Editor’s note: FM Perspectives are industry op-eds. The views expressed are the authors’ and do not necessarily reflect those of Facilities Management Advisor.
Artificial intelligence (AI) is showing up in more conversations about HVAC energy savings. That is not surprising. Heating and cooling are expensive, controls data is increasingly available, and many buildings still have room for operational improvement.
But facilities managers have a different test than the software demo.
The real question is not whether a dashboard looks smart. It is whether the building stays comfortable on a hot afternoon, whether operators can understand what changed, whether equipment is protected, and whether the team can override the system when the day does not go as planned.
That is why facility teams should look closely at what an AI HVAC tool actually does before giving it a role in optimization. Some tools mainly predict, monitor, or flag issues. Others help recommend or make control decisions over time.
Both can be useful. They are not the same thing.
Here are five questions facilities managers should ask before trusting AI with HVAC optimization.
1. Is the system only predicting conditions, or is it helping decide what to do next?
Prediction has real value. A tool that forecasts cooling load, detects abnormal energy use, or flags equipment behavior before failure can help a facility team act sooner.
Optimization goes one step further. It is about deciding what action should happen next.
In HVAC, that might mean when to start cooling, how aggressively to pre-condition a space, whether to adjust a setpoint, how to stage equipment, or when to reduce load ahead of a demand peak. Those decisions require more than knowing what is likely to happen. They require balancing energy use, comfort, equipment limits, occupancy, and operating priorities.
A forecast might tell the team that afternoon cooling demand will rise. A control-oriented optimization system should help answer a more practical question: What should we do now so the building performs well later?
This is where vendor language matters. If the product mainly offers dashboards, alerts, and forecasts, it may be an analytics tool. If it recommends or writes limited control actions within approved boundaries, it is closer to an optimization tool.
Facilities managers do not need every tool to be an optimizer. They do need to know which one they are buying.
2. How does it handle delayed effects?
Buildings do not respond instantly. Thermal mass, equipment ramp-up time, weather, humidity, occupancy, and operating schedules all create lag.
That lag is where many HVAC decisions become difficult.
Pre-cooling may increase energy use in the morning but reduce afternoon peak demand. Relaxing a setpoint may save energy for an hour but create comfort calls later. Running equipment less aggressively may reduce consumption now but slow recovery after occupancy changes.
A useful optimization system should not focus only on the next few minutes. It should understand that a control decision made now may show its full impact later in the day.
This is one reason control-oriented AI approaches, including reinforcement learning, are being explored for HVAC. The important point for facilities managers is not the algorithm name. The important point is whether the system is built for sequential decisions: actions, outcomes, feedback, and adjustment over time.
A practical question for vendors is: Does the system optimize for the next hour only, or does it consider the next several hours of operation?
The goal is not to win a short-term energy calculation. The goal is to run the building better over the full day.
3. Does it work with the existing BAS or BMS?
Most facilities are not looking to replace their building automation system (BAS). They already have schedules, alarms, trend data, control logic, operator screens, and maintenance workflows built around the BAS or building management system (BMS).
An AI HVAC project should fit that reality.
In many buildings, the practical model is a supervisory layer. The AI system sits above the existing BAS or BMS, reads operating data, and then recommends or writes limited changes based on rules the facility team has approved.
Before deployment, facilities managers should understand the integration clearly: These are operational questions, not just technical ones. They determine whether the tool will fit into daily building operations or become another system the team has to work around.
A strong AI HVAC project should make the existing controls environment more useful. It should not turn the BAS into a black box behind another black box.
4. What guardrails protect comfort, equipment, and operators?
AI should not mean unrestricted control.
Commercial buildings have comfort expectations, lease obligations, maintenance routines, safety considerations, and equipment constraints. Any optimization system should operate inside clear boundaries.
Those boundaries may include minimum and maximum setpoints, humidity limits, occupied schedules, equipment runtime limits, staging rules, alarm conditions, lockouts, manual shutoff procedures, and operator override rights.
Operator control is especially important. Facility teams should be able to see what the system is doing, understand why a recommendation was made, and override it when needed.
There will always be exceptions: a tenant event, a maintenance activity, a complaint-heavy day, an unusual weather pattern, or an equipment issue. On those days, the system should support the people responsible for the building. It should not fight them.
Trust is built when operators can inspect, limit, and override automation. A useful AI system should reduce operational burden, not create a control layer that the team is afraid to touch.
5. How will savings be measured?
Energy savings claims can sound simple. Measuring them is not.
HVAC energy use changes with weather, occupancy, hours of operation, equipment condition, tenant behavior, and building use. A lower utility bill does not automatically prove that optimization worked. The weather may have been milder. Occupancy may have changed. A schedule may have shifted.
Facilities managers should ask how savings will be measured against a credible baseline. The exact method may vary by building, but the principle should be clear: Compare performance against what the building would likely have used under similar conditions.
Comfort should be part of that evaluation. Lower energy use is not a success if it leads to hot and cold calls, humidity issues, poor recovery, or added equipment stress.
This matters especially when savings are used to justify a pilot or ongoing service fee. Before the project begins, both sides should understand what data will be used, how the baseline will be built, how comfort will be tracked, and what counts as a successful outcome.
Good optimization should be judged over time, with energy, comfort, reliability, and operator experience all in view.
Turning AI into a Practical HVAC Tool
AI can help facility teams operate buildings more efficiently. But the most important question is not whether a product uses AI. It is what kind of operational problem the product is built to solve.
For HVAC, true optimization is not just prediction. It is better decision-making over time in a system where actions have delayed consequences.
That can be valuable when it is deployed carefully: on top of existing BAS infrastructure, within clear guardrails, with operators still in control, and with savings measured against a credible baseline.
Facilities managers should approach AI HVAC optimization with practical questions: If the answers are clear, AI can become more than another dashboard. It can become a practical tool for reducing energy waste while keeping buildings comfortable, safe, and manageable.
Chuan He is the founder of ClimaMind, a company focused on BAS-compatible HVAC optimization for commercial buildings. His work focuses on supervisory AI approaches that operate within existing building automation infrastructure while respecting comfort, safety, equipment, and operator constraints.