On a typical Tuesday at 10:15, a badge system reports 1,140 people in a headquarters building. The desk booking tool shows 68 percent of seats reserved. A walk across the fourth floor finds 31 people and 140 empty chairs. Three sources, three answers, and the facility manager is about to defend a floor-consolidation plan to a CFO with none of them. That gap is the reason building occupancy analytics has become a budget line rather than a nice-to-have: badge data counts entries, booking data counts intentions, and neither counts bodies in rooms.
Occupancy is not one number. In practice it splits into three layers, and most projects that disappoint tried to answer a question from the wrong layer.
A university asking whether to close a library wing at 22:00 needs counts and dwell. A corporate workplace team choosing between 400 and 600 desks needs counts at the floor level over a full quarter. A council running a public swimming pool needs a real time occupancy monitoring feed with a hard ceiling. Write the question down first; the sensor choice follows from it.
A desk occupancy sensor (typically passive infrared under the desk surface) answers presence for one seat. It will not tell you how many people are in the open-plan area, because visitors, standing conversations and people at shared tables are invisible to it. A ceiling-mounted AI counting sensor at a doorway or above a zone answers count, distinguishing individuals even when they walk in as a group. Wi-Fi and Bluetooth association counts are useful for trend lines across a campus but should never be reported as headcount; a person with a laptop, phone and watch can appear three times, and a visitor with Wi-Fi off appears zero times.
For anything where the number itself carries a decision, insist on a stated accuracy figure in the contract. Vemco's AI sensors carry a contractual minimum of 96 percent, and typically deliver 98 to 99 percent when lighting, entrance layout and how people move through the space allow it. The difference between those figures is worth understanding for your own building: on a 300-person floor, 96 percent means a possible error of about 12 people at any moment, which is irrelevant to a consolidation decision but matters if a room limit is 300 and you are alerting at 290.
A zone count is only as good as its perimeter. If a floor has a main stair, a lift lobby and a fire stair that people quietly use as a shortcut to the car park, and you sensor only the first two, the count will drift all day and show 40 people at 19:00 in an empty building. Every uncontrolled opening must either be counted or physically discouraged. Practitioners typically walk the floor at 07:30 and 17:30 during the survey phase specifically to find the doors nobody mentioned in the drawings.
Here is the observation most implementers learn the hard way: the largest source of error in a building count is rarely the sensor, it is cleaners, security rounds and facilities staff who cross zone boundaries dozens of times per shift. A caretaker checking six meeting rooms adds twelve door crossings to the morning data. Staff exclusion, where uniformed or badge-tagged personnel are recognised and removed from the count, is the feature to ask for, alongside a scheduled overnight reset so that any residual drift is cleared before the next day begins.
Raw counts are not analytics. Once the data is flowing into space utilization software such as VemSpace, four derived figures do most of the work:
Collect for at least six to eight weeks before drawing conclusions, and for universities make sure the window includes both a teaching week and an exam period; the two look like different buildings. Workplace occupancy analytics based on a fortnight in August will undersize every office in northern Europe.
Measurement pays for itself slowly through better space decisions and quickly through energy. Most BMS schedules ventilate a floor from 06:00 to 20:00 whether 200 people are present or 12. When an occupancy monitoring system feeds live zone counts into the building management system, as VemFusion does with HVAC, BMS and security platforms, ventilation and cooling can follow actual load per zone. The savings show up first on floors with early leavers: an area that empties at 15:30 stops being conditioned at full rate for four and a half hours it does not need.
The same feed supports safety. A lecture theatre licensed for 400, a sports hall or a public library reading room can trigger an alert at a predefined limit so front-of-house staff act before the number is exceeded rather than reconstructing it afterwards. Security teams also use the after-hours count to confirm a building is genuinely empty before arming, instead of relying on a badge-out that half the workforce forgets.
What is an occupancy sensor? An occupancy sensor is a device that detects whether people are present in a space and, in the case of counting sensors, how many. Simple types report presence for lighting or desk-availability control; AI-based counting sensors report headcount for a zone or entrance and form the basis of building occupancy analytics.
How do occupancy sensors work? Passive infrared sensors detect body heat moving within a field of view, which suits desks and small rooms. Ceiling-mounted AI sensors analyse a downward image to identify and count individual people crossing a virtual line or standing within a zone, and can exclude recognised staff so the count reflects genuine occupants rather than door traffic.
If you are weighing a floor consolidation, a ventilation schedule rewrite or a capacity limit for a public space and want to know which sensor mix and thresholds would give you a defensible number, talk to Vemco about a building occupancy analytics scope for your site.