A store manager tells you Saturday was "absolutely packed". The till says revenue was flat against the same Saturday last year. Both statements can be true at the same time, and without a footfall counter you have no way of knowing which problem you actually have: more people who bought less, or fewer people who bought more. Those two situations call for completely different decisions on staffing, range and layout, and the POS report alone cannot separate them.
That gap between "how many people came in" and "how many people paid" is the reason the footfall counter exists. What follows is how the device actually produces that number, where each sensor technology is a sensible choice, and where the count quietly goes wrong.
A footfall counter does one thing: it registers a person crossing a defined virtual line, in a defined direction, at a timestamp. Everything else, from hourly traffic curves to conversion rate to live occupancy, is arithmetic done on those crossing events. The sensor does not count "traffic" in a loose sense. It counts an object it has classified as a human passing from zone A to zone B.
Three properties matter more than the brochure suggests:
The technology inside the housing determines what the device can and cannot do. Most estates end up with more than one type, because a mall's main atrium, a fashion store's double doors and a car park lift lobby are not the same counting problem.
Modern sensors count at the edge. The image is processed on the device, the crossing is logged, and only the event data, not video, leaves the unit. That matters for GDPR and for network load: a sensor sending a few kilobytes of counts per hour is a very different conversation with the IT director than one streaming footage to a server.
The counts then go to a people counting system that does the work the sensor cannot. It reconciles multiple sensors covering one wide entrance so a person is not counted twice at the overlap. It applies opening hours so the night security round does not appear as visitors. It runs staff exclusion, so employees who cross the line forty times a shift are removed from the customer figure. And it joins the count to transactions from the POS to produce conversion rate per hour, per entrance, per store. In Vemco's case that layer is VemCount, and it is deliberately sensor-agnostic: a Xovis unit in the flagship and a Milesight unit in the stockroom corridor land in the same dataset.
For a mall operator the same pipeline runs at portfolio scale. Every tenant's door count and every common-area entrance feed a single footfall data set, which is what makes tenant benchmarking and turnover-rent verification defensible rather than anecdotal.
Anyone selling a footfall counter will quote a headline percentage. The honest version has two parts. Vemco's contractual minimum is 96%. Under good conditions, meaning adequate lighting, a mounting height the sensor was designed for, a layout without pinch points directly under the counting line, and ordinary visitor behaviour, installations typically run at 98 to 99%. Nobody should promise you a flat 99% before they have seen the door.
Accuracy is also not fixed at commissioning. Here is the observation every implementer eventually learns the hard way: a sensor validated at 99% in October can drift in February, because low winter sun through a glass facade throws hard-edged shadows across the counting zone that were simply not there during the summer install. The fix is usually a small adjustment to the detection zone or a shift of the counting line a metre inward, but only if someone notices. The same applies when a new promotional display is placed just inside the entrance and visitors start stopping directly on the line, generating repeated in-out crossings.
When validating, count manually for at least 200 crossing events during a busy period, not 20 during a quiet one. Group behaviour is where counters fail, and you will not see groups at 9:15 on a Tuesday.
Across a large retail people counting estate, the most damaging error is not a sensor that miscounts by 3%. It is a sensor that stops reporting and nobody notices for three weeks. A store whose footfall drops to zero while POS keeps running shows a conversion rate of several thousand percent, and if regional reports aggregate averages, that store's figure can be silently blended into a number that looks plausible. A power cycle after an electrical contractor's visit, a PoE switch swap, a firmware update that reset the network configuration: all of these produce gaps.
This is why sensor health monitoring belongs in the software layer, not in someone's Friday afternoon checklist. VemCount flags a sensor that has gone quiet or whose count pattern has changed abnormally before the gap reaches a report, so the conversation with the store is "your sensor has been offline since Tuesday", not "why does your March conversion look impossible".
Return to the Saturday from the opening. With a trusted count, the flat revenue splits into one of two stories. Footfall up 18% and conversion down five points means the store was understaffed at the fitting rooms, or the promotion drew browsers rather than buyers. Footfall down 12% and conversion up means the marketing spend that weekend did not bring anyone to the door and the sales team compensated. The first is a labour-scheduling decision, the second a marketing one. Footfall data is what tells you which meeting to call on Monday.
For facility managers the same events become live occupancy, which drives cleaning rotas and HVAC schedules. For mall operators, footfall counters at tenant doors and common entrances turn the leasing conversation from "the centre is busy" into a per-tenant capture rate the prospect can check against their own trading model.
How does a footfall counter work? A ceiling-mounted sensor, most commonly 3D stereo or time-of-flight, builds a depth image of the area beneath it and identifies each object shaped and sized like a person. When that object crosses a virtual line in a given direction, the sensor logs a timestamped in or out event. Software then aggregates those events, removes staff and applies opening hours to produce visitor counts, occupancy and conversion.
How accurate are footfall counters? Well-installed 3D sensors typically achieve 98 to 99% under good lighting, correct mounting height and normal visitor behaviour, with Vemco contracting to a 96% minimum. Accuracy can drift after installation due to changing light, new fixtures near the entrance or crowding patterns, so periodic manual validation against several hundred crossings is sensible. Cheaper break-beam or Wi-Fi methods fall well below these figures.
If you are specifying a people counting system for a new estate, or trying to work out why two stores with the same sensor model report very different accuracy, talk to Vemco about a site-by-site sensor assessment. We will tell you which entrances need stereo vision, which can take a cheaper unit, and how to keep the count trustworthy after the installers have left.