A sensor shows presence, not value

The gap this collection is about

Occupancy analytics counts presence in space over time. Everything else — utilisation, demand, whether a floor is worth keeping — is inferred, and the inference is where the decisions actually get made.

The technical evidence in “A sensor shows presence, not value” describes the building, while the effort required to install, test and maintain it is project work. When teams research employee monitoring software, employee monitoring software can provide time and project context for that operational effort without replacing the sensor platform as the source of truth for physical occupancy.

What is directly observed is narrow: something was detected in a zone at a time, for how long, and sometimes how many. What gets inferred is everything else. That the detection was a person rather than a cleaning trolley. That one detection is one person rather than somebody carrying three devices. That a desk with a bag on it is in use, or is not. And above all, that absence means not needed, which is the largest leap and the one driving the most expensive decisions.

For a public, independent reference related to “A sensor shows presence, not value”, consult the ISO standards catalogue. Its principles provide a useful check on scope, terminology, governance and the claims made during procurement or review.

Presence is not value

A room used twice a week looks wasteful on a utilisation report. Whether it is depends on what happens in it, and no sensor can see that.

Low occupancy has at least six plausible readings: the space is genuinely unnecessary, it is badly designed, it is hard to book, nobody knows it exists, it is used for something invisible to the sensor, or it is used rarely and the rare use matters enormously.

High numbers are just as ambiguous and get far less scrutiny, because they confirm what facilities teams want to hear. A crowded floor can mean genuine demand, or it can mean people have nowhere else to go.

Three numbers that get confused

Occupancy is how many are present. Utilisation is how much capacity was used over a period. Capacity itself has three competing definitions — design, comfortable and fire safety — and which one you use changes every ratio downstream.

A room at four people out of twelve is fully occupied for booking purposes and a third utilised, and both statements are true.

Then there is the denominator. Working hours or all hours, weekdays or all days, bookable hours or total. Switching from a round-the-clock basis to working hours roughly triples every figure without anything changing in the building — which is why benchmark comparisons between organisations are close to meaningless unless the definitions travel with them, and they rarely do.

Peak and average drive opposite decisions

A building at 45% average occupancy and 92% peak is two different buildings depending on which number reaches the decision. Average drives "we have too much space"; peak drives "we cannot fit everybody". Both are usually true at once, and a building sized to the average fails on its busiest day.

Never publish one without the other. A report quoting average alone will be used to justify reduction, and nobody will ask about the busiest day until after the decision is made.

Calibration decides whether any of it is real

Sensors produce numbers immediately. Whether those numbers mean anything is settled in the first few weeks, and that work cannot be done retrospectively: a sensor counting a walkway for six months has produced six months of inflated figures that nobody can un-inflate.

The work is unglamorous. Count people physically in your busiest spaces, at varied times, over a week. Compare against the sensor for the same minutes. Set timeouts from what you observed rather than from the supplier default. Expect to move a quarter of the devices. Establish your error band and state it whenever figures are used afterwards.

And set an alert for any device reporting nothing, because a dead sensor reports zero — which looks like an empty desk rather than a fault.

Which method answers which question

Six ways to count, each with different blind spots, and the choice should follow the question rather than the supplier's catalogue.

Badge and access data you already hold. It answers building-level headcount and nothing about specific spaces. It counts cards rather than people, most systems record entries without exits, and two people through one door on one swipe is common enough to matter.

Booking systems record intention, not presence. The comparison between booked hours and sensed hours is the single most useful figure the two sources produce together, because the no-show rate is almost always larger than anybody believed.

Infrared detects movement, not presence. Somebody reading or on a call stops registering once the timeout passes. Cheap, adequate for binary desk-level questions, and incapable of counting.

Thermal and depth sensors count people without producing an identifiable image, which is why they are chosen for rooms and entrances. Verify what the device actually transmits rather than accepting the datasheet.

Device counting is free because the access points are already there, and unreliable because people carry different numbers of devices, laptops persist after their owners leave, and modern devices randomise their identifiers.

Cameras are the most capable and carry the highest cost in trust. Several data protection regimes require you to consider whether a less intrusive method would do, and for most occupancy questions one would.

Low utilisation usually means the wrong space, not too much

This is the finding that most often survives proper analysis, and it is almost never the first proposal.

The data rarely shows uniform emptiness. It shows patterns: large rooms used by small groups, open desks empty while quiet space is contested, one floor busy and another ignored. That is a mismatch picture, not an excess picture, and the response is different.

Subdividing a twelve-person room that hosts meetings of three typically raises measured utilisation sharply and resolves the shortage people complain about. Most buildings are short of somewhere to take a call, somewhere genuinely quiet, and somewhere for two people to talk briefly — all cheap to create, and all absorbing demand that is currently hitting meeting rooms.

Space given up cannot usually be taken back, certainly not at the same price. The asymmetry is severe: keeping too much costs rent, releasing too much costs the ability to operate. Redesign first, measure the result, and reduction remains available afterwards better informed. The reverse order is not available.

Where it stops being about space

Occupancy analytics and workplace monitoring use overlapping technology for different purposes. The distinction is not technical: it is which question is asked. "How is this space used" produces an aggregate count about a place. "Where is this person" produces an answer about an individual. The same sensor does both.

Programmes cross that line gradually and by request rather than by design. A manager asks whether their team is coming in. HR asks during a dispute. Security asks who was on the floor. Each request is individually reasonable and each one granted changes what the system is.

What holds the line is partly procedural — a written purpose, a reporting floor, a named person who refuses — and partly technical. Not retaining the desk-to-person mapping makes individual attribution impossible rather than merely prohibited, which is a much stronger position.

It also protects the data. Once people know occupancy figures affect them, behaviour adapts: badge in and leave, sit at the sensed desk, book the room and not use it. The readings stop describing the building and start describing the incentive.

What the core notes deliberately avoid

Product rankings inside the core notes. Named products appear only in separate comparison guides, so the method and governance advice remains independent of a particular supplier.

No accuracy percentages. The figures in circulation were measured under favourable conditions, in a different building, by the party selling the device. What matters is how you verify accuracy in yours, which is a week of counting.

And no claim that measurement removes judgement. Every reading needs interpretation, and the interpretation is where the decision lives.