Room by room: measuring a hotel that measures back
This article describes an illustrative hotel energy pilot pattern: a room-by-room digital twin fed by metering and BACnet data. The public Smart City Labs twin is a sample model, not evidence of a live customer deployment or measured result.
Most building-energy programs begin with a model. An engineer feeds floor area, glazing, occupancy assumptions, and climate data into a simulation, and out comes an estimate of where the waste should be. The estimate is often decent. It is also unfalsifiable at the resolution that matters, because when the utility bill disagrees with the model, nobody can say which room, which hour, or which piece of equipment is responsible for the gap.
The Smart City Labs pilot inverts the starting point. Instead of modeling the hotel and checking against a monthly bill, it meters the hotel and lets the building describe itself: consumption per room, continuously, as it happens. The model comes second, built from measurements rather than in place of them.
A twin that should be fed, not merely drawn
The reference pattern earns fidelity from per-room meters and building-system data, with timestamp and provenance attached to each governed reading. The public twin demonstrates the interface with illustrative data; it is not a live property telemetry feed.
In a real deployment, measured values, modeled values, stale signals, and missing coverage must be visibly distinct. Provenance helps a reviewer trace a reading to its source; it does not establish sensor calibration or completeness on its own.
Measure before claiming waste
With validated per-room measurement, teams can compare occupancy, schedules, commands, and meter readings. Turning those differences into a waste or savings claim still requires a baseline, weather and occupancy normalization, coverage checks, and an agreed measurement protocol.
No consumption or savings result is claimed here. The public example describes an evaluation shape a real pilot could use.
Acting on a building is different
Sensing a hotel is one thing. Acting on one is another. A managed pilot should begin act-only-with-approval, with bounded setpoint authority, verified read-back where possible, duplicate controls, and an explicit ambiguous-outcome path. These are proposed safeguards for a real deployment, not a claim that the public sample is actuating HVAC.
Approval-gated actuation is not a training-wheels phase to outgrow quickly. A real pilot should link proposal, approval, attempted action, observed controller response, and later measurements. Any expansion of authority should depend on deployment evidence and safety review, one action class at a time.
What the pilot is really testing
The open question is how much of a governed software pattern transfers to physical systems. Events can share an envelope contract, but hardware adds commissioning, device state, network timing, read-back limits, occupants, and independent safety requirements. Those differences must be tested rather than assumed away.
Keep reading: What running on live commerce taught us · Warehouse seconds