HVAC AI Energy Savings establishes a per-site, weather-normalized baseline from your historical telemetry, optimizes setpoints and schedules against day-ahead grid pricing, and tracks variance so you know mid-month whether savings are on target — not after the utility bill arrives.
Most vendor-claimed energy savings are the difference between a modeled 'pre' scenario and a modeled 'post' scenario. Neither number comes from a meter. An M&V specialist or a utility rebate program won't accept that methodology, and a CFO who has seen two cycles of 'we'll save you 20%' should be skeptical of claims that can't be verified. HVAC AI Energy Savings is designed for the audience that needs to stand behind the number.
The baseline is established using your building's own historical telemetry: 12–36 months of consumption data, weather-adjusted using NOAA degree-day data specific to the building's location. The resulting baseline is IPMVP Option B-compatible — energy savings are measured at the system boundary, not estimated from whole-building models. For portfolios where Option B metering isn't available at every system, an Option C regression model is built from whole-building utility data. Both approaches produce a defensible savings claim.
On top of that baseline, the optimization layer runs continuous. Setpoints are trimmed during low-occupancy periods and pre-loaded before occupancy peaks — not on a fixed schedule, but driven by the occupancy signals your building already produces (access card events, calendar feeds, CO2 sensors). Peak shaving logic reads day-ahead grid pricing and demand-charge forecast data, then adjusts pre-cool and pre-heat windows to shift compressor load out of demand-peak windows. Demand charges represent 30–50% of a commercial HVAC electricity bill; targeting them directly is often more impactful than chasing kWh alone.
Variance tracking runs continuously against the established baseline. If a building is running 12% above its savings target at mid-month, the agent surfaces that gap with an explanation — a hot spell that exceeded the weather model, a setpoint override that wasn't reverted, a piece of equipment running inefficiently. You course-correct in the current billing cycle, not after you've already missed the target. Utility rebate exports are formatted for the major programs — Con Edison, PG&E, BC Hydro, Xcel Energy — so rebate applications submit the telemetry data the utility program actually requires.