Most articles about IoT-enabled maintenance focus on the win: a sensor catches a failure early, downtime gets avoided, everyone’s happy. Fewer people write about what a plant looks like eighteen months in, after the dashboards have filled up and technicians have quietly started ignoring half the alerts on them.
It’s worth talking about, because it’s common, and because it’s avoidable.
The Pilot Worked. Now What?
Early in a predictive maintenance rollout, sensor data usually earns its keep fast. A vibration sensor flags a bearing on its way out, a tech checks it, the catch is real. Confidence goes up. That early win is usually the thing that gets a pilot expanded to the rest of the plant.
The trouble starts during the expansion. More assets get instrumented, and alert volume climbs faster than anyone’s actual capacity to chase each one down. Thresholds set conservatively during commissioning — often by an integrator who’s never run that specific line on a Tuesday afternoon with the operator pushing the changeover faster than spec — start firing for conditions that are just normal noise for that asset. A pump that always runs a little rough on startup throws the same alert as a pump with a bearing actually failing.
The first few false alarms, a tech checks out properly. By the tenth one on the same asset, they’ve stopped believing the alert. By the thirtieth, they’ve muted that line on the dashboard entirely — and, often, they’ve started side-eyeing alerts from other assets too, because it’s the same system that cried wolf on the pump that’s now flagging the gearbox.
That’s alarm fatigue. It’s well documented in aviation and healthcare, and it shows up in industrial maintenance for the same reason: when false positives don’t get filtered out by the system, the human filters them out instead, and humans aren’t precise about it. Trust doesn’t erode asset by asset. It erodes across the board.
Less Instinct, Nothing to Replace It
A plant running zero condition monitoring still has technicians who trust their own senses — the sound a bearing makes right before it goes, the smell of insulation overheating, a panel that’s running hotter than it should. A plant with poorly tuned sensors can end up somewhere worse: techs who’ve partly handed that judgment over to a system, and then learned the system can’t be trusted either. Less instinct, and nothing solid to replace it with.
The Danger of “Phantom Compliance”
There’s a quieter problem underneath that one, and it’s the part that doesn’t show up on a leadership dashboard. We’ve heard a version of this more than once from manufacturing teams evaluating maintenance software, enough that it sounds like a pattern rather than a one-off: a maintenance manager describing their team as having a strong PM compliance number on paper — alerts acknowledged, work orders closed — while breakdowns kept happening on the same equipment anyway. When you dig into what “closed” actually meant, it often turned out technicians were marking alerts resolved without a real inspection, because clearing the queue before end of shift was the path of least resistance, and explaining to a supervisor why an alert “felt like noise” wasn’t worth the conversation. The paperwork looked healthy. The floor told a different story.
That gap is the real risk. Leadership reads compliance rate as a proxy for asset health, when in some plants it’s actually a proxy for how good technicians have gotten at clearing their screens.
Tuning Isn’t a Setup Step. It’s a Job.
The fix sounds simple — tune the thresholds — but the reason it doesn’t happen is more about ownership than math. Tuning usually gets treated as a one-time commissioning task, owned by whoever installed the sensors. Once the integrator packs up and leaves, nobody owns revisiting thresholds as shift patterns, product mixes, or asset age change underneath them.
The plants that hang onto trust in their monitoring programs tend to do a few things on purpose. They treat the first three to six months after any sensor goes live as a tuning window, not a results window, and they say so out loud to the team — alerts will be noisy at first, here’s how long that’s expected to last. They give technicians a fast way to flag an alert as a false positive without a long form to fill out, and somebody actually reviews those flags every week and adjusts thresholds instead of letting them pile up unread. They also track something most dashboards don’t surface by default: how many alerts on a given asset turned out to be real versus noise. A line where almost every alert turns out to be nothing isn’t a technician problem. It’s a tuning problem, and the technician is the one quietly absorbing the cost of it every shift.
What Actually Matters Before You Buy Anything
Most procurement checklists for condition monitoring focus heavily on hardware accuracy and integration breadth. Fewer buyers look at the human workflow. They don’t ask if a technician on the floor can challenge a bad alert in two taps, or if the platform can flag management automatically when an asset’s alert pattern looks more like a tuning error than a mechanical failure. It’s not a feature that demos well, but managing the human element of the data is what determines whether the investment actually pays off.
It’s usually the difference between a connected maintenance program that gets more valuable every year and one that turns into a screen full of notifications nobody reads anymore. The sensors were never the hard part. Keeping people’s trust in what the sensors say is.
