‍
In CMMS-integrated Facilities Management operations, human-in-the-loop means a trained ROC (Remote Operations Centre) reviewer holds supervisory authority over an autonomous FM agent, intervening only when the system flags a fault as low-confidence, high-risk, or outside its authorised operating scope. The agent resolves the remaining cases independently within defined operating boundaries.

For asset owners and compliance-driven FM teams, the important question is not whether a human is in the loop. It is what the human is in the loop for.

A reviewer who signs off on every fault becomes a bottleneck. A reviewer who governs exceptions becomes an oversight function. That distinction determines whether an AI system is scalable, auditable, and operationally safe inside a CMMS or CAFM environment.

In Xempla's Autonomous Maintenance program, one year of operational data across 25,000 assets — including healthcare estates and distributed renewable infrastructure shows the supervisory model operating at roughly a 1-in-7 escalation rate. Around 14% of faults require human review; the remaining 86% are resolved autonomously by the Omi autonomous maintenance agent within authorised operating boundaries.

Those boundaries are explicit. They define which asset classes, fault categories, confidence thresholds, and compliance scenarios the agent is authorised to act on without human intervention. When a fault falls outside those conditions, the system does not trigger a generic escalation. Instead, it surfaces the specific reason for uncertainty to the ROC reviewer, who makes the operational decision.

This distinction matters because many early FM AI deployments misunderstood human-in-the-loop as continuous human approval. In practice, that model collapses under operational volume. Reviewers begin approving actions at the speed of the queue rather than the speed of judgement. The result is slower response times, increased operator fatigue, and limited operational learning for the system itself.

That structure is what creates defensible governance under SLA review, compliance audit, and internal risk assessment. Assurance comes not from the number of human approvals, but from the traceability of where operational authority sits at each decision point.

As FM AI systems mature, the role of the human operator does not disappear. It becomes narrower and more accountable focused on exceptions, operational risk, and governance oversight.
‍

FAQ

What is human-in-the-loop in FM AI systems?

Human-in-the-loop in FM AI systems places a trained ROC reviewer in a supervisory role over an autonomous FM agent. The agent operates independently within predefined boundaries for asset class, fault type, confidence threshold, and compliance category, escalating only the faults that fall outside them. The reviewer governs exceptions rather than approving routine maintenance actions.