If ventilation standards already specify how much outdoor air a space needs, why do so many buildings still over-ventilate, under-ventilate, or both — sometimes in the same day?
1. Introduction
Ventilation is one of the most standardized parts of HVAC design. Codes specify exactly how much outdoor air a space needs, down to the person and the square foot. And yet buildings following those same codes routinely over-ventilate empty rooms, under-ventilate crowded ones, or manage to do both in the same day, in the same space.
That's not a compliance failure. It's a design limitation — and understanding why is the key to understanding what AI actually changes here.
2. How Ventilation Is Conventionally Controlled
Ventilation standards such as ASHRAE 62.1 specify outdoor air rates based on a combination of floor area and design occupancy — the maximum number of people a space is expected to hold. A conference room designed for 20 people gets an outdoor air rate sized for 20 people, calculated once at design time and built into the AHU's fixed damper position or a simple schedule.
That rate is deliberately conservative — sized for the room at its busiest, on a schedule that assumes the room is roughly that busy whenever it's in use. It's a reasonable engineering approach given the information available at design time: nobody sizing the system in advance knows exactly when the room will be empty, half full, or over capacity on any given day.
3. Where This Goes Wrong in Both Directions
Over-ventilation happens whenever the room is running below its design occupancy — which, for most conference rooms and meeting spaces, is most of the time. A fixed system still conditions and delivers outdoor air sized for 20 people even when the room holds two, or none. This is the same category of waste identified in Part 2: energy spent conditioning space nobody's using.
Under-ventilation happens in the opposite case: when actual occupancy exceeds the design assumption. A 20-person room hosting a 45-person town hall doesn't get more outdoor air just because more people showed up — the fixed system delivers the same rate it always does, sized for an occupancy the room has already exceeded. CO₂ and other occupant-generated pollutants build up faster than the system was ever designed to remove them.
Same room, same fixed system, two very different failure modes — depending entirely on who happens to be in the room that hour.
Fixed Ventilation: Two Failure Modes
Same conference room — design occupancy: 20 people
Figure 1 — Under a fixed ventilation rate, both failure modes produce poor outcomes: wasted energy when under-occupied, degraded air quality when over-occupied.
4. Why CO₂ Is the Practical Proxy
Directly counting occupants in real time — with cameras or dedicated people-counting sensors — is possible, but expensive and often impractical to deploy at scale across every room in a building. CO₂ offers a practical shortcut: occupants continuously exhale CO₂, so indoor CO₂ concentration rises predictably with occupant density and falls as outdoor air dilutes it.
CO₂ sensors are inexpensive, reliable, and already standard equipment in many demand-controlled ventilation (DCV) systems. A rising CO₂ trend is, in practice, one of the most reliable low-cost signals a system has for "more people just arrived than this room usually holds."
5. From Reactive DCV to Predictive Ventilation
Demand-controlled ventilation isn't new — plenty of buildings already modulate outdoor air based on live CO₂ readings, increasing ventilation once levels start climbing. That's a real improvement over a fixed schedule. But it's still reactive: the system only responds once CO₂ has already started rising, which means occupants spend the first part of any high-occupancy event breathing air the system hasn't caught up to yet.
What AI adds is the same shift seen in Parts 2 and 3: moving from reacting to anticipating. Using the same kind of inputs — calendar bookings, historical occupancy patterns, and live CO₂ trends — the system can identify that a room is about to be busy before anyone walks in, and begin increasing outdoor air ahead of the event instead of chasing the CO₂ curve after it's already climbing.
Figure 2 — From Occupancy Signals to Ventilation Rate
Figure 2 — AI combines calendar bookings, live CO₂ trends, and historical occupancy to predict demand and adjust ventilation rate before occupancy rises.
6. Why This Doesn't Replace Ventilation Standards
AI-based ventilation control doesn't override or replace codes like ASHRAE 62.1 — those minimums remain the compliance floor, exactly as designed. What AI changes is how efficiently the system operates within and above that floor: throttling down when a space is under-occupied instead of running the fixed design rate regardless, and ramping up ahead of high-occupancy events instead of only reacting once a threshold is crossed.
7. A Practical Example
Consider a conference room designed for 20 people, with a target indoor CO₂ level of 1,000 ppm — a common operational reference point used in ventilation design, not a universal health or safety threshold, but a practical marker of when a space is being adequately diluted with outdoor air relative to its occupancy.
A 45-person town hall is booked from 10:00 to 11:00 — more than double the room's design occupancy.
Under fixed ventilation, the outdoor air damper holds its standard position all day. When the meeting starts at 10:00, CO₂ begins climbing immediately and the fixed-rate system — sized for 20 people, not 45 — can't keep pace. Levels peak well above the 1,000 ppm target before the meeting ends and gradually settle back down afterward. Outside the meeting, the same system keeps ventilating an empty room at that same fixed rate all day, wasting energy the rest of the time.
Under predictive ventilation, the system sees the 10:00 booking in advance and begins increasing outdoor airflow at 9:40 — 20 minutes ahead of occupancy. By the time attendees arrive, the room is already being flushed with additional outdoor air, and CO₂ stays close to the 1,000 ppm target throughout the meeting instead of spiking past it. For the rest of the day, while the room sits empty, the system throttles ventilation down well below the fixed system's constant rate, since there's no scheduled or detected occupancy to justify it.
