Every facility-management vendor now claims to be “AI-powered”. Some of it is real, some of it is a chatbot bolted onto a ticket queue. If you run buildings, the useful question isn’t “does it have AI?” — it’s “which decisions get better, and can I audit them?” Here is an honest map of where AI actually earns its keep in facility operations today.
Where AI already works
Smart ticket routing and triage
Classifying a request (“water leak, pantry, L7”) against a service catalogue, picking the right SLA and assigning the best-placed technician or vendor is a pattern-matching problem — exactly what machine learning is good at. Done well, it removes the dispatcher bottleneck and cuts first-response time dramatically. The tell of a real implementation: the system shows why it routed a ticket, and a human can override it.
Anomaly detection on building telemetry
If your BMS, meters and sensors feed a platform, baselines per asset per season are easy to learn. A chiller drawing 11% above its seasonal baseline is invisible in a monthly bill but obvious to a model watching intervals. The output should be a drafted work order with the evidence attached — not a dashboard nobody opens.
Predictive and condition-based maintenance
Runtime hours, vibration trends and delta-T degradation predict failure earlier than a quarterly calendar. You don’t need exotic sensors to start: even DG-set runtime and energy-draw patterns move you from “every 90 days” to “when the asset says so”. Expect fewer emergency call-outs, not zero.
Drafting the paperwork
Shift reports, board packs, invoice narratives, RCA summaries — assembling them from structured operational data is exactly the drudgery language models should absorb. The rule: AI drafts, humans approve, and the underlying numbers come from the system of record, not the model’s imagination.
Where to stay skeptical
- Fully autonomous operations — nobody serious lets a model restart a fire pump. Look for human-in-the-loop approvals.
- Chatbots as the headline feature — a conversational front-end on a bad workflow is still a bad workflow.
- Black-box scoring — if a vendor can’t explain why the model flagged an asset, you can’t defend it in an audit.
- AI that needs perfect data first — good platforms improve your data as a side-effect of daily work; they don’t demand a two-year cleanup project.
Questions to ask any vendor (including us)
Which decisions does the AI touch, and what happens when it’s wrong? Is every suggestion logged with its evidence? Can we switch a capability off? Where does our data go, and is it used to train models shared with other customers? A vendor comfortable with those questions is selling software; one who isn’t is selling a demo.
Facyliti applies AI where it demonstrably pays — routing, anomaly detection, predictive maintenance and drafted reporting — with every suggestion logged, explainable and approved by your team. See it live.
