Most commercial buildings condition and ventilate space on a fixed schedule, sized for a level of occupancy that rarely occurs. In hybrid-work offices, attendance between the busiest and quietest weekday commonly varies by half or more. Replacing the schedule with live occupancy data lets HVAC respond to actual demand, with expected energy savings of 20 to 40 percent on affected systems.
The mechanism is well understood and the mechanical plant is rarely the obstacle. What usually blocks it is the occupancy data itself: whether it is accurate enough to sequence against, whether it updates quickly enough to matter, and whether it arrives somewhere the control system can actually read it.
This article covers where the waste comes from, how demand-controlled ventilation actually works, and what the data has to look like for it to function.
For the architectural context, start with our guide to on-premise occupancy intelligence.
Where the waste comes from
HVAC is typically the largest single energy consumer in a commercial office building. In most of those buildings, it runs on assumptions rather than information.
The schedule assumption. Plant starts at 6am and runs to 7pm, Monday to Friday, because that is what the sequence was commissioned to do. It does not know that the fourth floor is empty every Friday.
The design occupancy assumption. Ventilation rates are calculated from design occupancy, which is the maximum a space was built to hold. A 12-person meeting room ventilates for 12 people whether there are 12 in it, three in it, or none.
The zone averaging assumption. A single zone sensor stands in for an entire floor plate, so a floor that is packed at one end and empty at the other gets conditioned to the average of the two.
None of these were bad decisions. They were the only decisions available when the alternative was no information at all. Hybrid work made them expensive.
The pattern is consistent across enterprise portfolios: a sharp midweek peak, quiet Mondays and Fridays, and a building that conditions all five days identically. Every hour of that gap is energy purchased to condition air for people who are not in the building.
How demand-controlled ventilation actually works
Demand-controlled ventilation, usually shortened to DCV, modulates outdoor air supply based on actual occupancy instead of design occupancy. It is well established in standards including ASHRAE 62.1, and it is a required strategy under an increasing number of energy codes.
The traditional method uses carbon dioxide sensors as an occupancy proxy. CO2 rises as people breathe, so the control system infers occupancy and increases outdoor air.
That works, with two significant limitations.
It lags. CO2 concentration takes time to build. A meeting room fills at 9:00 and the CO2 sensor registers meaningful change several minutes later. The ventilation response trails the actual demand by the length of that lag, which is precisely the period when the room most needs it.
It infers rather than measures. CO2 gives a rough proxy for how many people are present. It cannot tell you the room is empty and the reading is residual, or that four people just walked in.
Direct occupancy counting removes both problems. Instead of inferring demand from a downstream effect, the control system receives the headcount. Ventilation is calculated against the people actually present, and it responds when they arrive rather than several minutes after.
The two approaches are complementary rather than competing. Occupancy data drives the ventilation rate. CO2 and other indoor air quality measurements verify that the rate is achieving the intended result.
Beyond ventilation
DCV is the clearest case, but it is not the only one.
Optimal start and stop. Plant starts based on when the building is actually occupied rather than a fixed clock time. In a building where Friday attendance runs at a third of Wednesday’s, the difference over a year is significant.
Zone-level setback. Floors and wings that are genuinely unoccupied are set back rather than conditioned to full comfort. This requires occupancy resolution at the zone level, not one sensor per floor.
Meeting room conditioning. Meeting rooms are usually the most over-conditioned spaces in a building, because they are sized for peak and used at a fraction of it. They are also where occupancy data has the highest resolution.
Lighting. Occupancy-driven lighting control is mature technology, but it is normally implemented with separate PIR sensors doing nothing else. If accurate occupancy data is already on the network for HVAC, lighting can use the same signal.
No-show release. A booked room that nobody entered can be released automatically. This is a utilization gain rather than an energy one, but in large enterprise portfolios it commonly runs to hundreds of recovered room-hours per month.
What the data has to look like
This is where most projects fail, and it is worth being blunt about it.
Occupancy data that is good enough for a quarterly utilization report is often not good enough to drive plant. Three requirements separate the two.
Accuracy at the space level
A utilization report averages error out across weeks of readings. A ventilation sequence acts on a single value at a single moment. If the headcount is wrong, the ventilation rate is wrong for as long as the error persists.
Ask for accuracy measured at the space level rather than the sensor level, and ask how it was measured. Accuracy in the high nineties is the practical threshold for driving plant.
Update latency
A control sequence can only be as responsive as its slowest input.
This is where the sensing technology choice becomes decisive. Battery-powered sensors extend their life by transmitting infrequently, often on intervals measured in minutes. That is a sound engineering decision for a device designed to report utilization for five years on a cell. It also puts a hard ceiling on how quickly automation can respond, and no amount of integration work removes it.
Mains-powered sensing with update intervals measured in seconds gives the control system something it can actually sequence against. XY Sense sensors update every two seconds.
Delivery into the control system
Occupancy data sitting in a cloud dashboard cannot drive a chiller. It has to arrive in the BMS as a readable point.
