What problem does it solve?
Space heating is the single largest energy end use in US commercial buildings, at about 32% of total energy use in 2018, with ventilation adding roughly another 10% (US EIA). A building that runs this equipment on fixed schedules and setpoints does not adapt as occupancy, weather and electricity prices change hour to hour. According to BrainBox AI, rising energy costs and regulations led Cammeby's International, a real estate investment company, to look at AI for its 32 storey office property in New York City's financial district.
Facility teams can tune a building management system by hand, but re tuning every zone continuously as conditions change takes staff time. Without that continuous attention, a building can end up running wider, more conservative margins than a continuously optimized building would need, and can miss the chance to shift load to the hours when the local electricity grid is cleanest.
- Space heating was about 32% of total US commercial building energy use in 2018, the largest single end use; ventilation and lighting were next at about 10% each (US EIA).Use of energy in commercial buildings (2018)
How does it work?
- Connect to the existing building management system. An edge device or cloud connection reads live data points, such as temperatures, valve positions and equipment status, typically over the BACnet protocol, without replacing hardware.
- Learn the building's thermal behaviour. The AI models how each zone responds to outside weather, occupancy and equipment changes, specific to that building rather than a generic template.
- Predict and act ahead of time. The system forecasts the next hours of demand and adjusts valve positions, fan speeds and equipment sequencing continuously, cooling or heating the building before the need arrives.
- Shift load where a grid signal is available. With a marginal emissions or price signal, the AI pre cools the building during the hours when the grid is cleanest, then lets the temperature drift during the hours when the grid relies more on fossil fuels, so the building's thermal mass carries the load instead of the equipment.
- Report through the existing tools. Facility engineers watch the AI's decisions through customised graphics inside their own building management system, rather than a separate interface, and can adjust or override a setpoint if needed.
- Audience
- Back office
- Autonomy
- Autonomous
- Adoption
- Early adopters
- Channels
- Internal tools, API and system to system
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Energy savings | Too few to pool | 10% to 15.8% | 2 | 2 vendor |
| Cost savings | Not pooled | USD 42,951 | 1 | 1 vendor |
Value drivers: Compliance quality.
Indicative value
An office building with 300,000 square feet of AI controlled HVAC
USD 5625 to USD 75,000
Annual HVAC energy cost avoided per year
How this is calculated
Formula: squareFeet * hvacCostPerSqFt * savingsShare. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Square feet under AI control squareFeet, square feet | 150,000 | 500,000 | Editorial assumption. Cammeby's International's controlled space was 251,104 square feet (BrainBox AI), within this range. |
| HVAC energy cost per square foot per year hvacCostPerSqFt, USD per square foot per year | 0.75 | 1.5 | Editorial assumption. BrainBox AI's Cammeby's International figures ($42,951 saved from a 15.8% reduction in HVAC related electricity, across 251,104 square feet over an 11 month period) imply a total HVAC electricity cost of about $271,800 over that period ($42,951 / 0.158), or roughly $1.08 per square foot over the 11 months and about $1.18 per square foot annualised, assuming the dollar saving is proportional to the 15.8% consumption reduction; replace with your own utility spend. |
| Share of HVAC energy cost saved savingsShare, fraction of HVAC energy cost | 0.05 | 0.1 | At or below the two reported results on this page (a 15.8% reduction in HVAC related electricity consumption at Cammeby's International and a 10% savings in HVAC related energy at Loyola University's Schreiber Center, both BrainBox AI), to allow for a different building's baseline and scope of HVAC energy cost. |
What it leaves out: HVAC energy cost avoided only. It leaves out the software subscription, any emissions credit or compliance benefit, and the risk that an aggressive deployment without a hard comfort band produces occupant complaints.
Who already uses it?
2 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Loyola University Chicago
United States · Education · 2024
Loyola University's Schreiber Center, a LEED Gold certified, 10 storey mixed use building that houses Loyola's Quinlan School of Business in Chicago, ran a year long proof of concept combining BrainBox AI's HVAC optimization with WattTime's Automated Emissions Reduction signal, which pre cools the building during low emissions events, when the local grid has surplus renewable energy, and lets it drift when the grid relies more on fossil fuels. The project was carried out with UC Berkeley's Center for the Built Environment, and BrainBox AI points to a fuller study in an ASHRAE guide on the role of grid interactivity in decarbonization.
- Energy savings: 10%, over the year long proof of concept
"we were able to achieve a 10% savings in HVAC-related energy and a 10% reduction in HVAC-related CO2e emissions through our AI for HVAC solution"
Claimed by: vendor
Cammeby's International
United States · Real estate · 2023
Cammeby's International, a real estate investment company, deployed BrainBox AI's AI Control solution across a 32 storey, 386,315 square foot office building in New York City's financial district, built in 1983, controlling 251,104 square feet of air handling units, variable air valves, outdoor and exhaust fans, and the hot and chilled water systems. Over an 11 month period in 2023, the deployment cut HVAC related electricity consumption, cost and emissions without disrupting daily operations, according to the building's facility engineers.
- Energy savings: 15.8%, over an 11 month period in 2023
"Our AI Control solution drove a 15.8% reduction in HVAC-related electricity consumption, saving $42,951, and mitigating 37.14 tCO2eq."
Claimed by: vendor - Cost savings: USD 42,951, over an 11 month period in 2023
"Our AI Control solution drove a 15.8% reduction in HVAC-related electricity consumption, saving $42,951, and mitigating 37.14 tCO2eq."
