Updated June 2026

Commercial buildings in the United States waste approximately 30% of the energy they consume, according to the EPA’s ENERGY STAR program. As facilities that operate around the clock with variable occupancy, high ventilation requirements, and massive glass curtainwall envelopes, waste in airport terminals is even more pronounced. HVAC systems alone account for 40–60% of total energy consumption, and the financial consequences of inefficiency extend well beyond the utility meter.

The question is no longer whether AI has a role in managing these complex energy environments. The more interesting question is where, specifically, AI-driven monitoring uncovers problems that conventional building management cannot. Based on our first hand experience with a large east coast terminal, the impacts are significant.

The invisible conflicts inside building systems

Most commercial HVAC systems are configured by independent contractors during commissioning and rarely revisited as a whole. The result is a common but hard-to-detect failure mode: subsystems that actively work against each other.

Consider a real-world example from a recently commissioned airport terminal. Circuit-level AI monitoring of 24 rooftop HVAC units revealed that cooling and humidification setpoints had been configured in conflict. The cooling system was removing moisture that the humidification system was simultaneously adding, and vice versa. Neither system was malfunctioning in isolation and both were responding correctly to their individual setpoints. But together, they were locked in a cycle of wasted energy that no periodic walk-through or monthly utility bill would ever surface.

The estimated cost of this single conflict, up to $25,600 per year, depending on seasonal load. The fix was a setpoint adjustment, not a capital expenditure. It’s not difficult to imagine this was one of many such issues identified, with compounding benefits.

This is the kind of finding that separates AI-driven energy monitoring from traditional BMS (Building Management System) approaches. A BMS reports what each system is doing. A smart AI approach based on careful data collection efforts can identify when what systems are doing doesn’t make sense in combination.

Demand charges: the line item that punishes you for one bad interval

For many commercial electricity customers, demand charges—fees based on the highest rate of power drawn during any single 15-minute interval in a billing period—account for 30% to 70% of the total monthly electric bill. The mechanics are unforgiving: one spike in a single 15-minute window sets the demand charge for the entire month, regardless of how efficiently the facility operated the rest of the time.

In airport terminals and other large facilities with dozens of independently controlled HVAC units, the risk multiplies after a power outage. When utility power is restored, every unit attempts to restart simultaneously, creating a massive coincident load spike. In the airport terminal deployment mentioned above, AI monitoring identified this exact vulnerability: 32 rooftop units with no stagger-start sequencing. A single simultaneous restart event could produce a demand spike large enough to reset the facility’s supply capacity charge—a significant financial exposure on a $2.4 million annual electricity supply contract.

The solution is a controlled stagger-start sequence, spacing unit restarts over a window of minutes rather than seconds. It’s an operational protocol change, not a hardware investment. But without circuit-level monitoring and AI-driven pattern recognition, the gap between what should happen during a power restoration and what actually happens is invisible until it shows up on a bill.

Tenant loads and the metering blind spot

In multi-tenant commercial facilities—airport terminals, mixed-use buildings, shopping centers—landlords and operators typically meter energy at the building level or at the floor level, rarely at the individual tenant circuit. This creates a persistent blind spot: tenants whose actual energy consumption dramatically exceeds their allocated or expected share.

In the airport terminal deployment, AI analysis of circuit-level data flagged a single food-and-beverage tenant operating at approximately four times its expected electrical load. The estimated excess cost was up to $32,500 per year. Without circuit-level visibility, this overuse would have been absorbed into the building’s overall energy costs indefinitely, spread across all tenants through common area maintenance charges or simply absorbed by the operator.

This type of finding has implications beyond energy cost. Sustained overloading of a circuit can accelerate wear on distribution equipment, increase the risk of thermal events, and shorten the useful life of switchgear and conductors. Identifying it is both a financial and a safety question.

Power factor: a quiet drag on electrical efficiency

Power factor—the ratio of real (working) power to apparent total power drawn from the grid—is one of the most overlooked efficiency metrics in commercial buildings. A power factor of 1.0 means every amp of current drawn is doing useful work. Most commercial buildings operate between 0.75 and 0.85 without correction, and utilities in many jurisdictions impose penalties or surcharges when power factor falls below 0.9.

