Osmos Global Publication · Osmos Perspective
Occupancy AI Must Respect Metric Boundaries
Attendance, utilisation and demand forecasts answer different questions
JLL reported 56% utilisation in its 2026 occupancy benchmark, while CBRE reported 53% in a separate global study.[3][4] The figures are not directly comparable because portfolios and methods differ.
Analysis
An AI model trained on ambiguous labels will reproduce the ambiguity. Badge attendance, observed workpoint use and bookings cannot be treated as interchangeable ground truth.
Osmos Global analysis: occupancy analytics require a metric contract defining space, time, capacity, population and exclusions.
Leadership implication: model accuracy is meaningless when the target variable is not operationally defined.
Decision lens
Recommended action. Approve the measurement dictionary before approving the occupancy model.
Cite this
Osmos Global Research & Knowledge Centre (2026). Occupancy AI Must Respect Metric Boundaries. Osmos Perspective, Osmos Global. https://www.osmosglobal.org/articles/occupancy-ai-must-respect-metric-boundaries
Keep reading
Smart-Building Pilots Need an Operating Owner
An experiment becomes useful only when someone owns the decision it is meant to improve.
1 Sept 2026 · Osmos Global Research & Knowledge Centre · 5 min read
An Alert Is Not a Maintenance Outcome
Analytics creates value through verified correction, not the number of faults displayed.
1 Sept 2026 · Osmos Global Research & Knowledge Centre · 5 min read
Sensor Coverage Is Not Data Quality
Connected points need identities, context and a known level of trust before they can support decisions.
1 Sept 2026 · Osmos Global Research & Knowledge Centre · 5 min read
Download this paper
The full PDF, formatted for circulation. Downloads are for members, so that we know who our research reaches.
Discussion
Add what you are seeing on the ground.