Complete Guide to Iot Occupancy Sensors

Walk through most office buildings at 4 p.m. on a Friday and you will find the HVAC running at full capacity for floors that emptied at lunchtime. The BMS is doing exactly what its schedule tells it to do — and the schedule was written for an occupancy pattern that stopped existing years ago. That gap between assumed occupancy and actual occupancy is where IoT occupancy sensors earn their budget line, and it is also where most deployments quietly underperform because the wrong sensor type was specified for the wrong job.

This guide is written for teams who already know the elevator pitch. Instead, we will look at what actually separates a deployment that feeds reliable data into automation logic from one that produces dashboards nobody trusts after month three.

Presence detection versus people counting: the distinction that shapes everything

The term "occupancy sensor" covers two fundamentally different technologies, and conflating them is the most common specification error we see in tender documents.

• Presence sensors (PIR, ultrasonic, mmWave radar) answer a binary question: is anyone in this zone? They are cheap, private by design, and perfectly adequate for switching lights in a storage room. They cannot tell you whether a meeting room holds two people or twelve.
• People counting sensors (typically ceiling-mounted AI vision or stereo optical units at entrances and thresholds) maintain a running count. They answer the questions that drive real decisions: how many people are on this floor right now, when does the canteen peak, which of our forty meeting rooms are actually used at capacity?

If your use case is demand-controlled ventilation, capacity compliance, or space consolidation decisions, you need counting — not presence. A PIR sensor will report a floor as "occupied" whether one security guard or two hundred employees are present, and your air handling unit will respond identically to both situations.

Accuracy claims: what to demand in writing

Every vendor quotes an accuracy figure. Very few will put it in a contract. Insist on both a contractual floor and an honest explanation of what moves the number. As a benchmark: Vemco commits to a minimum of 96% counting accuracy contractually, with 98–99% typically achieved when conditions allow — good lighting, sensible sensor placement, and predictable visitor flow. Any vendor promising a flat 99% regardless of environment is telling you something about their sales process, not their sensors.

Two accuracy killers deserve specific attention during procurement:

• Staff contamination. In a university building, cleaning and facilities staff can cross a counting line dozens of times per shift. Without staff exclusion — where AI sensors filter out badge-wearing or otherwise identified personnel — your utilisation figures inflate and your capacity alerts fire falsely. Ask how exclusion is handled and whether it requires staff cooperation to work.
• Threshold geometry. Wide entrances, glass doors reflecting sunlight, and lobbies where people linger under the sensor all degrade counts. A good integrator flags these during the site survey, not after commissioning.

The integration layer is where the ROI lives

A sensor that only feeds a dashboard pays back slowly. A sensor that feeds automation pays back on the next utility bill. The economics of IoT occupancy sensors change entirely once the data reaches your building systems in real time. This is why the middleware question matters more than the sensor brand: how does occupancy data reach your HVAC controls, your BMS, your security platform?

Platforms like Vemco's VemFusion exist precisely for this handoff — connecting live occupancy data to HVAC, BMS and security systems so that ventilation follows actual headcount rather than a timetable. In offices, universities and public buildings, this is where the waste from always-on systems gets cut: a lecture hall that ventilates for 200 people only when 200 people are actually in it, an office floor where the BMS steps down setpoints the moment the last person badges out through a counted exit.

During evaluation, ask vendors three specific integration questions:

• Latency: how fresh is the count when it reaches the BMS? Fifteen-minute batch updates are useless for demand-controlled ventilation.
• Protocol support: can it speak to what you already run — BACnet, REST APIs, MQTT — or does it demand a proprietary gateway?
• Alerting logic: can you set a predefined occupancy limit per zone and trigger real-time alerts when it is breached? This is essential for capacity compliance and safety use cases, and surprisingly often missing.

A practitioner's note on drift

Here is something rarely mentioned in vendor material: bidirectional counting sensors accumulate drift. If a sensor at a floor entrance is 98% accurate, the small residual error in "in" versus "out" counts compounds over a day, and by evening the system may believe three people remain on an empty floor. Mature deployments handle this with scheduled midnight resets, cross-validation between sensors, or reconciliation against access control events. Ask your vendor how they handle drift correction. If the answer is a blank look, keep shopping — this single detail separates integrators who have run occupancy systems in production from those who have only demoed them.

From counts to floor plans: the space utilisation layer

Automation is the fast payback; portfolio decisions are the big one. Twelve months of zone-level occupancy data, analysed through a space utilisation tool such as VemSpace, tells you which floors you can consolidate, which meeting room formats your organisation actually uses, and whether that lease renewal needs all six floors or four. These are seven-figure decisions, and they should not rest on badge swipes — which count arrivals, not presence — or on a facilities manager's walk-through impressions.

The teams getting the most from this data share one habit: they define the decisions they want to make before choosing sensor placement. Counting at building entrances answers headcount questions. Counting at floor and zone thresholds answers consolidation questions. Room-level sensing answers meeting-space questions. Each layer costs more, so map the sensor plan to the decisions, not the other way round.

A realistic rollout sequence

• Pilot one floor with counting sensors at every threshold, and validate accuracy against manual spot counts for two weeks.
• Connect one HVAC zone to live occupancy data and measure energy delta against a comparable uncontrolled zone.
• Only then scale — with drift correction, staff exclusion and alert thresholds proven, not promised.

Teams that skip the pilot phase almost always end up re-commissioning sensors later, at higher cost and with credibility already spent internally.

If you are specifying an occupancy sensing project — whether it is demand-controlled ventilation for a campus, capacity alerting for a public building, or utilisation data ahead of a lease decision — talk to a team that has commissioned these systems in production environments. Contact Vemco Group at vemcogroup.com/contact-us to discuss sensor selection, accuracy validation and BMS integration for your specific building.