In practice that means BACnet/IP objects the BMS can subscribe to, published by an on-premise occupancy gateway on the building’s own network, or UDMI messages a smart building platform can consume. An API that a systems integrator has to build a custom bridge for will technically work, but it adds a layer that has to be maintained by somebody for the life of the deployment.
What this is worth, by city
The savings percentage is only half the answer. What matters is what it converts to on the bill, and that swings enormously by market because tariffs vary by a factor of three.
The table below models the same building in nine cities: a three-floor tenancy of 4,500 m² (roughly 48,000 sq ft), typical of an enterprise office footprint. Every figure is in USD so the markets can be compared directly.
Figures are illustrative. They model one building type across nine tariff and climate conditions to show the range, not to predict any particular building’s result.
How these figures were calculated
Anyone presenting this to a CFO should be able to defend every input, so here they are.
Building. 4,500 m² across three floors. Scale linearly for a different footprint: a 9,000 m² tenancy doubles every number.
HVAC energy intensity. A baseline of 100 kWh/m² per year, adjusted for climate. The baseline comes from CBECS data showing US commercial buildings averaging around 80 kBtu per sq ft per year across all energy sources, with heating, cooling and ventilation accounting for roughly 40 percent of that. Climate adjustment ranges from 0.70 for San Francisco’s mild year-round conditions to 1.45 for Singapore’s continuous cooling and dehumidification load.
Tariffs. New York 21.9, San Francisco 25.8 and Dallas 8.4 US cents per kWh, all from EIA commercial rates for April 2026. Singapore 31.91 SGD cents from the SP Group regulated tariff for Q3 2026, before GST. Dublin 24.8 euro cents and Paris 17 euro cents from Eurostat non-household bands. London 21 pence, mid-point of the 18 to 24 pence range quoted for large half-hourly commercial contracts in 2026.
Savings range. 20 to 40 percent applied to HVAC load.
What the table does not account for
Treat these as order-of-magnitude figures for building a case, not as a quote.
The climate adjustments are engineering judgment rather than measured data for each city. Tariffs are market benchmarks, not contracted rates, and your own rate is the single input that moves the answer most. Demand charges, which can be a large share of a commercial bill in the US and Australia, are excluded entirely. Gas-fired heating is folded into the energy intensity rather than priced separately, which understates heating-dominated markets where gas is cheap relative to power.
Where the plant cannot modulate, the saving is zero regardless of what the table says. Check that before anything else.
The part that is no longer optional
Everything above frames this as a savings decision. In a growing number of markets it is now also a compliance one, and that changes the arithmetic. If a building is going to have to reduce measured energy use regardless, the question stops being whether to spend and becomes what to spend it on.
Here is where the nine markets in the table above actually stand.
Market | Requirement | Exposure |
|---|---|---|
Dublin, Paris and the rest of the EU | Building automation and control systems mandatory for non-residential buildings where HVAC rated output exceeds 290 kW. Threshold drops to 70 kW at the end of 2029. | National law since the 29 May 2026 transposition deadline. Enforced through national regimes. |
New York | Local Law 97 emissions caps on buildings over 25,000 sq ft. Plus benchmarking, audits and sub-metering under LL84, LL87 and LL88. | $268 per metric ton of CO2e over the cap, every year. $0.50 per sq ft per month for failing to file. |
London | MEES. EPC E minimum to let. EPC B confirmed for 2031 on lettings above 1,000 m². | Up to £150,000 per breach depending on rateable value, plus publication on the public exemptions register. |
Singapore | Mandatory Energy Improvement regime for buildings over 5,000 m² that exceed the EUI threshold three years running. | Audit within 90 days, an improvement plan, measures delivered within three years, and the gain held for a year after. |
Sydney | Commercial Building Disclosure. Office space over 1,000 m² needs a Building Energy Efficiency Certificate before sale, lease or sublease, with the NABERS rating shown in all advertising. | The rating is public and it prices into the transaction. |
San Francisco | State benchmarking and disclosure, plus a growing set of city and state building performance standards across the US. | Varies by jurisdiction. |
Texas | No state-level building performance standard. | None currently. |
India is absent from the table because the Energy Conservation Building Code is adopted state by state, so a Bangalore position needs a local read rather than a national one.
Europe now requires the control system itself
Most of these rules set a target and leave the method to the owner. Cut emissions. Improve the certificate. Publish the rating.
Europe went further. Article 13 of the recast Energy Performance of Buildings Directive requires qualifying non-residential buildings to operate a building automation and control system that can continuously monitor, log and analyze energy use, allow that use to be adjusted, benchmark efficiency, detect losses in building systems, and communicate with the systems connected to it.
That obligation is live today for buildings where HVAC rated output exceeds 290 kW, which covers most large offices. At the end of 2029 the threshold falls to 70 kW, which pulls in a far larger share of the stock, including buildings nobody currently thinks of as being in scope.
The law tells owners to have a control system. It says nothing about what to feed it, and that gap is where the money is won or lost. A BACS running on scheduled assumptions and design occupancy will satisfy the regulation and deliver very little. The same system fed accurate live headcount will hit the same compliance requirement and take 20 to 40 percent off the HVAC bill while doing it. Both buildings pass. Only one of them stops paying for empty space.
