A lighting schedule can remain unchanged for two years after a floor shifts from 9-to-5 occupancy five days a week to hybrid work. Facilities teams are not necessarily overlooking it. The problem is the missing feedback loop: no signal shows when scheduled lighting has drifted from actual occupancy, so the gap goes unobserved.
The Static Schedule Problem
Commercial lighting is usually managed by zone through a BMS or separate lighting controls. Schedules set occupied mode, with full illumination, versus unoccupied mode, with lights off or at minimum. Commissioning teams usually set them from tenant lease terms and building hours.
HVAC has temperature feedback: a zone that stays too warm or cold eventually prompts a complaint. Lighting has no matching signal. Full illumination at 8:30pm in an empty building creates no complaint. It simply raises the utility bill, where the cost blends into the building's monthly energy total.
Lighting waste is therefore hard to see. It appears diffusely in the electricity bill without a clear cause. In a typical commercial office, lighting uses 25-35% of total electrical consumption. Measurement and verification studies suggest 15-25% of that occurs during low- or zero-occupancy periods. Thus, lighting schedules can account for 4-9% of total building electrical use.
Why Sensors Alone Fall Short
Point-in-time sensors, including motion and PIR sensors, target a narrower issue: lights staying on after a space has been empty for 10 minutes. They work well for that case and are common in modern commercial buildings at zone or room level. Their strength is reducing short-cycle waste.
Pattern waste is different. It occurs when a schedule supplies occupied-level light during periods with low average occupancy. Motion detected as someone passes keeps the lights on for that visit. But when Friday afternoons are typically 20% occupied and the schedule runs full lighting from 7am to 7pm every day, the sensor cannot revise the schedule. It can only briefly extend lighting in a zone already scheduled to be dark.
Pattern waste calls for a schedule change, not only a response to presence. Machine learning applied to occupancy time series can supply what sensors alone cannot.
Building a Lighting Occupancy Model
A useful lighting model needs zone-level occupancy time series, usually combining motion logs, BMS occupancy inputs, and available calendar data. Two to four weeks covering representative days and times can reveal stable patterns: zones reliably occupied in specific windows, and zones with variable or low use during nominal occupied hours.
The model does not simply label every time slot occupied or empty. It produces probabilities. Zone 6 on the third floor might show 85% occupancy Monday through Wednesday from 8am to 5pm, 25% on Thursday and Friday afternoons, and near-zero on Saturday. Those probabilities guide schedule changes. The occupied-light threshold should exceed 50% to avoid people entering a dark zone and waiting for motion detection, but remain below 90% to produce useful changes.
For most commercial offices, occupied-mode scheduling often uses a 60-70% probability threshold. Below it, the zone uses reduced lighting or sensor-only control, while motion sensing covers brief visits.
The Schedule Update Workflow
The benefit is not a single schedule revision. Occupancy changes with tenants, seasons, and building use. Refitting the model every two to four weeks keeps schedules aligned with current conditions instead of allowing year-over-year drift.
Operationally, the system reviews two to three weeks of occupancy data, recommends a schedule for each zone, marks proposed changes for facilities review, and applies approved updates in the lighting controls or BMS. Staff review a supported summary rather than investigating every zone. Each review cycle takes a few minutes.
Schedule changes are intentionally conservative. The system will not remove occupied-mode lighting unless occupancy probability stays below a low threshold across several observation periods. It targets persistent drift, not every possible lighting minute.
Lighting and HVAC Scheduling
Lighting and HVAC occupancy models work together. A zone scheduled as unoccupied for lighting generally does not need full HVAC setpoints. Its occupancy probabilities feed the HVAC scheduling engine, which decides between conditioning and setback.
Coordination can extend the energy effect beyond either system alone. A zone with 30% occupancy probability on Friday afternoons can use lighting setback and HVAC setback in the same window. Their electrical reductions in that zone over a typical month are additive. Across a portfolio, large floors with recurring low-occupancy periods are where that combined reduction accumulates.
What This Does Not Cover
Occupancy modeling does not remove the need for manual corrections during unusual events. A zone reconfiguration, tenant move-out, or building event that changes normal use may require a manual override or retraining trigger. The model detects gradual drift well, but abrupt one-time changes require an adjustment to the training window.
For routine drift, such as schedules misaligned after 2-3 years of occupancy change, automated modeling is the efficient route to accurate schedules. Emergency events and unusual weeks still call for manual override. The two tools serve different situations.