Occupancy Sensing for HVAC, Lighting and Space Planning

A client says they want occupancy data. Ask what decision it will support and you get three answers in one sentence: to stop conditioning empty areas, to work out whether the third floor is needed, and to show the executive team a dashboard. Those are three different measurements with different accuracy requirements, different sensors, different costs and, importantly, different privacy exposure. Treating them as one requirement is why so many occupancy projects deliver a dashboard nobody acts on.

The message up front: decide first whether you need presence, count or utilisation. Presence is binary, cheap, reliable, and enough for most control purposes. Count is substantially harder and more expensive, and it is what demand-controlled ventilation and real space planning need. Utilisation is count aggregated over time, and its value depends entirely on whether anyone will act on it. Buy for the decision you will actually make, pick a technology whose known failure mode you can live with, and settle the privacy position in writing before the first sensor is ordered rather than after somebody notices it.

1. Presence, count and utilisation are three different things

Almost every disappointment in this area traces back to conflating these.

• Presence. Is anybody here, yes or no. Cheap, well understood, and sufficient for switching lighting, setting back a zone, releasing a meeting room booking or enabling ventilation. The classic failure is the opposite of dangerous: a sensor that misses a still person and switches the lights off.
• Count. How many people are here. Considerably harder, because it requires distinguishing individuals rather than detecting motion. This is what ventilation rate modulation and genuine space planning need, and it is where accuracy claims should be examined most sceptically.
• Utilisation. Count or presence aggregated over time and compared with capacity: how often a space is used, by how many, at what times. A reporting product built on the layer below it, and the one most often bought before anyone has decided who will act on the answer.

A fourth exists and should be named in order to be excluded: identity, meaning who specifically is where. Access control systems and desk booking already hold this, and almost no facilities use case requires it. Collecting it is a materially different proposition legally and culturally, and I would advise keeping it firmly out of scope unless there is a specific, documented reason that survives a conversation with your legal and people teams.

2. The sensing technologies, and what each actually tells you

The table below is editorial assessment of how these behave in buildings, not vendor specification. Treat accuracy claims from any supplier as a hypothesis to be tested in your own space, because performance depends heavily on geometry, mounting height and how the space is used.

3. Feeding HVAC: where occupancy earns its money

The energy case for occupancy sensing is almost entirely in the air side, and it works through a small number of mechanisms:

• Occupied and unoccupied setpoint bands. Widening the acceptable temperature band when a zone is empty, rather than conditioning it to comfort conditions all day. Needs only presence, and it is usually the largest and cheapest win.
• Ventilation modulation. Reducing outdoor air to unoccupied or lightly occupied zones. This needs count or a credible proxy, and it interacts with air quality and statutory ventilation requirements, so the minimum rates and the control strategy must be set by the mechanical designer against whatever standard your jurisdiction adopts, not inferred from a sensor reading.
• Terminal unit control. Closing down airflow to unoccupied zones within an occupied floor. The mechanics belong with the sequences described in BMS in HVAC: controls, points and sequences, and the plant itself in what a VAV system is.
• Optimised start informed by actual patterns. Using observed arrival times rather than a fixed schedule, so the building warms or cools to suit how it is really used.

Two cautions that apply to all of them. First, occupancy-driven control only pays where the plant can actually turn down; a system running at a fixed rate regardless of demand will not save anything no matter what the sensors say. Second, every setback has a recovery cost, and aggressive setbacks on a building with slow recovery produce comfort complaints that get the whole strategy disabled. The energy modelling context sits with energy management systems for buildings.

4. Feeding lighting, and why it is the easy case

Lighting is where occupancy sensing has the longest track record and the simplest economics: presence detection plus a time delay, usually with daylight dimming alongside it. The engineering is mature, the payback is well established, and modern luminaire-level sensing makes the granularity much finer than it used to be.

