Connected sensing has become inexpensive enough that deployment is no longer the constraint. Which has produced a recognizable failure pattern: substantial sensor estates generating substantial data that nobody has an established process for acting on.
Where it demonstrably works
Condition-based maintenance. Vibration, temperature and current monitoring on rotating equipment to identify degradation before failure. The most mature application, with the clearest economics, because the alternative — scheduled maintenance or run-to-failure — has a measurable cost.
Energy and utility monitoring. Sub-metering to identify consumption by area, system and time. Frequently pays back quickly, largely because it surfaces equipment running when it should not be.
Environmental monitoring where conditions matter to a product or process — cold chain, clinical storage, controlled environments. Effective partly because the compliance requirement forces someone to own the alerts.
Occupancy and space utilization, used to inform real estate and facilities decisions.
The design questions that determine outcome
What decision does this data inform, and who makes it? If there is no answer, the deployment will produce dashboards nobody opens. This is the question that separates the projects that pay back from the ones that get quietly abandoned.
What is the alerting threshold, and who receives it? An alert with no owner is noise, and after a few weeks of noise the whole system is ignored.
How long does data need to be retained? Condition monitoring needs history to establish baselines. Compliance monitoring may need years. Retention drives storage architecture and is awkward to change retrospectively.
How are these devices secured and maintained? Sensors and gateways are network endpoints with long lifetimes, infrequent updates and often weak default configuration. They belong on segmented networks with a firmware maintenance plan, or they become the route into the environment they were installed to monitor.
A sequencing recommendation
Start with one asset class and one decision. Instrument it, establish the baseline, prove the decision improves, then extend. Estate-wide deployments before the decision process exists generate cost and data in roughly equal measure, and value in neither.