AI-Based Thermal Comfort: From One Setpoint Per Zone to Comfort That Adapts to People

Walk into almost any open-plan office and you'll hear some version of the same argument: someone near the window thinks it's too warm, someone near the vent thinks it's too cold, and the thermostat between them insists the room is exactly 72°F.

All three of them are right. That's the problem. The HVAC system isn't necessarily malfunctioning. The thermostat is simply measuring conditions at one location and controlling the zone around that measurement. But people don't experience the thermostat. They experience the space around them. And that difference is where AI-based thermal comfort begins.

1. How Comfort Control Works Today

Most zones — a room, an open floor plate, sometimes an entire wing — are controlled to a single setpoint, read from a single thermostat or sensor. That number is chosen to satisfy the average occupant in that zone, under average conditions, as best as one number can.

This isn't a design failure. It's the practical outcome of controlling temperature with one sensor and one setpoint per zone — a real engineering constraint, not an oversight. The system is doing exactly what it was built to do.

2. Why One Setpoint Satisfies Almost No One Exactly

Thermal comfort was never just about air temperature. The engineering models behind it — most notably Predicted Mean Vote (PMV) — account for radiant temperature, humidity, air speed, clothing insulation, and metabolic rate, alongside air temperature itself. Two people in the same room, at the same air temperature, can have genuinely different comfort experiences because every one of those other variables differs between them.

Position in the room compounds this further. A desk near a south-facing window picks up solar radiant heat the thermostat across the room never senses. A desk near a supply vent sits in a localized cold draft the same thermostat also never senses. The single sensor is only ever describing conditions at its own location — everywhere else in the zone is an approximation.

3. What Changes When Feedback Enters the Loop

One missing input in conventional control is the occupant. A thermostat has never asked anyone whether they're comfortable — it only measures air temperature at one point and compares it to one setpoint.

AI-based comfort control adds occupant feedback as a genuine input — a simple too-warm/too-cold response through an app or a desk control, sometimes supplemented by desk-level or wearable sensors. This is the same category of input introduced in Part 2 (historical data, tariffs, equipment status), just personal rather than building-wide: not "what does this building usually need," but "what does this specific person, in this specific spot, usually need."

4. From Fixed Setpoint to Adaptive Zoning

Once feedback accumulates over time, the system isn't just holding one number for the whole zone anymore — it's learning where within that zone conditions consistently run warmer or colder than the setpoint suggests, and why. A row of desks near the window that repeatedly reports "too warm" in early afternoon, or a section near a vent that consistently reports "too cold" on windy days, becomes a recognizable pattern rather than an occasional complaint.

Where the zone has sub-zone actuation available — VAV boxes, reheat coils, diffuser dampers — the system can bias conditions toward those learned patterns: slightly more cooling near the window in the afternoon, slightly less aggressive airflow near the vent on windy days. The zone-level setpoint doesn't disappear; what changes is how evenly that setpoint's intent is actually delivered across the physical space.

5. Why This Doesn't Replace the Thermostat or the Standard

AI-based comfort control doesn't replace the engineering framework established by ASHRAE 55. The standard provides methods for determining satisfactory thermal environmental conditions, considering both environmental and personal factors. What AI can change is how intelligently the HVAC system responds to those conditions and occupant patterns.

AI doesn't ventilate to a different standard and it doesn't comfort to a different target. It delivers the same standard and the same target — just more consistently, to more people in the zone.

6. A Practical Example

Consider an open-plan zone — ten desks deep — with the thermostat located at the center desk and a 72°F setpoint. Desks nearest the window pick up solar gain; desks nearest the supply vent sit in the strongest draft.

Under fixed single-zone control, the thermostat reads 72°F because that's where it sits — but occupants at the window and near the supply vent may experience very different thermal conditions. Nobody at either end is being lied to by a broken system; they're both experiencing real local conditions the single sensor was never positioned to see.

Under adaptive, feedback-informed control, accumulated position-based feedback can help narrow that difference — not to zero, but meaningfully. The system recognizes that the window side repeatedly runs warmer during periods of solar gain, while the area near the supply vent repeatedly experiences cooler conditions. Same thermostat. Same zone. Same setpoint — but a more informed response to what people actually experience.

• Fewer comfort complaints — fewer occupants sitting meaningfully outside the zone's intended comfort range
• Fewer manual overrides and thermostat disputes — the system is already resolving some of the disagreement the setpoint alone couldn't
• Reduced energy waste as a byproduct — zones are frequently over-cooled or overheated system-wide just to satisfy the most sensitive occupant in the room, the same category of waste identified in Part 2, this time driven by comfort disputes rather than fixed schedules

None of this requires new equipment beyond what a VAV system already has. It requires better information about where, within the zone, conditions actually diverge from the setpoint's intent.

8. Real-World Evidence

Occupant feedback-based control has a genuine research base behind it. A 2024 peer-reviewed review of AI-based thermal comfort controls surveyed a range of individual studies incorporating occupant feedback into HVAC control. Reported results vary considerably by method, building type, sensing strategy, and control approach. Some studies report meaningful energy savings while maintaining or improving thermal comfort — but the review makes clear that AI-based thermal comfort control remains an active research area rather than a universally proven building solution.

That same review makes an honest observation worth repeating here: in most of the studies it surveyed, thermal comfort improvement was a byproduct of research primarily focused on energy management — not the main objective. Separate field research on personal comfort modeling — building individualized comfort predictions from occupant behavior and feedback rather than population averages — has demonstrated that machine learning can meaningfully predict individual thermal preference, reinforcing that the underlying mechanism is sound even where it wasn't the study's main focus.

As with earlier parts in this series, reported outcomes vary by building, sensor coverage, and how feedback is collected. The mechanism is well supported, but the outcome will depend on the building, controls, and available data.

9. Limitations

• True per-person comfort in a shared space has physical limits. Two desks sharing the same duct run can only be biased so far apart before one starts affecting the other — a real constraint of shared air distribution, not a software gap. The 2024 review is clear: no real-time solution yet exists for fully satisfying multiple occupants' individual thermal comfort within one shared zone. This remains an open research problem.
• Feedback systems can skew toward whoever bothers to give feedback. Feedback quality itself is a known limitation — occupant feedback can be affected by mood, stress, and general well-being, not just actual thermal sensation.
• It's still a zone-level compromise, just a better-informed one. Sub-zone bias narrows the spread; it doesn't grant every individual occupant their own personal climate.

In short: this narrows the gap between a single setpoint and real occupant experience. It doesn't remove the physical reality of sharing conditioned air across a shared space.

10. Looking Ahead

So far, each part of this series has looked at one system in isolation: energy, maintenance, ventilation, comfort. In Part 6, we'll look at what happens when all of these operate together as part of a connected, AI-integrated smart building.

References