Why this favors spending on the data layer
For any building in scope, the capital is committed. The open question is where it goes.
Spend it on measures that clear a certificate and the result is a compliance event: a cost incurred, a box ticked, and nothing that shows up on next year’s bill. Spend it on the control layer and the saving recurs annually, improves the position at the next assessment, and keeps improving as the requirements tighten.
They will tighten. New York’s 2030 to 2034 emissions limits run roughly 40 percent stricter than the current period, and Urban Green Council analysis suggests most covered buildings will exceed them without intervention. Europe’s BACS threshold drops in 2029. London moves to EPC B in 2031. Singapore’s energy intensity thresholds are reviewed as the stock improves. A building that scrapes over the line today faces the same problem again in four years, from a worse starting position and with less time.
The escape routes are also narrowing. Recent New York rulemaking restricts buildings in the fine mitigation process from meeting emissions caps with renewable energy certificates. Regulators increasingly want measured reductions in actual consumption rather than offsetting arrangements, which puts the value on anything that genuinely lowers the meter reading.
There is a timing argument as well, and for portfolios facing a deadline it is the decisive one. Occupancy-driven control cuts consumption without touching the building fabric. It deploys in weeks rather than the years that glazing, insulation or plant replacement demand, and it does not require the building to be emptied. It is one of the few levers that moves inside a compliance window instead of across a refurbishment cycle.
Building a business case
The savings figure will be specific to the building, but the calculation structure is consistent.
- Establish the baseline. Current HVAC consumption, separated from total building load.
- Quantify the occupancy gap. Actual measured occupancy against the design occupancy the systems are sequenced for. This is usually the number that changes the conversation, because it is larger than people expect.
- Identify the addressable load. Which systems can respond to occupancy: air handling, terminal units, zone setback, lighting.
- Apply a realistic savings range. 20 to 40 percent on affected systems is the expected range for occupancy-driven control. Where the building lands depends on how variable attendance is and how much the plant can modulate.
- Add the reporting value. Occupancy-linked energy data feeds Scope 1 and Scope 2 reporting, and supports NABERS, WELL and RESET certification. For many organizations this carries weight independent of the cost saving.
- Add avoided penalty exposure. If the building sits in scope of a performance standard, model the fine it avoids as well as the energy it saves. In New York that is a calculable number: tonnes over the cap multiplied by $268, every year, against limits that tighten in 2030.
One thing to check early: whether the plant can actually modulate. A constant-volume system with no variable-speed capability cannot respond to a demand signal no matter how good the data is. Confirm the mechanical capability before building a case on the savings.
Frequently asked questions
How much energy can occupancy-based HVAC control save?
Expected savings are 20 to 40 percent on affected HVAC systems. The range depends on how variable the building’s attendance is, how much of the plant can modulate, and how conservatively the existing sequences are set. Buildings with high hybrid-work variability sit at the upper end.
How much money can occupancy-based HVAC control save a typical office?
For a three-floor tenancy of about 4,500 square meters, modeled annual savings run from roughly $7,500 in a low-tariff market such as Bangalore to over $60,000 in Singapore, where cooling runs year-round and electricity is expensive. The contracted electricity rate and the climate move the answer more than anything else.
What is demand-controlled ventilation?
A control strategy that adjusts outdoor air supply based on actual occupancy rather than design occupancy. It is recognized in ASHRAE 62.1 and required under a growing number of energy codes. Traditionally driven by CO2 sensors, increasingly driven by direct occupancy counting.
Are CO2 sensors good enough for demand-controlled ventilation?
They work, but they lag. CO2 concentration takes minutes to build after people arrive, so the ventilation response trails actual demand during exactly the period the space needs it most. CO2 also infers occupancy rather than measuring it. Direct occupancy counting responds immediately and measures directly. The strongest implementations use occupancy data to drive the rate and CO2 to verify the outcome.
Can occupancy sensors control HVAC directly?
Not directly, and they should not. The occupancy data is published to the building management system as readable points, typically over BACnet/IP, and the BMS control sequences use it as an input. The occupancy system provides the signal. The BMS retains control of the plant.
Are buildings legally required to have building automation?
In the EU, increasingly yes. The recast Energy Performance of Buildings Directive requires non-residential buildings with HVAC rated output above 290 kW to have a building automation and control system capable of monitoring, logging, analyzing and adjusting energy use. That threshold has applied since the end of 2024 and drops to 70 kW at the end of 2029. Elsewhere the regulations tend to mandate an outcome rather than a technology: New York caps emissions with penalties per tonne, the UK sets a minimum EPC rating to let, Singapore requires audits and improvement plans for energy-intensive buildings.
Does this work with an existing BMS, or does it need a new one?
It works with an existing BMS in most cases, provided the BMS speaks BACnet/IP, which almost all do, and the plant can modulate. What usually needs to change is the control sequence, not the hardware.
