Occupancy sensors vs calendar data: which tells the

Room 3 shows Busy on the panel outside the door. Through the glass, the room is empty, and it has been empty since the meeting started 25 minutes ago. Somebody booked it on Monday, took the call from their desk instead, and never touched the booking again. Two people are standing in the corridor with a laptop looking for somewhere to talk, and the system is confidently telling them there is nowhere to go.

That gap, between what the calendar says and what is actually happening in the room, is the entire argument for occupancy sensors. This post covers what calendar data genuinely knows, what each sensor type adds and at what accuracy and privacy cost, what installation actually involves, and the cheap middle path that gets most 20 to 100 person offices most of the way there. For the record we build calendar-driven room displays and sell no sensors at all, so this is not an argument for buying either.

What calendar data actually knows

Your room mailboxes in Microsoft 365 or your calendar resources in Google Workspace hold a genuinely rich dataset, and most offices never look at it. From bookings alone you can answer:

Which rooms are booked, how often, and in which time bands.
How long the average booking is, and how that compares to room capacity.
How far in advance people book, which is a good proxy for whether booking feels easy or annoying.
How much of the week is consumed by recurring series that were set up two years ago.
Which rooms nobody ever picks, which is usually a naming or equipment problem rather than a demand problem.

That is a lot. It is free, it needs no hardware, and in Google Workspace the admin console will even report on resource use directly. If you have never pulled a per-room booking report, do that before you consider spending money on hardware. Most offices find at least one obvious fix in the first hour.

What calendar data cannot know

One thing, and it is the important thing: whether anyone was in the room. A calendar records intent. It does not record attendance. That produces two errors that push in opposite directions:

Booked but empty. The meeting moved to a call, or got cancelled verbally, or three of the four people are remote today. The calendar still counts the room as used.
Used but not booked. Two people grab a visibly free room for a ten minute conversation. Real demand, zero data.

The first inflates your utilization figures. The second hides demand. If you are deciding whether to knock through a wall and build another room, both errors matter, and they do not cancel out reliably.

What sensors add, type by type

"Occupancy sensor" is not one thing. The technologies differ enormously in what they can detect, how they fail, and how much of a privacy conversation they trigger.

Passive infrared (PIR)

PIR detects changes in infrared radiation, which in practice means it detects movement of a warm body. Cheap, low power, battery friendly, and by far the most common sensor in commercial buildings because it is what lighting controls already use.

Its failure mode is well documented: a person sitting still in a chair listening to a presentation may stop registering. Academic work on inferring occupancy from a single PIR sensor exists precisely because raw PIR output is not a headcount. PIR tells you "something moved recently" and nothing more. It cannot count people, and it cannot distinguish a person from a fan pointed at the sensor.

Ultrasonic and dual technology

Ultrasonic sensors emit high frequency sound and measure the Doppler shift in the reflection, which makes them better at detecting small movements (typing, breathing at close range) than PIR. Vendors commonly pair the two, because a dual technology sensor can require PIR for the initial trigger and accept ultrasonic to hold the reading, which cuts false negatives on a seated person and false positives from HVAC airflow.

Still no headcount. Still no distinction between one person and eight.

Radar and mmWave presence sensors

The current generation of presence sensors uses millimetre wave radar, which can detect micro movement including chest movement from breathing. Vendors quote high accuracy figures for meeting rooms. Verify those figures with a pilot in your own rooms before you rely on them, because accuracy depends heavily on mounting height, room shape and how the vendor defines a correct reading, and published figures are usually measured under conditions the vendor chose.

Thermal array sensors

A low resolution thermal array (often 8x8 or 32x24 pixels) sees heat blobs, not faces. This is the sweet spot many workplace vendors have settled on for people counting, because you can get a reasonable headcount without capturing an identifiable image. It is the option that survives a privacy review most easily among the counting technologies.

