Only about 22% of digital out-of-home networks run sensor technology for audience measurement, the foundation of any digital signage analytics, even though the global DOOH market was already worth $25.52 billion three years ago. Most screens still broadcast blind. They play a loop, hour after hour, with no idea who walked past, who looked, or whether the 11 a.m. spot reached anyone at all.
For an integrator, that blind spot is the opportunity. Advertisers and venue owners increasingly want proof that a screen did its job, and a network that can answer “who watched, and for how long” is worth more than one that cannot. This is where digital signage analytics comes in: a layer of sensors and software that measures real audiences in front of real screens. Sensors turn a signage network from a broadcast medium into a measurable one, and that measurability is exactly what you can package, prove, and sell.
This guide covers what the technology measures, how the sensors work, where they fall short, and how the data becomes revenue. If you are weighing whether to build it into your stack, it helps to know what you are signing up for before committing to custom digital signage software.
What is digital signage analytics?
Digital signage analytics is the practice of measuring what happens around and in response to your screens. The term covers two different things, and mixing them up leads to bad buying decisions. Before looking at sensors, it is worth separating the two.
Audience analytics vs operational analytics
Operational analytics tells you whether the network is healthy: whether the player came online, the right campaign played, and the screen is still showing content rather than a black rectangle. That is the realm of proof of play and device monitoring, which most content management systems already handle.
Audience analytics tells you something the CMS cannot see on its own: who stood in front of the screen and how they reacted. That data comes from a sensor watching the space, not from the player reporting its own status. Both matter. Only one of them puts a number on the audience.
| Operational analytics | Audience analytics | |
|---|---|---|
| Question it answers | Did the screen play the content? | Who saw the content, and how? |
| Data source | The media player and CMS logs | A camera sensor watching the space |
| Typical metrics | Uptime, proof of play, campaign logs | Viewers, dwell time, attention, demographics |
| Who cares most | Operations and SLA reporting | Advertisers, venue owners, media buyers |
| Built in by default | Usually yes | Almost never |
When someone says they want digital signage analytics, they almost always mean the second column. The first is table stakes.
The metrics that actually matter
A good audience sensor produces a small set of numbers, and each one answers a question an advertiser would ask.
- Viewer count – how many people stood in front of the screen during a spot, the raw reach number the industry calls people counting, and the signage equivalent of an impression.
- Attention time (dwell time) – the total time people spent looking at the screen rather than walking past it. A sensor that tells "present" from "looking" is far more useful than one that only counts bodies.
- Looked count – how many distinct people glanced at the screen at all. Someone who looks, turns away, then looks again still counts once, so the figure stays honest.
- Demographics – an estimated split by gender and age band for the people detected, which is what lets a buyer say a screen skews young, or female, or older commuters.
Put together, these turn a vague "lots of footfall" into something concrete:
This spot reached 240 people, 87% of them looked, average attention just under two seconds, and the crowd skewed 18 to 34. That is a sentence a media buyer can price.
Anonymous video analytics, explained
Anonymous video analytics (AVA) is the camera-based technology behind most sensor-driven digital signage analytics. A camera observes the area in front of the screen, and software estimates how many people are present, how long they look, and their rough age and gender, without ever identifying anyone. No image is stored, and no face is matched to a name.
The word anonymous is doing real work here. AVA detects that a face exists and estimates broad attributes from it. It does not ask whose face it is. We will come back to why that matters legally, but the short version is that AVA and facial recognition are different technologies that happen to share a camera. Counting an anonymous face is not the same as recognising a person, and conflating the two is the fastest way to lose a privacy-sensitive client.
How do sensors capture audience data?
The hardware behind sensor-based digital signage analytics looks deceptively simple. A camera sits above or beside the screen, and a compute device next to it does the thinking. What happens between the lens and the dashboard is where the engineering lives.
From camera to anonymous aggregates
In a typical setup, one camera covers one screen, and a single box does double duty. It plays the content on the display and analyses the camera feed at the same time, so the same device is both the media player and the sensor. All of the image processing happens on that box, locally, and the video never travels anywhere.
The processing runs as a short pipeline, repeated frame after frame:
- Detect people in the frame, separating human shapes from the background.
- Track them across frames so the same person is not recounted on every move. This step is what stops one shopper from inflating the count into ten.
- Find and read the face, estimating age band and gender for anyone close enough to classify.
- Aggregate the results per campaign, then send only the totals to the CMS on a regular schedule, usually hourly.
The frames are analysed in memory and discarded at once. What leaves the device is a set of counts, for example "five women aged 25 to 34 looked at this campaign between noon and 1 p.m." Never an image, never a clip. Many of these systems run on open-source detection models, the same families the established audience-measurement vendors rely on, which keeps the approach transparent and auditable.
The limits of computer-vision accuracy
No vendor's marketing page will tell you this, so here it is plainly: the sensor data behind digital signage analytics is directional, not forensic. The numbers are good enough to compare screens, dayparts, and campaigns, but they carry a margin of error you should design around. Most of it comes from the messy reality of a public space, not from bad software.
- Double counting under occlusion – when one person briefly hides another, the hidden person can be counted twice as they reappear. Busy walkways where people cross in both directions are the worst case.
- Reflections – a glossy floor or a shop window can bounce back a reflection the model reads as another person.
- Split detections – a single person is occasionally registered as two, especially at the edges of the frame.
- Gender and age drift – classification is an estimate, women are sometimes read as men, and exact age is unreliable.
- Hostile light – strong backlight, the classic shop-window glare, and a glass panel in front of the screen all degrade the image the model works from.
