Night setback is one of the oldest tricks in facility energy management. The logic is straightforward: when the building is unoccupied, raise the cooling setpoint (or lower the heating setpoint), reduce ventilation airflow, and let the space drift toward unoccupied conditions. Come morning, start recovery early enough that the space is comfortable again by first occupancy.
This works. It has worked for decades. Most well-run commercial buildings use some form of it. But there is a systematic problem that night setback creates — one that erases a meaningful fraction of the savings it generates. That problem is the morning demand spike.
The Physics of Thermal Recovery
When an HVAC system is told to recover a space from setback temperature to occupied setpoint, it cannot throttle that recovery — not with a traditional schedule-based BMS, anyway. The system runs at or near full capacity until the space reaches the target setpoint. The AHUs ramp up, chiller staging comes online, and for a 30–90 minute window every weekday morning, the building draws significantly more power than at any other point in the day.
On a hot August morning in the Southeast, a 150,000 sq ft office building that has run a cooling setback to 80°F overnight might need to recover 10–12°F of zone temperature before the first occupants arrive at 8 AM. If the BMS is set to start recovery at 6:30 AM, the cooling plant will operate at high staging from 6:30 until roughly 7:45 — drawing perhaps 400–500 kW in a building that normally peaks at 300 kW during occupied hours.
That 90-minute spike at 400–500 kW will very likely set the monthly peak demand reading. And if this building is on a demand rate where every kW of monthly peak costs $14–18, a single morning recovery event has potentially added $1,400–3,000 to the monthly bill.
The night setback saved some energy overnight. The morning spike may have cost more in demand charges than the setback saved in energy consumption. This is the hidden arithmetic of traditional night setback.
Why Fixed Recovery Start Times Make This Worse
Traditional night setback programs use a fixed start time for recovery. The BMS is programmed to begin occupied-mode conditioning at, say, 6:30 AM, seven days a week, regardless of outdoor conditions.
On a mild spring morning where the space has only drifted a few degrees from setpoint, 6:30 AM recovery start is more than adequate — the system barely works to get back to setpoint by 8 AM. Demand impact is minimal.
On a brutal July morning after a hot night, the same 6:30 AM start time may not be enough. The space might still be 3–4°F above setpoint at 8 AM. Occupants arrive to a warm building. The facility manager gets complaints. The response is to start recovery even earlier — say, 5:30 AM — which extends the high-demand period by another hour and drives demand charges even higher.
The fixed start time approach is trying to serve two conflicting goals at once: minimize energy use and ensure comfort. It cannot optimize for both simultaneously because it has no information about what conditions it is actually dealing with on any given morning.
What Predictive Scheduling Does Differently
A predictive scheduling system approaches this problem with two pieces of information that a fixed schedule lacks: a weather forecast for the next 24–48 hours and a thermal model of the building.
The thermal model is what matters most. It captures how quickly the building's thermal mass heats or cools under different outdoor conditions — the effective time constant of the building envelope and the HVAC system's capacity relative to the heat load. That model is built from historical data: how long did it actually take to recover from setback on mornings with similar conditions?
With a weather forecast and a calibrated thermal model, the system can calculate: given tonight's projected low temperature, tomorrow morning's expected outdoor temperature at the time recovery starts, and this building's measured thermal characteristics, what is the latest time we can start recovery and still hit setpoint by occupancy?
On a mild April morning, that answer might be 7:15 AM. On a hot July morning after an 82°F overnight low, it might be 5:45 AM. The start time is different every day, calibrated to the actual conditions rather than worst-case assumptions.
This removes the demand spike on mild mornings entirely — the system does not ramp up an hour before it needs to. On hot mornings, it still ramps up early, but it does so knowing that it needs to, rather than as a precaution against uncertainty.
The Demand Charge Arithmetic
The financial impact of this difference depends heavily on tariff structure. For buildings on flat energy-only rates, the savings come purely from reduced consumption during the earlier recovery window that predictive scheduling avoids. That is real but modest — maybe 2–5% of total energy cost.
For buildings on tariffs with demand charges, the math is more compelling. Consider a mid-size commercial office building in a territory where the peak demand charge is $15/kW/month. If predictive scheduling avoids a single 200 kW demand spike on a mild-weather morning when the traditional schedule would have started recovery an hour early, that is $3,000 in avoided demand charges in that one month. Over a year, across the months where mild conditions mean over-eager traditional recovery creates unnecessary peaks, the annual demand charge savings can reach 8–15% of total electricity spend.
We are not saying predictive scheduling is a complete replacement for night setback — the setback itself still saves energy overnight, and that benefit remains. What changes is the recovery phase: instead of a blunt full-ramp at a fixed time, recovery becomes a calculated ramp at the right time for the actual conditions that day.
What This Requires From Your System
Predictive scheduling is not a setting you flip on in your existing BMS. It requires a control layer that can push optimized start times to the BMS — typically through BACnet or Modbus write commands — based on daily recalculation. The BMS still executes the setpoint and schedule commands. The predictive layer tells it when to start.
The thermal model calibration takes time. In our experience, a building needs 4–6 weeks of data collection across varying weather conditions before the model is reliable enough to use in production. During that period, the system observes how the building responds to recovery under different outdoor temperature conditions and builds its time-constant estimates.
The data requirements are modest: outdoor temperature (from a weather API, not from a local sensor), zone temperature telemetry from the BMS, and HVAC system state data (is it running, at what staging level). Most buildings with any kind of networked BMS can provide this. Buildings running standalone pneumatic or early electronic controls without network access need integration work before predictive scheduling is possible.
A Scenario to Illustrate the Difference
Consider a 120,000 sq ft suburban office campus — two buildings, shared central plant, located in a region with hot summers and mild shoulder seasons. The facility runs a standard night setback program: cooling setpoint raised to 79°F from 8 PM to 6:30 AM, recovery start fixed at 6:30 AM every weekday.
In May and October, when overnight lows are in the mid-50s, the building barely drifts from occupied setpoint overnight. The 6:30 AM recovery start has the system running hard for an hour when 20 minutes of moderate operation would have sufficed. Monthly peaks are being set by this unnecessary ramp-up.
In July, the same 6:30 AM start time is barely adequate after hot nights in the mid-70s. Facility staff have added 30-minute buffer by adjusting to 6:00 AM during summer months — which helps comfort but extends the high-demand window further.
Under predictive scheduling: the May/October morning recovery starts at 7:05–7:20 AM on mild days, cutting the early-ramp demand spike by 60–70%. July mornings still start early, but the model calculates exactly how early is needed — not a fixed conservative buffer. The peak demand readings for the shoulder months drop noticeably. Annual electricity bill savings come primarily from lower demand charges in the 4–5 shoulder-season months where the fixed schedule was most over-conservative.
The energy consumption savings are secondary to the demand charge savings in this scenario. That ordering is common for buildings on commercial tariffs — demand charges drive the ROI, not consumption reduction per se.