Much of the HVAC waste in commercial buildings doesn't come from broken equipment. It comes from working equipment doing the wrong thing quietly: valves that don't fully close, schedules that don't match occupancy, sensors that drift a fraction of a degree each month.
Your building automation system records all of it. A mid-size commercial building logs thousands of data points every few minutes — valve commands, actual positions, temperatures, fan status, pressures. Much of the evidence of that waste is sitting in that trend data right now.
The problem was never collection. It's that no human can read millions of rows a week. That's the job AI is actually good at in buildings — not controlling anything, just reading everything. Here are six things it can find that alarms and walkthroughs usually miss.
1 · The gap between commanded, reported, and real
The BAS tells a valve to close; the valve reports 35% open. No alarm fires — alarms watch for failures, not disagreements. AI compares available command and feedback histories and flags persistent gaps. That's the easy case.
The harder case is when the equipment lies. A valve with a worn seat can report fully closed and pass hot water anyway — command 0%, feedback 0%, and nothing in the control system admits a problem. The evidence lives one step downstream, in physics: a discharge-air temperature running warmer than a closed coil can explain. Dampers fail the same way — a seized linkage or a broken actuator can return perfect feedback while the blades never move, and the only witness is a mixed-air temperature that doesn't match what the commanded outside-air fraction should produce.
When expected and actual physics disagree, something in the chain is lying — and it can be a few-hundred-dollar part quietly costing thousands a year.
This is how trend analysis catches mechanical failure, not just control failure: cross-checking every command and every feedback signal against the temperatures they should be producing.
2 · Two systems fighting each other
Simultaneous heating and cooling is the classic hidden fault: one coil adds heat the other coil immediately removes. Comfort holds, so nothing prompts an investigation. Detecting it requires watching heating and cooling behavior together, across systems, hour by hour. People review one screen at a time; AI reads every unit's heating and cooling calls side by side and flags any hour where both are active without a legitimate reason.
3 · Slow drift no snapshot can see
A sensor that reads 0.3°F lower every month never trips a threshold. Eighteen months later it's 5°F wrong and every control decision downstream is acting on a lie. Drift is invisible to any single reading — it only appears when you compare a sensor against its own history and against its neighboring zones over months. That's a pattern-over-time problem, which is precisely what machine analysis does well and quarterly walkthroughs rarely catch.
4 · Schedules that stopped matching reality
Someone extends HVAC hours for an event and never sets them back. A holiday calendar from two ownerships ago keeps running. AI cross-references what equipment actually ran against occupancy patterns and flags runtime that serves nobody — nights, weekends, break periods. In our experience this is often one of the first places money turns up, because overrides outlive the events they were made for.
5 · The baseload that never sleeps
Ventilation, cooling, and lighting are the largest electricity uses in commercial buildings, about half of the total together¹ — and a flat overnight load profile means some of it is running for an empty building. AI baselines what "asleep" should look like for each building and flags the ones that never get there.
6 · The dollar value of each finding
Detection alone isn't the win. A raw fault list adds to an already crowded workload. Ranked impact estimates help teams choose what to fix next. The step that changes behavior is attaching an impact estimate to every finding and ranking the list by dollar impact, severity, and ease of repair, so a facilities director knows the first three things to fix and what fixing them is worth.
Does it actually pay? The published evidence
This isn't a vendor claim — the field results are public. In Lawrence Berkeley National Laboratory's Smart Energy Analytics Campaign, organizations using fault detection and diagnostics software reported median energy savings of 9% after two years.² LBNL's meta-analysis of existing-building commissioning — fixing the same kinds of operational faults — found 16% median whole-building savings with a 1.1-year payback.³
What to demand from any AI you point at your building
One rule separates useful building AI from expensive dashboards: every finding must show its work. If the tool says a valve is leaking $18,000 a year, you should be able to open the trend data behind that claim and check it yourself — because your controls contractor will ask, and your CFO will ask. A black-box score nobody can interrogate ends up ignored, no matter how accurate it is. That's the principle behind AIR: every score opens to its findings, and every finding opens to its trends.
The other rule: it shouldn't require new anything. The data above — commands, positions, temperatures, runtimes — is already being logged by the BAS you own. If a vendor's first step is a hardware proposal, they're solving a different problem.
Common questions
- Does this replace my facilities team?
- No. It helps your team review more trend data and focus its investigation. Your team, or your contractor, still decides what to fix and makes the fix. The AI's job is making sure the fixable problems are visible and ranked.
- Do we need new sensors or meters to start?
- No. Analysis runs on the trend data your BAS already collects, shared as a trend export or through a live API connection installed onsite. No new sensors, and no control writes.
- How much trend data does it take?
- Common faults like fighting coils, stale schedules, and stuck valves can show up in weeks of trend data, not years. 90 days or more works best. A shorter sample may not reveal an issue, and if the analysis is inconclusive, we'll tell you.
1. U.S. EIA, 2018 Commercial Buildings Energy Consumption Survey, energy sources and end uses: ventilation, lighting, and cooling account for 49% of commercial building electricity use.
2. U.S. DOE / LBNL, Smart Energy Analytics Campaign results: median savings of 9% for fault detection and diagnostics software users after two years.
3. Mills, E., LBNL, Building Commissioning: A Golden Opportunity for Reducing Energy Costs and Greenhouse Gas Emissions (643 buildings): 16% median whole-building savings and a 1.1-year median payback in existing buildings.