Camera based people counting

The most accurate for headcount, and the highest risk category by a wide margin. Vendors will tell you the image is processed at the edge and discarded, and for reputable vendors that is true. It does not remove the need for a data protection impact assessment, and it does not remove the conversation with your works council if you have one. Putting a camera in a meeting room in Germany or the Netherlands without that conversation is a good way to get the whole project shut down.

Desk pucks and door contacts

Under desk pucks are a desk booking tool, not a meeting room tool, and they are worth mentioning only so you do not buy the wrong thing. Door contacts (a simple magnetic reed switch) are the cheapest signal in the building: they tell you the door opened, which for a small phone booth or focus pod is often enough to drive an occupied light.

The accuracy and privacy trade-off

PIR. Detects movement. No headcount. Main failure: misses a person sitting still. Privacy risk: low.
Ultrasonic and dual technology. Detects small movement. No headcount. Main failure: airflow false positives. Privacy risk: low.
mmWave radar. Detects presence and micro movement, rough headcount at best. Main failure: sensitive to mounting height and room shape. Privacy risk: low to moderate.
Thermal array. Detects heat signatures, gives an approximate headcount. Main failure: warm equipment and tightly packed seating. Privacy risk: moderate, usually defensible because no identifiable image is captured.
Camera and computer vision. Detects people, accurate headcount. Main failure: lighting and occlusion. Privacy risk: high. Expect a data protection impact assessment and, in Germany or the Netherlands, a works council conversation.
Door contact. Detects door state only. No headcount. Main failure: doors propped open. Privacy risk: very low.

What it costs to actually install

Here is where most articles invent numbers. Almost no workplace sensor vendor publishes list pricing, and the ones that do publish hardware prices do not publish the platform subscription that makes the hardware useful. So the honest answer is: get a quote, and check current pricing for anything you read anywhere, including here.

What you can plan for is the shape of the cost, which is usually four parts:

Sensor hardware, per room, sometimes more than one per room for larger spaces.
A gateway or bridge if the sensors use a low power radio (LoRaWAN, Zigbee, BLE mesh) rather than WiFi. One or more per floor.
A per room or per sensor software subscription, billed annually. This is frequently the largest line over three years.
Installation. For ceiling mounted sensors above a suspended ceiling that means someone on a ladder and a facilities ticket. For anything mains powered it means an electrician.

Add the soft cost that nobody quotes: battery replacement. A battery powered sensor with a quoted three to five year battery life will start dropping off your dashboard in year three, and somebody has to notice and go fix it. In a three site company that somebody is usually an office manager who did not sign up for this.

The number everyone actually wants

Strip away the dashboards and there is one figure that drives real decisions: what percentage of booked room time had nobody in it. Call it the ghost rate, the booked-but-empty gap, or the booking-to-occupancy ratio. It decides whether you have a capacity problem or a hygiene problem, and those two problems have wildly different price tags.

Vendor benchmarks vary and should be read with care, because the vendors publishing them sell the sensors that produce them. Worklytics, for example, publishes 2025 benchmark tables putting occupancy at roughly 69 to 78 percent of booked time across company sizes, which implies a gap somewhere around a fifth to a quarter of booked time. Take the direction seriously and the decimal places not at all.

The practical rule: if your booked-but-empty gap is meaningfully above 20 percent, adding rooms is the wrong move. You are not short of space. You are short of a mechanism that hands unused space back.

The honest middle path: check-in and auto-release

Sensors are one way to detect that a room is not being used. A person confirming they turned up is another, and it costs almost nothing.

A check-in mechanism works like this: the booking exists, and at start time the display asks somebody to confirm. If nobody confirms within 10 to 15 minutes, the booking is released and the room shows free again. You now have, for every booking, a yes-or-no attendance record. That is not as rich as sensor data, but it answers the one question calendar data could not.

What you give up compared to sensors:

No headcount, so you cannot see that your 12 person boardroom routinely holds two people.
No detection of unbooked use, so pirate occupancy stays invisible.
It depends on people actually tapping, which means it needs a policy and a couple of reminders in the first fortnight.

