compound into something more valuable than any single report: a continuous, evidence-based understanding of how your location works - and where it can work better.

In the video below, you see our Market Development Director Mark King talk about PFM's CCTV AI capabilities at our Customer Event at Future Stores in London in June 2026.

Taken together, these insights transform CCTV from a passive recording system into a continuous source of business intelligence.

For retailers, this means understanding how merchandising decisions influence shopper behaviour. For shopping centres, it means measuring the impact of events, tenant changes or new entrances. Airports gain a clearer picture of passenger flow. Municipalities can better understand how pedestrians and cyclists use public spaces.

The camera becomes more than a security device. It becomes a decision-making tool.

Is camera-based analytics GDPR-compliant?

Whenever AI and cameras are discussed together, privacy inevitably enters the conversation, and we understand that very well. Responsible innovation depends on trust. At PFM, we pride ourselves that our solutions are 'privacy-by-design'.

Fortunately, modern GDPR-compliant video analytics is designed very differently from the surveillance systems many people imagine.

Processing takes place locally on dedicated hardware installed at the customer's own location. Video does not need to be streamed into the cloud for analysis. People and cars are anonymised in real time, and the platform transmits metadata rather than images. What enters dashboards is occupancy data, footfall counts, dwell time and movement statistics, not identifiable footage.

"The camera sees movement," says l' Amie. "The software transforms it into data and immediately discards the image in milliseconds."

This privacy-by-design approach allows organisations to gain operational insight while respecting individual anonymity, demonstrating that intelligent buildings do not need to become surveillance environments.

Why PFM chose this direction

At PFM, our ambition has always been to help organisations unlock the full potential of their locations through movement intelligence.

Dedicated precision sensors remain an essential part of that vision. Cost-effective people counters continue to play an important role. But if we truly want to help organisations understand how their spaces perform, we also need to embrace the infrastructure that already surrounds us.

By combining perception software with PFM's movement intelligence platform, we can transform existing IP cameras into intelligent business sensors. Organisations can begin with CCTV people counting and footfall analytics, then expand into dwell time analytics, occupancy monitoring, customer flow analysis, heatmaps and advanced operational insights, all using the cameras they already own.

Data flows securely from the camera into our intelligence platform built on more than 40 years of experience turning footfall data into actionable insight.

Our guiding principle is simple: deliver data that is Accurate, Reliable and Consistent. This ARC standard gives customers the confidence to work actively with data generated through CCTV and translate it into business intelligence that supports better operational and strategic decisions.

Our platform PFM Advantage brings these insights together in one connected environment. Supported by our data analysts, who continuously work with and monitor customer data, the platform helps organisations build a richer and more complete understanding of how their locations perform over time.

Importantly, this is not about replacing every existing solution with camera analytics: Some environments will always require the unparalleled precision of dedicated people counting sensors. Others may simply need straightforward counters.

Our responsibility is not to sell a particular technology. It is to recommend the technology that best helps each customer solve their challenge. That philosophy is what we live by.

Every camera is waiting to become something more

History has a habit of making technological revolutions look obvious in hindsight. Today it feels perfectly natural that a smartphone is also a camera, a map, a wallet and a navigation system - yet each of those capabilities emerged as software unlocked value from hardware people already owned.

The same shift is now reaching the millions of CCTV cameras watching our shops, airports, campuses and cities. For thirty years they have recorded what happened. Now they can help decide what happens next. The investment has already been made, the infrastructure is already in place.

As Michel l' Amie puts it:
"Our ambition is to unlock location potential everywhere. Every camera is a sensor waiting to be activated."

Yours included.

Frequently Asked Questions:

Can I use existing CCTV cameras for people counting?
Yes. Any IP camera that supports the standard RTSP protocol can be used for people counting, with no camera replacement required. A single AI compute box installed at your location connects to the cameras you already have and starts generating footfall counts, occupancy levels, dwell times, heatmaps and crowd density data.

How accurate is AI people counting with CCTV cameras?
AI people counting on existing CCTV cameras achieves 95%+ accuracy, sufficient for the vast majority of operational use cases. Where guaranteed 99%+ accuracy is non-negotiable , e.g. for billing contracts or hard SLAs, adding dedicated 3D sensors such as Xovis remains the right choice, and that's what PFM will recommend. The two technologies are complementary, not competing.

Is AI video analytics GDPR-compliant?
Yes. AI video analytics can be fully GDPR-compliant when built privacy-by-design: video is processed locally, anonymised in real time, and never stored or transmitted. All processing happens on hardware at your own location, faces and license plates are pixelated at the moment of analysis, and the image is immediately discarded. What leaves your building is a number, not a face. Full technical and legal documentation can be provided.

What hardware do I need for CCTV people counting?
No new cameras — just one AI compute box that connects to your existing CCTV infrastructure. Any RTSP-compatible IP camera works, and standard Full HD resolution is sufficient. Because there's no sensor installation project, an organisation with hundreds of cameras can be generating insights within weeks rather than months.

How do I know if AI video analytics will work with my cameras?
You can test it with your own footage before committing: PFM runs a short clip (ten minutes is enough) from your cameras through the system and shows you the results. You see the detection accuracy and the real-time anonymisation working in your actual environment and lighting conditions, before any decision is made.