The Occupancy Data Problem Nobody Wants to Admit
Every smart building vendor claims accurate occupancy data. Almost none of them deliver it. The reason is structural: single-source occupancy sensing is fundamentally unreliable in commercial environments. PIR motion sensors miss stationary occupants. Wi-Fi counting drifts with device behavior changes. CO2-based estimation fails in spaces with high ventilation rates. Each technology has a failure mode that corrupts the data every downstream optimization depends on.
BEAST has analyzed occupancy data quality across dozens of commercial deployments in APAC. The consistent finding is that single-source occupancy accuracy in real-world conditions ranges from 55-75% at zone level. This means your AI-HVAC system is making control decisions based on data that is wrong a quarter to nearly half the time.
The 4-Layer Fusion Architecture
The solution is not a better single sensor — it is sensor fusion. The 4-layer model combines four independent occupancy signal sources, each with different strengths and failure modes, into a unified occupancy estimate that is more accurate and resilient than any individual source.
Layer 1 — HVAC Telemetry: supply and return air temperature differentials, VAV box positions, and chilled water flow rates provide indirect occupancy signals based on thermal load. Always available but slow to respond and confounded by solar gain. Layer 2 — CO2 Concentration: metabolic CO2 correlates with occupant count and activity level, providing good density estimation but lagging occupancy changes by 15-30 minutes. Layer 3 — Wi-Fi/Network Analytics: connected device counts provide real-time presence detection with good spatial resolution but are affected by device behavior variations. Layer 4 — Motion/PIR/Computer Vision: active motion detection provides instantaneous presence confirmation but misses stationary occupants.
How Fusion Eliminates False Positives and Negatives
The power of fusion is cross-validation, not averaging. When Wi-Fi reports 30 occupants but CO2 shows ambient levels, the fusion engine flags a likely false positive from persistent device connections. When PIR reports no motion but HVAC shows elevated thermal load, the engine recognizes stationary occupants that motion sensors miss. Each layer compensates for the others' failure modes, producing fused estimates with demonstrated accuracy of 85-95% at zone level in production environments.
The fusion algorithm assigns dynamic confidence weights to each layer based on environmental conditions. During morning arrival, Wi-Fi and motion receive higher weights for speed. During steady-state afternoon occupancy, CO2 and thermal data receive priority for density accuracy. These weights are learned from historical data and continuously updated.
Implementation Without Rip-and-Replace
Most commercial buildings already have three of the four layers deployed — just not connected. HVAC telemetry exists in the BMS. Wi-Fi infrastructure exists for connectivity. Motion sensors exist for lighting control. Only CO2 sensing may require additional deployment, and modern wireless sensors install at $200-400 per zone with no wiring. The integration challenge is a software problem — extracting occupancy-relevant signals from each system, normalizing them to a common temporal and spatial resolution, and running the fusion algorithm in real-time.
The Downstream Impact
Accurate occupancy data is not the end goal — it is the foundation. HVAC optimization saves an additional 8-12% when operating on fused occupancy versus single-source data. Space utilization analytics become actionable when accuracy crosses the 85% threshold. Demand-controlled ventilation achieves its designed efficiency only when occupancy inputs reflect reality. The 4-layer fusion model is not a sensor project — it is the data quality investment that unlocks the full ROI of every other smart building initiative.