The age question deserves a note. Ask a model for an exact age and it jumps around, reading the same person as 27 one second and 31 the next. Age bands smooth that noise into something stable, which is why the whole industry, from Quividi to V-Count, reports in ranges rather than single years. Treat the output as a reliable trend line, not a headcount certified to the last person, and it will serve you well.
Privacy and GDPR by design
For a lot of clients, privacy is the first question, not the last. The reassuring part is that a well-built sensor is private by architecture, not by promise.
Three design choices do the heavy lifting. The camera feed is processed on the device and never transmitted. Individual frames are analysed and then deleted, so nothing is recorded. And only aggregate counts reach the CMS, never data about a specific person. There is no facial recognition and no memory of individuals, so a shopper who returns the next day becomes a fresh, anonymous count.
Because no personal data is stored or used to identify anyone, anonymous video analytics generally sits on the right side of the GDPR and regimes such as CCPA. That is not a blanket exemption. You should still post clear notice that measurement is in use, and in some venues an opt-out is the sensible minimum. On-device processing, no stored images, and aggregate-only reporting are what turn "we use cameras" from a red flag into a selling point.
Turning analytics into revenue
Measurement is not the goal. What integrators and their clients want is the business case that digital signage analytics unlocks, and it shows up in three places.
Proof of performance and higher-value inventory
A screen with no audience data is sold on guesswork, usually by location and a rough footfall estimate. A screen with audience data is sold on evidence. When you can show an advertiser that their spot reached a measured number of viewers, with a known attention time and demographic profile, the conversation changes from "trust us" to "here are the numbers."
That evidence has direct commercial value. Proof of play confirms a spot ran; audience data confirms it was seen, and by whom. Together they justify higher CPMs and make programmatic buying viable, because programmatic DOOH demand wants exactly this kind of audience signal to bid against. Networks that feed verified audiences into the programmatic pipeline, often by connecting their screens to SSP demand, command better rates than those selling blind loops.
From measurement to reactive content
Once you can see the audience, the next step is responding to it. The simplest version is dayparting informed by real data: if the morning crowd skews one way and the evening crowd another, the schedule follows. The more advanced version is content that reacts in near real time to who is in front of the screen.
That reactive layer is where measurement meets automation. Audience signals can trigger content rules on their own, and pairing them with AI-driven campaign management moves a network from reporting what happened to adjusting while it happens. It is also where pulling in real-time data feeds pays off, since one screen can react to both its audience and the world around it. Worth flagging: this is a bigger build than measurement alone, and most networks earn the right to it by getting measurement solid first.
Build vs buy: the role of a software partner
You have two routes to digital signage analytics. Off-the-shelf measurement vendors get you running quickly, but you rent the capability, your data lives in someone else's platform, and integrating it with your own CMS is rarely clean. Building your own gives you control over the data, the dashboards, and how analytics sits inside the rest of your stack, at the cost of real computer-vision and software work.
For most integrators the honest answer is in between: own the parts that differentiate you, the CMS, the reporting, the client-facing layer, and bring in specialists for the heavy computer-vision lifting. If audience measurement is going to be part of what you sell, it belongs inside software you control, not bolted on from a black box you cannot see into. That is the kind of build a digital signage software development partner exists for, turning a sensor proof of concept into something you can deploy across a network and put in front of clients.
FAQ - Digital signage analytics
What is digital audience measurement?
Digital audience measurement is the use of sensors and software to count and describe the people who see a digital screen, rather than guessing from foot traffic. In digital signage it usually means a camera-based system that records how many viewers were present, how long they looked, and their estimated age and gender, all anonymously. The output is aggregate audience data, not information about any individual.
What are the metrics for digital signage?
Audience-focused digital signage analytics tracks a small set of core metrics:
- Viewers – people detected in front of the screen.
- Attention time – how long they looked.
- Looked count – how many distinct people glanced at the screen.
- Demographics – estimated gender and age-band split.
- Plays and proof of play – how many times each spot ran.
Operational metrics like uptime sit alongside these, but the audience metrics are the ones advertisers pay attention to.
How are audiences measured?
Audiences are measured by a camera sensor mounted near the screen, feeding a computer-vision model that detects people, tracks them to avoid double counting, and estimates attention and demographics. All of this runs on a local device, and only aggregate numbers are sent onward. No footage is stored and no one is identified, which is why the method is called anonymous video analytics.
Does digital signage audience measurement use facial recognition?
No. Properly built systems use face detection, not facial recognition. Detection notices that a face is present and estimates broad attributes like age range and gender; recognition tries to match a face to a specific identity. Anonymous video analytics stops at detection, stores no images, and keeps no record that could re-identify anyone, which is what keeps it compliant and worth stating up front to any privacy-conscious client.
How much do digital signage audience sensors cost?
There is no single price for people counting sensors, because cost scales with the deployment rather than the hardware alone. The main drivers are:
- Camera quality – low-light performance and the right field of view matter more than raw resolution.
- Compute per screen – each screen needs a device that can play content and run the model at once.
- Number of screens – costs are largely per endpoint, so a network multiplies quickly.
- Software and CMS integration – getting the data into dashboards your clients will use is often the largest line item.
- Ongoing support – calibration, updates, and fleet monitoring.
For a network, the software and integration work usually outweighs the hardware over time.
How accurate is digital signage audience measurement?
Accuracy is good enough to compare screens, campaigns, and times of day, but it is an estimate, not a census. Counting is the most reliable part; demographic classification is softer, which is why age is reported in bands rather than exact years. Real-world conditions such as crowding, reflections, and harsh backlight add a margin of error. Used as a trend signal rather than an exact headcount, the data is dependable enough to price and plan inventory.
