Change a bearing every six months regardless of how it's wearing, that's preventive. Change it when a vibration signature says it's wearing, that's predictive. Same bearing, same technician, same wrench. The only thing that's different is what triggers the work order, and that one difference carries the entire economic argument.
Why preventive maintenance loses money on both ends
A calendar interval is a guess, set conservatively because the cost of guessing wrong in the other direction, a surprise failure, is usually higher than the cost of servicing early. That conservatism is the problem. Most assets on a preventive schedule get serviced well before they need it, which means paying for parts and labor, and often planned downtime, on equipment that had life left in it. Perth County Ingredients is the case for doing preventive maintenance at all, moving off pure reactive work saved it $40,000 a year, but a calendar schedule still doesn't know the difference between a bearing that's fine and one that's about to seize.
And a calendar still misses the failures that don't respect a schedule. Electrical faults, contamination events, and sudden mechanical failures happen between intervals just as often as on them. Preventive maintenance over-services healthy assets and still gets blindsided by random ones. Both problems have the same root cause: the trigger is time, not condition.
Why predictive maintenance isn't the default answer either
Predictive maintenance fixes both problems by triggering on the actual state of the asset instead of a guess about it. That fix isn't free. It requires sensors on the equipment, which costs money to install and maintain. It requires a model trained on failure history, which means the asset needs to have failed, or come close, often enough to teach the model what failure looks like, a requirement most plants can't meet for most of their equipment. And it requires someone whose job is to act on the alert, which is a workflow most maintenance teams have to build, not just buy.
Fiberon shows what it looks like when that setup pays off: $274,000 saved and 178 hours of downtime avoided in eight months, a run-rate of roughly $411,000 a year, on top of an existing maintenance program. That number is real, and it's also the result of a plant that had the sensor budget, the failure history, and the response workflow already in place. Most plants evaluating predictive maintenance don't yet have all three.
The comparison
| Preventive | Predictive | |
|---|---|---|
| Trigger | Calendar or runtime interval | Measured condition |
| Data required | A rough failure-rate estimate | Sensor coverage plus failure history |
| Setup cost | Low, mostly process discipline | Higher, sensors, integration, a model |
| Risk of over-servicing | High, by design | Low |
| Risk of missing random failures | High, doesn't detect them | Low, if the failure mode is measurable |
| Best-fit assets | Cheap parts, low consequence of failure, no clean signal to measure | Expensive or safety-critical assets with a measurable failure mode |
The honest answer: most plants run both
This isn't a decision you make once for the whole plant. It's an asset-by-asset call, and the rule is straightforward: an asset deserves predictive maintenance when failure is expensive or dangerous, the failure mode is physically measurable, vibration, temperature, current draw, and you have or can build enough history to train a model. Everything else, cheap components, assets with no clean measurable signal, or equipment that fails too rarely to have a usable failure history, stays on preventive, or in some cases stays reactive on purpose because fixing it after it breaks is genuinely cheaper than instrumenting it.
That's why the deployments in our Predictive Maintenance benchmark, 62 of them, cluster on critical rotating equipment rather than spreading evenly across a plant's asset list. That benchmark publishes no downtime median, deliberately: only five deployments report a parseable downtime figure and three are the identical round number 80%, which is a fact about how vendors write case studies rather than about what plants achieve. What the reported savings do support is a median of $200,000 a year across 8 deployments. The plants getting those results didn't replace preventive maintenance, they layered predictive on top of it, on the subset of assets where the economics actually work.
Where to go next
- For the mechanics of how a predictive system actually operates, see what is predictive maintenance.
- Whichever approach an asset is on, the work order and history live in a CMMS, and knowing the difference from an EAM matters once you're managing more than one site, see CMMS vs EAM.
- The full deployment list, filterable by industry, is on the Predictive Maintenance use-case hub.
- Vendors active in this category are ranked with their evidence on the manufacturing operations management software hub.
AI Roadmap — Eight workflows in your plant where AI would pay back this quarter. We know which. 12 pages, in 72 hours.