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Live read and write access to the building management system's HVAC data points
- A record of the HVAC equipment and how zones map to it
- At least a few weeks of historical operating data to calibrate the thermal model
Systems to integrate
- Building management system, typically over BACnet or an equivalent protocol
- Weather forecast data
- Optionally, a grid emissions or price signal for load shifting
Complexity: Medium
The AI needs a working connection to the building's existing controls, commonly BACnet, and a controls contractor's cooperation to expose the right points. In a typical rollout, most of the effort is integration and change management with the building's engineering team, rather than the AI modelling itself.
- 1
Start with the highest value system
At Cammeby's International, the building's own team set out to target savings particularly in its chilled water loop, according to BrainBox AI's case study. Naming the equipment that carries the most load up front gives an early, defensible result to point to, even though the rollout itself went floor by floor across all the connected equipment, air handling units, variable air valves, fans and both the hot and chilled water systems, rather than one system at a time.
- 2
Work with the controls contractor, not around them
Program the AI's decisions into graphics the facility team already uses inside its own building management system. At Cammeby's International, BrainBox AI rolled out floor by floor in close cooperation with the building's Chief Engineer, so daily operations were not disrupted.
- 3
Measure a real baseline first
Establish a measured baseline period before claiming a saving, so the figure holds up against normal seasonal swings. BrainBox AI reported Cammeby's International's 15.8% reduction over an 11 month results period in 2023, without disclosing how the baseline itself was built.
- 4
Add a grid or price signal once the basics work
Once core optimization is stable, a marginal emissions or price signal, as Loyola University's Schreiber Center used with WattTime, lets the building shift load to cleaner or cheaper hours by pre cooling ahead of time and drifting during the dirtier hours.
- 5
Keep occupant comfort as a hard constraint
Set temperature and ventilation bands the AI may not cross regardless of the savings opportunity, and track comfort complaints alongside the energy numbers.
Guardrails
- Hard temperature and ventilation bands the AI cannot exceed, enforced in the building management system itself, not only in the AI's own logic
- No changes to life safety, fire, or code required ventilation and pressure sequences
- A facility engineer can see and override every AI decision through the existing building management system graphics
KPIs to instrument
- HVAC related energy consumption and cost, measured against a comparable prior period rather than a single year over year comparison
- Occupant comfort complaints during and after rollout
- Emissions avoided when a grid signal drives load shifting
Human in the loop
Facility engineers watch the AI's live decisions through their own building management system graphics and can override any setpoint, and a reliability or sustainability team reviews measured savings against the baseline on a schedule.
Common failure modes
- Savings measured against the wrong baseline
- Weather and occupancy swing year to year; a saving claimed against a single prior year, rather than a normalised baseline, will not survive scrutiny.
- Comfort complaints erase trust in the program
- Aggressive pre cooling or setpoint drift without a hard comfort band produces complaints that get the whole program switched off; keep comfort bands non negotiable from day one.
What are the risks and rules?
EU AI Act
Minimal risk
Optimizing HVAC equipment is not listed in Annex III. Point 2 covers AI safety components in the management and operation of critical infrastructure such as the supply of water, gas, heating or electricity, and a building's own HVAC controller does not manage infrastructure at that level. The system also does not decide credit, employment, access to essential services or another high risk use. Under Article 6(1), an AI system that is a safety component of a product covered by Annex I harmonisation legislation, such as the Machinery Regulation, and that needs third party conformity assessment, would be high risk regardless of Annex III; this is why the guardrails on this page keep the AI out of fire, life safety and code required ventilation sequences rather than letting it override them.
Rules that apply
Controls to put in place
- A documented safe operating envelope, covering temperature, humidity and ventilation rate, that the AI cannot leave, enforced independently of the AI's own decisions
- A logged history of every setpoint or sequence change the AI made, so a disputed comfort complaint or an unusual energy month can be traced to a specific decision
Frequently asked questions
- How much energy does AI HVAC optimization actually save?
- BrainBox AI reports an 11 month measured 15.8% reduction in HVAC related electricity consumption at Cammeby's International's office tower in Manhattan, saving $42,951 and avoiding 37.14 tonnes of CO2 equivalent. At Loyola University's Schreiber Center in Chicago, a year long project found a 10% savings in HVAC related energy and 10% lower HVAC related CO2e emissions. Results vary by building age, climate and how the baseline is measured.
- Does it need new hardware?
- Both deployments on this page connected to the building's existing management system over BACnet, but Cammeby's International needed a new edge device: BrainBox AI's case study says the solution was deployed through its own edge device, communicating over BACnet IP. At Loyola University's Schreiber Center, BrainBox AI says it integrated its system with the building's existing HVAC controls via BACnet, with no edge device mentioned.
- Can it help a building use more renewable energy?
- Indirectly. Loyola University's Schreiber Center paired HVAC optimization with WattTime's marginal emissions signal to shift some cooling to the hours when the local grid relies more on renewables. BrainBox AI reports a 15% average reduction in HVAC emissions during marginal emissions events, including the following four hour drift period.
- Is this high risk under the EU AI Act?
- Usually not. HVAC optimization is not listed in Annex III's high risk categories, including point 2 on critical infrastructure for water, gas, heating or electricity supply, since a building's own HVAC controller does not manage infrastructure at that level, and it is not a safety component of a product under Annex I harmonisation legislation such as the Machinery Regulation. If it did perform that kind of safety function, Article 6(1) could make it high risk regardless of Annex III, which is why a well designed deployment keeps it out of fire, life safety and code required ventilation sequences.
How to cite this page
Blits.ai AI Use Case Library, "AI for building HVAC and energy optimization", last verified 28 September 2026, https://www.blits.ai/ai-use-cases/building-energy-optimization. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 28 September 2026: First published