Two things are worth taking from the lighting world into the rest of the building. One is that the time delay matters more than the sensor: too short and people sit in the dark, too long and the saving evaporates, and tuning it per space type is the whole job. The other is that luminaire-level sensors are, incidentally, a dense grid of presence data across the whole floor, often at a finer resolution than anything bought specifically for occupancy analysis. Whether that data can be extracted depends on the lighting control system's interfaces, and it is worth asking before buying a separate sensor network to measure the same thing.

5. Feeding space planning and utilisation

This is where the money is at portfolio scale, because property is expensive and most organisations do not know how much of theirs is used. The questions utilisation data can genuinely answer:

• Peak against average. Both matter and they lead to different decisions. Average occupancy drives disposal and consolidation decisions; peak drives how much space is needed on the busiest day, which is what people actually complain about.
• Booked against used. Comparing room booking records with measured occupancy. This usually produces the most immediately actionable finding in any utilisation study, and it is cheap because half the data already exists.
• Pattern by day and by time. Which days are busy, when people arrive and leave. This feeds both space decisions and the HVAC scheduling above.
• Space type performance. Whether the large meeting rooms are used for large meetings, which they commonly are not, and whether the quiet spaces are used at all.
• Change over time. The only way to know whether a workplace change achieved anything. Which means a baseline, taken before the change, and almost nobody has one.

The methodological trap is the short study. A two-week sensor deployment during an atypical fortnight produces a confident number that is wrong, and it is wrong in a way nobody can detect afterwards. Occupancy varies by season, by week of the month, by project cycle and by whatever was happening in the organisation that month. Either measure for long enough to see that variation, or be explicit that you have a snapshot and treat it as one.

6. The warehouse and industrial case, which is different

Occupancy monitoring in a warehouse is not a smaller version of the office problem. The drivers are different, and so are the useful technologies.

• Conditioning and lighting large volumes. A high-bay space is expensive to light and sometimes expensive to condition, and occupancy is genuinely sparse and mobile. Zone-level presence detection on aisles can pay quickly, and it is one of the clearer cases for the technology.
• Activity rather than headcount. What operations usually wants is not how many people are in the building but where work is happening, which overlaps with travel and congestion analysis rather than with occupancy sensing proper. That is the subject of warehouse heat maps.
• Safety separation. Detecting people in areas where powered equipment operates is a safety function, and it has a completely different reliability requirement from a comfort or energy function. A sensor whose failure mode is an occasional missed detection is acceptable for switching lights and is not acceptable as a protective function. Protective detection is the province of functional safety engineering and the standards adopted for it in your jurisdiction, and it should be specified and validated by a competent safety engineer, not derived from a building occupancy project.
• Environmental monitoring alongside it. In storage environments the occupancy question usually arrives attached to temperature, humidity and air quality monitoring, which is covered in environmental monitoring in warehouses.

7. Privacy, and settling it before procurement

This is the part that derails occupancy projects after installation, when it would have cost nothing to resolve beforehand.

The legal position depends on your jurisdiction and on what you collect, and I am not going to summarise anyone's data protection law here, because the duties differ by country and a confident generalisation would be wrong somewhere. In the European Union and the United Kingdom, personal data processing falls under the GDPR regime and there are specific obligations around workplace monitoring; many other jurisdictions have their own regimes with different requirements. Establish which applies to you, and take advice on it rather than inferring from a vendor's reassurance.

What does transfer between jurisdictions is a set of design decisions that reduce the exposure substantially:

• Count, do not identify. Choose technologies that cannot distinguish individuals where that is sufficient. A thermal array or a time-of-flight sensor answers most questions a camera would, without producing an image of anyone.
• Aggregate at the edge. Where a device does process images, prefer one that derives a count on the device and transmits only the number. What never leaves the sensor cannot be requested, leaked or repurposed.
• Set a spatial floor. Decide the smallest area you will report on, and enforce it. Floor or zone level reporting answers space planning questions; desk-level reporting in a small team is individual monitoring in all but name, whatever the sensor technology.
• Set a retention period and automate it. Utilisation analysis needs months of aggregate data, not years of raw records.
• Separate it from identity systems. The moment occupancy data is joined to badge or booking records, you have created personally identifiable monitoring, possibly without intending to. If that join is needed, it is a deliberate decision with its own governance, not an integration convenience.
• Tell people, in advance, in plain language. What is measured, what it is not, who sees it, how long it is kept, and what decisions it informs. Projects that do this encounter far less resistance than projects discovered after the fact, and the credibility cost of being found out is difficult to recover.

