AI-Powered Predictive Maintenance for Elevators: Hype vs. Reality in 2026
"AI-powered predictive maintenance" has become the single most-searched phrase in vertical transportation this year — every major OEM now has a dashboard, an app and a press release about it. Building owners ask us almost weekly whether it's worth paying for. The honest answer: parts of it are genuinely useful, and parts of it are a sensor feed with a machine-learning label stuck on top.
What the technology actually does well
Vibration, current-draw and door-cycle sensors feeding a trained model are genuinely good at spotting trend deviations — a bearing that's slowly drifting out of tolerance, a door operator drawing more current than its baseline, a brake that's taking marginally longer to set. Caught early, these are inexpensive fixes. Caught late, they're breakdowns.
Where the hype outruns the substance
- Most "AI" failure predictions are still threshold alerts on raw sensor data, not genuine predictive models trained on failure history.
- A system is only as good as its installed base — a small population of elevators gives any model very little to learn from.
- Vendor dashboards often report "health scores" with no visibility into the underlying calculation, which makes them hard to act on independently.
Our view: predictive monitoring is a genuinely valuable layer on top of good maintenance practice — not a replacement for it. We evaluate the underlying sensor set and data history before recommending any platform to a client, rather than taking a vendor's health score at face value.
What to ask before you buy in
Ask any vendor pitching predictive maintenance how many of your building type are in their training data, what specific failure modes the model detects (not just "anomalies"), and whether the raw sensor data is exportable if you switch AMC providers. If they can't answer clearly, you're paying for a dashboard, not a prediction.
Want an independent read on how this applies to your building?
Book a Consultation