Static setback schedules were designed for a world where building occupancy was predictable. In that world, a commercial office building fills up at 8 AM, empties at 6 PM, and the HVAC system could be pre-programmed to match those transitions with fixed time windows. That world existed, in a narrow sense, in single-tenant commercial buildings with stable anchor tenants and no hybrid work patterns. It doesn't describe most commercial buildings operating today.
The Setback Schedule Assumption
A setback schedule is a time-based HVAC control program: raise the cooling setpoint to 80°F at 10 PM, return it to 72°F at 6 AM. Simple, auditable, and tuned once during BMS commissioning — then left to run indefinitely.
The problem isn't the concept of setback. Temperature setback during unoccupied hours is sound HVAC engineering. The problem is the assumption that the restore transition — the moment the building moves from night setback back to occupied setpoint — can be scheduled at a fixed clock time, independent of what the building will actually do on any given day. That assumption fails routinely, and in exactly the conditions where demand charges are highest:
- Cold overnight after a warm week: The building has more thermal inertia to overcome than the Tuesday-average the schedule was set for. HVAC runs at full staging for 90 minutes instead of 45, creating a demand spike in the first morning peak-tariff window.
- Low-occupancy day with full pre-conditioning: A company holiday or Friday remote-work day leaves the building at 25–35% occupancy, but the BMS schedule pre-cools for 100% capacity. Maximum demand charges captured on a minimum-occupancy day.
- Event-driven occupancy surge: A conference or off-site event brings 150% of normal occupancy on a day the schedule treats as average. Under-conditioning creates complaints and then a manual override that spikes demand when operators push setpoints down rapidly.
- Shoulder-season swing day: In Minneapolis, April brings days that start at 35°F and reach 68°F by 2 PM. The setback schedule designed for a stable summer cooling season has no mechanism to handle the heating-to-cooling mode switch mid-day without overcycling.
None of these are edge cases. They occur throughout the billing month. And because demand charges are set by the single highest 15-minute interval recorded, any one of them can inflate your total demand charge for the month — regardless of how well-managed the other 29 days were.
Where the Demand Charge Actually Comes From
Most facilities managers understand conceptually that demand charges are based on peak consumption, but the mechanics are worth being precise about. Commercial utility bills typically include a demand charge line item calculated as: the highest 15-minute average kilowatt demand recorded during the billing period, multiplied by the utility's demand charge rate (typically $8–22/kW in the Midwest).
That single 15-minute window sets the demand charge for the entire billing month. A building with excellent demand management 29 days out of 30 that experiences one morning ramp-up spike on the 30th day pays demand charges as though it spikes every day.
The HVAC morning setback restore is the most common source of that spike. When the system transitions from an 80°F setback to a 72°F target with 200 people arriving in 45 minutes, it stages every available cooling unit simultaneously. The resulting kW draw in that 15-minute window frequently exceeds anything the building records for the rest of the month — including peak summer afternoons when occupancy is higher but HVAC has been running continuously rather than cold-starting.
Peak-tariff time-of-use structures compound this. Midwest commercial accounts on TOU tariffs often face on-peak energy charges of $0.18–0.28/kWh during morning and afternoon windows (roughly 9 AM–9 PM on weekdays), versus $0.07–0.12/kWh off-peak. A pre-conditioning ramp that starts at 6 AM avoids the on-peak energy rate entirely. One that starts at 7:45 AM catches the building at maximum staging exactly when the on-peak window opens.
What a 72-Hour Thermal Forecast Changes
A thermal demand forecast doesn't replace the concept of setback. It replaces the fixed clock timing of the restore transition with a physics-grounded, building-specific, day-specific pre-conditioning window.
Instead of "pre-cooling begins at 6:00 AM every weekday," the forecast delivers: "on Wednesday, with a forecast outdoor dry-bulb of 34°F at 5 AM rising to 52°F by noon, 68% expected occupancy by 9 AM, and a building thermal mass recovery rate of approximately 0.8°F per hour under those conditions — begin pre-cooling at 4:50 AM with a setpoint target of 69.5°F, coasting to 71°F by 8:45 AM."
