The pitch for AI predictive maintenance is "fix it before it breaks." The reality is more specific — and it depends almost entirely on one thing you can check before you sign anything.
AI predictive maintenance for commercial buildings uses live equipment and sensor data to forecast a developing fault before it causes a failure, so a chiller, pump, or air handler is serviced on condition rather than on a fixed calendar or after it has already broken down. It works only when your equipment already exposes the data the model needs to read.
Three maintenance strategies, and where AI fits
Predictive maintenance is the third of three long-established strategies, and it helps to name them before comparing vendors:
- Run-to-failure — you fix equipment after it breaks. Cheapest to plan, most expensive when a failure hits a tenant-critical system at the wrong time.
- Preventive (calendar-based) — you service on a fixed schedule regardless of condition. Predictable, but it both over-services healthy equipment and can still miss a fault that develops between intervals.
- Predictive (condition-based) — you service based on the equipment's actual measured condition. This is the approach formalized in reliability engineering as condition monitoring (ISO 17359) and reliability-centered maintenance (SAE JA1011).
"AI" predictive maintenance is condition-based maintenance where the condition signal is inferred by a model watching many data points at once, rather than by a technician reading a single gauge on a walkthrough.
What the AI layer actually does
In a commercial building, the model usually sits on top of the data your building automation system (BMS) and any added IoT sensors already produce — supply-air temperatures, pump pressures, motor current, vibration, runtime hours. It looks for the patterns that precede a known failure mode: a fan drawing more current than its operating history suggests, a chiller approach temperature drifting, a valve hunting.
This is the same family of analytics as automated fault detection and diagnostics (FDD), which ASHRAE has developed standardized methods for and which the U.S. Department of Energy and its national laboratories have studied in commercial buildings as a way to surface faults that would otherwise go unnoticed. The honest framing is that the model flags a probable developing fault and prioritizes it — it does not, by itself, guarantee a specific downtime or cost outcome, which depends on how your team acts on the alert.
The prerequisite that decides everything: are your points open?
A predictive model cannot predict on data it cannot see. Before evaluating any vendor's accuracy claims, the deciding question is whether your equipment exposes its data:
- Open and well-tagged — your BMS speaks an open protocol such as BACnet (ASHRAE Standard 135) or Modbus, and points are accessible. The analytics can start.
- Open but messy — the protocol is there but points are unlabelled. Workable, but budget for a normalization pass (open semantic models like Project Haystack and Brick Schema exist for exactly this) before any prediction is meaningful.
- Closed or proprietary — a locked-down system that does not expose points. Here the integration starts one step earlier, at the controls layer, not the AI.
A vendor demonstrating impressive predictions on their reference building tells you little until you confirm the same data is available on yours.
Schedule a work order, or act on the equipment? Two risk tiers
"Predictive maintenance" hides an important line. A model that raises a prioritized work order for a human to review is low-risk and the usual starting point. A model that automatically changes equipment behavior — throttling, rescheduling, resetting setpoints in a closed loop — is a different commitment. Before you enable any automated action, get two things in writing: who owns the consequence when an automated decision is wrong, and what the defined fail-safe is when the system is unavailable.
Frequently asked questions
What is AI predictive maintenance for commercial buildings? It is condition-based maintenance where a model watches live equipment and sensor data — from the BMS and any added IoT sensors — to forecast a developing fault before it causes a failure, so equipment is serviced on its actual condition rather than on a fixed calendar or after a breakdown.
How is predictive maintenance different from preventive maintenance? Preventive maintenance services equipment on a fixed schedule regardless of condition; predictive maintenance services it based on its measured condition. Preventive can over-service healthy equipment and still miss faults that develop between intervals, while predictive targets the equipment that is actually showing a problem.
What does my building need for AI predictive maintenance to work? The equipment must expose its data points over an open protocol — most commonly BACnet (ASHRAE Standard 135) or Modbus — or a vendor API. If points are open and well-tagged, analytics can begin; if the system is closed or proprietary, you may need a protocol gateway or integrator to open a data path first.
Sources
- ISO 17359, Condition monitoring and diagnostics of machines — General guidelines — the standard framing of condition-based monitoring referenced above.
- SAE JA1011, Evaluation Criteria for Reliability-Centered Maintenance (RCM) Processes — the reliability-centered maintenance framework referenced above.
- ASHRAE — standardized methods for fault detection and diagnostics (FDD) in building systems; ASHRAE Standard 135 (BACnet) for open data communication.
- U.S. Department of Energy and national-laboratory research on automated fault detection and diagnostics in commercial buildings, referenced above as a driver, not a quantified guarantee.
- Project Haystack and Brick Schema — open semantic data models for tagging and normalizing building equipment data.
Want to pressure-test a predictive-maintenance pitch against your own building before you commit? Start with the AISB tools at tools.ai-smart-buildings.com.
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.