Figure 3 — Conference Room: Fixed vs. Predictive Ventilation
Full-day operation — 45-person town hall (10:00–11:00) in a room designed for 20
Fixed damper: 80% all day
AI damper: 25–40% when empty
AI drops to 620 ppm
Figure 3 — Fixed ventilation runs at 80% damper all day: CO₂ spikes to 1,750 ppm during the 45-person meeting, and energy is wasted when the room is empty. AI adjusts the damper dynamically: throttles down to 25–40% when empty (saving energy), pre-ventilates before the meeting (dropping CO₂ to 620 ppm), and increases to 75–85% during the meeting to keep CO₂ near the 1,000 ppm target.
Same room, same code minimums, same equipment — a materially different outcome in both directions: better air quality during the unusually large meeting, and less wasted energy the rest of the day.
📊 View data table for Figure 3
| Time | Fixed CO₂ | AI CO₂ | Fixed Damper % | AI Damper % |
|---|---|---|---|---|
| 7:00 | 800 | 700 | 80% | 40% |
| 8:00 | 800 | 680 | 80% | 35% |
| 9:00 | 800 | 620 | 80% | 30% |
| 10:00 | 950 | 1,010 | 80% | 75% |
| 11:00 | 1,750 | 1,100 | 80% | 85% |
| 12:00 | 1,450 | 980 | 80% | 70% |
| 14:00 | 1,000 | 830 | 80% | 45% |
| 16:00 | 850 | 730 | 80% | 35% |
| 18:00 | 800 | 660 | 80% | 25% |
Values shown for illustrative purposes — actual results vary by building.
8. Where the Value Comes From
- Reduced over-ventilation energy waste, by throttling down outdoor air (and the conditioning load that comes with it) when a space is unoccupied or under design capacity — directly extending the savings sources identified in Part 2
- Better air quality during unusually high-occupancy events, by increasing outdoor air ahead of demand instead of only after CO₂ has already climbed past target
- Lower lag time between an occupancy change and the system's response — the same reacting-to-anticipating shift that drove the savings and detection improvements in Parts 2 and 3
Neither benefit requires new ductwork, new AHUs, or new dampers — the same equipment, operating on better-timed decisions.
9. Real-World Evidence
Demand-controlled ventilation itself has a long, independently documented track record. A field study coordinated by the Air Infiltration and Ventilation Centre (AIVC) measured DCV performance across four real school and office buildings over multiple weeks, comparing it against constant air volume operation. The results showed fan energy reductions in the range of 25–55%, along with meaningfully lower heat losses, while indoor CO₂ levels stayed within acceptable ranges throughout — evidence that reduced airflow during low-occupancy periods didn't come at the cost of air quality.
That study measured conventional, reactive DCV rather than the predictive, AI-assisted layer described in this article — the predictive extension is newer and less extensively documented in independent field studies. But it's built directly on top of a control strategy with a well-established evidence base: the mechanism (throttling ventilation to actual demand) is proven; what predictive AI changes is only the timing of when that throttling happens, shifting it earlier relative to occupancy rather than in response to it.
aivc.org →
10. Limitations
- CO₂ is a proxy, not a direct air quality measurement. It correlates closely with occupant density but says nothing about VOCs, particulate matter (PM2.5), or other pollutants that may be present regardless of occupancy
- Sensor placement and calibration matter. A poorly placed or drifting CO₂ sensor gives the system a misleading picture of the room — the same data-quality dependency raised in Part 3
- Prediction is only as good as the occupancy signal behind it. An unscheduled, unbooked gathering gives the system no calendar signal to act on ahead of time — it still has to fall back on reacting to CO₂ once it starts climbing
In short: this improves how efficiently and how quickly ventilation responds to occupancy. It doesn't replace the need for adequate sensor coverage, correct calibration, or a broader air quality strategy where contaminants beyond CO₂ are a concern.
Key Takeaways
- Fixed ventilation sized for design occupancy produces two failure modes: energy waste when under-occupied, poor air quality when over-occupied
- CO₂ is the most practical and cost-effective proxy for real-time occupancy detection
- Conventional DCV improves on fixed schedules but remains reactive — responding after CO₂ has already risen
- Predictive AI ventilation anticipates occupancy using calendar data, acting before the event rather than after
- ASHRAE 62.1 minimums remain the compliance floor — AI operates within and above them, not around them
- Field evidence for conventional DCV shows 25–55% fan energy reductions without sacrificing air quality
- CO₂ sensors don't detect VOCs, PM2.5, or other pollutants — a broader air quality strategy requires additional sensing
11. Looking Ahead
Ventilation is only one part of how occupants experience a space. In Part 5, we'll look at thermal comfort — how AI moves beyond a single setpoint per zone toward control that accounts for how comfort actually varies from person to person.
In spaces you manage, which failure mode shows up more often in practice: energy wasted ventilating empty rooms, or air quality suffering when a space runs over its design occupancy?