What you get: the booked-but-empty number, rooms handed back automatically, and no hardware on the ceiling. Both Microsoft and Google ship room release mechanisms of their own, and several display vendors implement check-in on the panel.

A decision list

Pull your calendar data first. It is free and it will answer half your questions.
Add check-in and auto-release. Run it for a month. Now you have a real ghost rate.
If the ghost rate falls and contention disappears, stop. You had a hygiene problem and you fixed it.
If contention persists and you are about to make a real estate decision worth six figures, sensors are worth their cost, because you need headcount and unbooked-use data that check-in cannot give you.
Choose the least invasive sensor that answers your actual question. If you only need occupied or not occupied, do not buy a camera.

The buy side of this decision

Since this post is on our blog, the disclosure. Lobby is a meeting room display that syncs in real time with Google Workspace and Microsoft 365 room resources. It shows what is booked, lets people book from the door, and gives you per room booking data. It runs on a 7.5-inch e-ink panel from USD 69 per room that you buy once direct from the manufacturer and own (less at volume, USD 62.99 each at 10 or more), or in any browser tab if you already have a tablet or TV. Pricing is on our pricing page: free forever for up to three displays with 15 minute e-ink refresh and no credit card, then USD 30 per month billed yearly (USD 360 a year, or USD 50 month to month) for unlimited displays, and USD 60 per month billed yearly (USD 720 a year, or USD 100 month to month) for Pro Unlimited with 5 minute refresh, EU hosting for e-ink devices, custom branding, QR room feedback and a Slack /book command. Yearly billing saves 40% against the month-to-month rate. Setup is under 10 minutes, self-serve, no sales call.

What it does not do, plainly: Lobby has no occupancy sensors and does not read anyone else's. No PIR, no radar, no cameras, no headcount. It also does not do desk booking, visitor management, or room and AV automation, there is no enterprise SSO or SCIM at scale, no MDM on the e-ink hardware, and no on-premises Exchange without a hybrid configuration. If your question is "how many people were physically in room 4 last Tuesday", Lobby cannot answer it and no amount of configuration will change that. If your question is "is this room free right now and who has it next", that is the job. And if you want a booked-versus-turned-up signal without a ceiling install, check-in on a door display is the cheap version of the sensor answer.

TL;DR

Calendar data is free, rich and already in your tenant. Pull it before buying anything.
Calendar data records intent, not attendance. It cannot tell you whether the room was occupied.
PIR detects movement and misses still people. Ultrasonic and dual technology fix some of that. Radar is the current default for presence. Thermal arrays give approximate headcount without identifiable images. Cameras are the most accurate and by far the highest privacy risk.
Sensor cost is hardware plus gateway plus subscription plus installation, and vendors rarely publish list prices. Get a quote and verify current pricing.
The number that drives decisions is the booked-but-empty gap. Above roughly 20 percent, you have a hygiene problem, not a capacity problem.
Check-in plus auto-release gets you that number for near zero cost. It will not give you headcount or unbooked use.
Buy sensors when a large real estate decision genuinely depends on data check-in cannot produce. Otherwise start cheap.

Sources

Vendor benchmark and sensor specification figures were verified 5 August 2026. Sensor pricing is almost always quote-based, so confirm current figures directly with each vendor.

Track room and resource use in Google Calendar, Google Workspace Admin Help.
Set-CalendarProcessing, Microsoft Learn (room mailbox booking behaviour).
Types of occupancy sensors and applications, Milesight (sensor technology overview, vendor source).
The complete guide to occupancy sensors for your workplace, VergeSense (vendor source).
Predicting room occupancy with a single passive infrared (PIR) sensor through behavior extraction, ACM UbiComp 2016.
2025 benchmarks for meeting room utilization, Worklytics (vendor benchmark).
Seeed reTerminal E1001 product page, for the USD 69 unit price.
Guidelines 4/2019 on Article 25 data protection by design and by default, EDPB.

Occupancy Sensors vs Calendar Data: Which One Tells You the Truth About Your Rooms