At the airport terminal in question, the power factor measured across the monitored HVAC units averaged 0.6—well below the 0.9 threshold that most utilities consider acceptable. At that level, a significant portion of the electrical capacity serving those units is consumed by reactive power that does no useful work, effectively reducing the real capacity available and increasing demand charges. Correcting power factor from 0.7 to 0.95 in a typical commercial building can reduce demand charges by 20–30%, often with a payback period of one to three years.

The challenge is that power factor degradation is invisible without circuit-level monitoring. It doesn’t trigger alarms. It doesn’t cause equipment to fail. It simply costs more every month.

What AI monitoring actually does differently

The common thread across these examples is not the sophistication of the AI algorithms—it’s the resolution of the data. Traditional building management operates at the system level: the chiller is running, the air handler is on, total building consumption is X kilowatt-hours. AI-driven monitoring operates at the circuit level, ingesting data from individual units and correlating it across systems, time periods, and operational states.

This distinction matters because the most expensive inefficiencies in complex facilities are emergent—they arise from the interaction of systems, not from the failure of any single component. A cooling-vs.-humidification conflict is invisible at the system level because both systems are working. A demand spike from simultaneous restarts is invisible in monthly consumption data because total energy use didn’t change. A tenant overloading a circuit is invisible at the building meter because the building’s total load is within expected range.

AI adds value not by replacing facility engineers, but by giving them visibility into the interactions and patterns that no human operator could monitor continuously across thousands of data points from multiple often disconnected systems.

Threading and making use of all available data types can reveal statistical correlations in the data to seed machine learning algorithms in Peaxy’s Machine Learning Manager module, part of our energy analytics solution. See our previous article on anomaly detection using ML-enabled data pipelines for more information.

The data challenge is real

None of this works without the right data infrastructure. A large commercial facility can produce millions of data points per day across electrical, mechanical, and environmental sensors. The data must be ingested in real time, tagged and threaded correctly to individual assets and circuits, and stored at sufficient resolution to support time-series analysis and anomaly detection.

This is a non-trivial engineering problem. Many facilities have some monitoring in place—a BMS, a set of utility meters, perhaps a handful of submeters—but the data is siloed, aggregated at too coarse a level, or simply not retained at the resolution needed for AI analysis. Bridging this gap—what might be called the “first mile” of data capture—is often the most consequential step in any AI-driven efficiency initiative, and one we focus on closely.

The computational demands are significant as well. At-scale monitoring across a portfolio of facilities requires big-data architecture capable of handling high-frequency time-series data, active data curation to maintain quality, and cost-conscious storage strategies that balance retention depth against analytical needs.

Where this is headed

The regulatory environment is accelerating the urgency. New York City’s Local Law 97, now in its first year of imposing financial penalties, charges building owners $268 per metric ton of CO₂ equivalent above their annual emissions cap for buildings over 25,000 square feet. ASHRAE 90.1-2022 adoption is accelerating across states, with the DOE estimating 14% energy savings over the 2019 edition. SEER2 and EER2 are now the standard efficiency metrics for commercial HVAC equipment, reflecting real-world operating conditions rather than laboratory ideals.

For airport terminals specifically, the pressure is compounding. Terminals consume a disproportionate share of an airport’s total energy—roughly 60% of a facility that may draw up to 180 million kWh per year. Commercial electricity prices rose 10% in the first half of 2025. For operators managing annual electricity budgets in the millions, even single-digit percentage improvements in efficiency translate to six-figure savings.

AI-driven monitoring doesn’t eliminate the need for experienced facility engineers, good commissioning practices, or thoughtful building design. What it does is surface the problems that those professionals would fix immediately—if only they could see them. The gap between what building systems are doing and what operators think they’re doing is where the money is. A comprehensive AI solution focused on appropriate levels of data capture closes that gap.