"Predictive maintenance" is the line item every IoT vendor wants on your budget, and the one most likely to disappoint — not because the technology doesn't work, but because it gets sold as a building-wide religion instead of an asset-specific tool. Bolt sensors onto everything, the pitch goes, and failures vanish. In practice the ROI lives in a narrow band: a handful of assets where the cost of unplanned failure is high and the failure mode is detectable. Everywhere else you're paying to monitor things that were fine on a clipboard.
Here is the honest version for commercial buildings.
The economics that actually decide it
Predictive maintenance pays when three things line up on the same asset:
- High cost of unplanned failure. A chiller or main air handler going down mid-summer costs comfort, tenant goodwill, and emergency-rate labor. A $200 exhaust fan does not.
- A detectable, gradual failure mode. Vibration, temperature drift, current draw, or fault-code patterns that trend before they break. Random electronic failures give you nothing to predict.
- Enough runtime data to learn "normal." Without a baseline, no algorithm can flag the abnormal.
Miss any one and "predictive" collapses back into "expensive condition monitoring."
Where it wins in a commercial building
- Central plant — chillers, boilers, cooling towers, primary pumps. High failure cost plus rich sensor data is the strongest case in the building.
- Major air handlers / RTUs — comfort-critical, with detectable bearing, belt, and coil degradation.
- Critical-environment assets — data halls, labs, healthcare. Failure cost is extreme, so even modest prediction accuracy pays for itself.
Where it usually doesn't
Terminal units, small fans, plumbing fixtures, lighting. Low failure cost, cheap to run-to-failure or handle on a simple preventive schedule. Instrumenting these is exactly how predictive-maintenance programs blow their budget — and their credibility.
The number to run before you buy
Predictive-maintenance ROI is a function of avoided-failure cost, not sensor count. The same payback logic that governs an AI-HVAC retrofit applies here: implementation cost ÷ (avoided emergency repair + avoided downtime + deferred replacement). If a vendor can't tie the program to specific high-value assets and a defensible avoided-cost number, the ROI is a slide, not a forecast.
→ Run the asset-level payback in the free AI-HVAC ROI calculator and start your program where the math is strongest — not where the sensor catalog is biggest.
Research compiled by the AISB agent fleet from primary sources; every claim verified against the public record. Cost figures are labeled industry estimates. Full source list available on request — hello@ai-smart-buildings.com.