The short answer
Two categories pay reliably: leak detection, because the avoided damage is large and immediate, and occupancy, because it changes how cleaning and energy are allocated.
The rest depend entirely on whether anyone acts on what the sensors say, which is an operational question rather than a technical one.
Where the return is clearest
| Sensor | Return | Confidence |
|---|---|---|
| Water leak detection | Avoided damage and downtime | High |
| Occupancy and footfall | Cleaning and energy allocation | High |
| Temperature in critical areas | Avoided spoilage or complaint | High where it matters |
| Dispenser levels | Fewer complaints, fewer wasted visits | Moderate |
| Air quality | Wellbeing, sometimes compliance | Depends on the obligation |
| Asset tracking | Less time searching | Depends on how much is lost |
The confidence column matters more than the return column. A high return that depends on behaviour change is less certain than a modest one that does not.
The costs that decide it
- Battery replacement across hundreds of devices, on a cycle
- Gateway coverage, which older buildings make awkward
- Connectivity subscriptions per device or per gateway
- Integration into the system people already use
- Device failure, and detecting it
- Someone whose job includes acting on the output
The last item is the one that determines whether the deployment still works in two years. Sensors with no owner become a dashboard nobody opens.
Start with a real pilot
- One building, one sensor type, one clear question.
- Record the baseline before fitting anything.
- Run with the existing process unchanged for a period.
- Compare what the sensors say against what you assumed.
- Change the process, then measure again.
Running unchanged first is what tells you whether the sensors reveal anything you did not already know. Skipping it means you cannot separate the sensor effect from the process change.
Avoid the dashboard trap
The common failure is a well-built dashboard that nobody opens after the first fortnight. Data that arrives as a task in someone's existing workflow gets acted on; data that requires opening another system does not.
Integrate into whatever your supervisors already use, even if the integration is cruder than a purpose-built view.