8. Where the data should actually land

A recurring pattern: occupancy sensors are installed, they come with their own cloud dashboard, and the data lives in a silo nobody logs into after month three. Deciding the destination before the procurement avoids this.

• To the BMS, for control. Presence and count used in sequences must reach the controller layer, which means a real protocol integration rather than a dashboard. The mechanics are in IoT and BMS integration, and the addressing detail in BACnet objects, device instances and addressing. Specify which objects carry the values, and at what update rate, rather than leaving it to the installer.
• To an analytics layer, for insight. Utilisation reporting, trend analysis and comparison across sites belong above the BMS rather than inside it, for the reasons in from BMS to building analytics.
• To the space records, for planning. Utilisation is only meaningful against an accurate space register: area, capacity, type, department. If the space data is out of date, the utilisation percentages are arithmetic on a fiction, and in my experience the space register is the weaker of the two data sets more often than the sensors are.
• Not to three places with three different definitions. Agree what counts as occupied, over what interval, and define it once. Two dashboards reporting different occupancy for the same floor is the fastest way to lose the credibility of the whole programme.

The security dimension applies here as to any added network of devices on a building network, and is covered in BMS cybersecurity for connected buildings.

9. The honest part: the data is rarely the obstacle

Occupancy projects are usually presented as measurement problems and are usually, in the end, organisational ones.

The technology works well enough. You can establish within a reasonable margin how much of a building is used and when. What tends not to happen is the decision that was supposed to follow. Utilisation data says a floor is half used, and the floor stays, because giving up space is a departmental loss, the lease has years to run, the headcount forecast assumes growth, and somebody senior likes the view. The data was never the constraint.

This is not an argument against measuring. It is an argument for establishing, before procurement, who owns the decision the data is meant to inform, what they have already committed to doing with it, and what result would actually change their mind. Where that person cannot be named, a smaller and cheaper study is the right answer, because the output is going to be a dashboard either way and a dashboard does not need a full sensor deployment.

The same applies to the control case, with better odds. Occupancy-driven setbacks and ventilation reduction do save energy, reliably, and they get disabled when they cause complaints. So the organisational question there is who handles the complaints and whether they have been told what the strategy is, which is cheaper to arrange than any sensor.

The idea to walk away with

Decide which of the three questions you are answering, then pick the cheapest technology whose failure mode you can live with, then decide where the data lands and who acts on it. In that order. Most occupancy projects run it backwards: a technology is selected, a dashboard appears, and the question it answers is worked out afterwards.

The cheapest useful version of this, for most organisations, is to start with data they already hold. Room booking records, access control counts and luminaire-level lighting sensors together give a surprisingly good picture of how a building is used, at no capital cost, and comparing booked against actual use tends to produce an actionable finding within days. Buy sensors to answer the questions that exercise leaves open, which is usually fewer than the original proposal assumed.

Final thoughts

Of everything in this article, the distinction between presence, count and utilisation is the part worth carrying into your next supplier conversation. Suppliers move between the three words freely, because their products usually do one well and the others approximately, and the compound pitch is more appealing than the specific one.

So ask the narrow question. Not "does this give us occupancy data", but "does this give me a headcount per zone accurate enough to modulate a ventilation rate, and what is the evidence for that in a space shaped like mine". The answers separate quickly, and the suppliers who answer precisely are usually the ones worth continuing with.

Related reading: Smart building IoT monitoring, From BMS to building analytics, IoT and BMS integration, Environmental monitoring in warehouses, BMS in HVAC: controls and sequences, Warehouse heat maps.