That level of specificity — 70 minutes earlier than the default schedule, a specific target temperature, tuned to Wednesday's actual weather and occupancy prediction — is the difference between a controlled pre-cooling ramp that stays below the peak demand threshold and a cold-start spike that sets the monthly demand charge.
The pre-cooling window is where demand charges are won or lost. A 72-hour thermal forecast gives you the window. A fixed setback schedule doesn't know the window exists.
Why the Horizon Is 72 Hours, Not 24
The 24-hour forecast drives the actual staging schedule for tomorrow — it's the operationally critical output. At 24-hour horizon, weather forecast accuracy is high enough (and MAPE below 8% is achievable on most building types) that the staging windows it generates are specific and actionable.
The 48- and 72-hour horizons serve a different function. They provide early identification of high-demand days two to three days ahead — the days where outdoor temperatures, occupancy patterns, or both will create a demand charge exposure that requires preparation beyond a normal pre-cooling window.
These advanced signals are operationally significant for facilities teams in two scenarios:
- Demand response program participation: Utility demand response programs (ISOs and co-ops operating demand curtailment programs) typically send dispatch signals with 30-minute to 2-hour lead times. A 72-hour forecast lets facilities managers pre-position HVAC thermal state before the dispatch signal arrives, so curtailment doesn't require setpoint sacrifice from an already-warm building.
- Event-driven occupancy planning: Facilities managers scheduling large building events — move-in weekends, conferences, tenant events — can see 72 hours ahead whether the thermal demand on that day is likely to require extended pre-conditioning and coordinate accordingly, rather than reacting with manual overrides on the day.
Accuracy degrades at the 72-hour horizon (MAPE below 12% is the target), but even degraded forecast accuracy outperforms a fixed schedule that carries no weather or occupancy information at all. The 72-hour horizon is not used for precise staging schedule execution — that's the 24-hour output's job. It's used for planning headspace.
The Accuracy Threshold for Actionable Staging
For a forecast to translate reliably into staging schedule reductions, it needs to be accurate enough that the pre-cooling window it specifies doesn't produce systematic overshoot or undershoot. Overshoot — pre-cooling too deep, too early — wastes energy and can create a different demand spike when occupancy arrives and HVAC reduces staging. Undershoot — not pre-cooling far enough — fails to catch the morning ramp peak and the demand charge impact is unchanged.
The benchmark for operational adequacy on 24-hour thermal demand forecasts is MAPE below 8%. This figure comes from the load forecasting literature on commercial building short-term forecasting — it's the threshold at which pre-cooling windows generated from the forecast consistently produce better outcomes than fixed schedules in the peer-reviewed evidence base, including work from Lawrence Berkeley National Laboratory's building energy research program.
Heatvelo's validation methodology uses time-series cross-validation: the model is trained on 80% of historical building data, tested on the remaining 20% in temporal order (no look-ahead), and MAPE is calculated building-type by building-type. We share per-building backtest results with operators before live forecasting begins.
One important qualification: MAPE performance is building-specific, not universal. Buildings with stable occupancy patterns and high-quality BMS historian data consistently reach or beat the 8% threshold. Buildings with erratic occupancy, low sensor coverage, or BMS historians with significant data gaps produce higher MAPE figures — and we tell operators this during backtest, rather than discovering it during the live pilot.
The Practical Argument
There is a version of this argument that requires no reference to MAPE, thermal mass time constants, or TOU tariff mechanics. It goes like this:
A fixed setback schedule sets the pre-cooling window to the same time every weekday, regardless of what the weather will be or how many people will be in the building. A 72-hour thermal forecast sets the pre-cooling window based on what the weather will actually be and how many people are actually expected. The first approach ignores available information. The second uses it.
For buildings where demand charges represent 20–35% of total utility spend — which describes most commercial office, retail, and mixed-use properties on commercial TOU tariffs — the economic case for using available information is straightforward. The data to do it is already being collected. The BMS historian has it. The weather API has it. The occupancy access control logs have it. What's been missing is the model that combines those three inputs into a forward demand curve and translates that curve into staging windows a facilities manager can act on. That's the gap Heatvelo was built to close.