Building management systems (BMS) have been collecting sensor data for decades. Temperature, CO2, occupancy, energy meters — commercial buildings have been instrumented for a long time. The problem was that most of this data sat in proprietary BMS controllers that didn’t talk to each other, couldn’t be integrated without expensive middleware, and certainly didn’t do anything intelligent with what they collected.

Edge computing is changing the architecture of smart buildings, and it’s doing it in a more practical way than the “smart building” hype of a decade ago suggested. The change isn’t about futuristic AI running the building — it’s about processing sensor data closer to where it’s generated, making faster control decisions, and integrating building systems that historically operated in silos.

Why Edge, Not Cloud, for Building Control

The fundamental constraint in building automation is latency and reliability. A cloud-dependent HVAC control loop that loses internet connectivity means loss of automated building control — unacceptable for a hospital, data centre, or office building with several hundred occupants. The latency of a cloud round-trip (even 50–100ms) is also too slow for some real-time control applications, like demand-response actions that need to activate within seconds of a grid signal.

Edge computing for buildings means local processing power — typically an industrial gateway or small server co-located in the building’s comms room — that handles real-time control logic locally, with cloud connectivity used for monitoring, analytics, and configuration rather than control.

The practical architecture:

Sensors (occupancy, CO2, temp, metering)
        ↓ (BACnet, Modbus, MQTT, wired M-Bus)
Edge Gateway / Local Controller
        ↓ (runs control logic locally)
BMS/Building Controller (HVAC, lighting, access)
        ↓ (cloud sync: non-critical, high latency OK)
Cloud Platform (analytics, reporting, remote config)

The edge layer handles the control-critical workloads; the cloud handles everything else. This gives you reliability and low latency at the control layer, plus cloud-scale analytics without the latency risk.

Occupancy Sensing and HVAC Optimisation

The highest-ROI application of smart building IoT is occupancy-based HVAC control. In most commercial buildings, HVAC systems run to fixed schedules: heating/cooling starts at 7am, ends at 7pm, regardless of whether anyone is in the building. In a hybrid working world, this means conditioning a building for 500 people when 80 are actually there, from 9am to 3pm.

Edge AI makes dynamic occupancy-based control practical. A network of CO2 sensors and PIR (passive infrared) motion detectors, feeding into an edge model that estimates occupancy density per zone, allows the BMS to condition only the zones that are actually occupied — and to respond in near real-time as people move through the building.

CO2 as a proxy for occupancy: CO2 levels rise predictably with occupancy. At 400ppm (outdoor baseline), a zone is empty. As occupancy increases, CO2 rises. By correlating CO2 readings across zones with known floor plan occupancy capacity, an edge model can estimate occupancy density without cameras, preserving privacy while still providing useful control inputs.

Demand-controlled ventilation (DCV) is the HVAC term for CO2-based airflow control. It’s not a new concept, but implementing it properly requires local processing: a PID control loop running on the edge gateway that adjusts air handling unit damper positions based on real-time CO2 readings across multiple zones. The cloud can’t do this reliably — the control loop needs to run at 1–5 second intervals locally.

Energy savings from occupancy-based HVAC control in commercial buildings are well-documented: 20–40% HVAC energy reduction is achievable in buildings with significant occupancy variation, which in 2026 means most office buildings with a hybrid working population.

BACnet and Modbus Integration

The reality of most commercial buildings is that the existing BMS is running BACnet over IP or RS-485, Modbus, or some proprietary protocol from the BMS manufacturer. Any edge computing layer has to integrate with this installed base.

BACnet (Building Automation and Control Networks) is the dominant open standard in commercial HVAC. Most modern HVAC controllers, air handling units, and chillers support BACnet/IP or BACnet MSTP. An edge gateway with BACnet stack libraries (Node-RED has BACnet nodes; dedicated gateways from vendors like Crestron, Siemens, and Delta Controls handle this natively) can read setpoints, sensor values, and alarm states, and write control outputs back to BACnet controllers.

Modbus TCP covers a lot of the older installed base — particularly energy meters, power monitors, and older building sub-systems. The edge layer aggregates Modbus data alongside BACnet, normalises it into a unified time-series format, and applies analytics and control logic uniformly.

MQTT is increasingly the bridge protocol for new sensor installations. CO2 sensors, occupancy counters, and smart power plugs from modern IoT vendors publish to MQTT topics. An edge MQTT broker (Mosquitto, EMQX) sits on the gateway, subscribes to sensor data, and feeds it into the control logic alongside BACnet and Modbus reads.

The unification of these protocols at the edge — rather than at the cloud — is one of the practical advantages of the edge architecture. Cloud platforms that aggregate building data typically require internet connectivity for every read; an edge layer stores locally and syncs when connectivity is available, giving you historical data even during outages.

Fault Detection and Diagnostics at the Edge

Beyond control, edge AI is increasingly used for HVAC fault detection and diagnostics (FDD). HVAC faults — failing damper actuators, refrigerant leaks, heat exchanger fouling, stuck valves — are expensive to detect and fix reactively but relatively easy to detect proactively if you’re monitoring the right signals.

An edge model trained on normal HVAC operating patterns for a specific system can flag anomalies: a variable air volume box that’s drawing more actuator power than expected (actuator starting to fail), a chiller running at reduced efficiency compared to its baseline curve (refrigerant loss or fouling), a cooling coil that’s not meeting setpoint despite full valve opening (blocked coil or control valve failure).

This isn’t cloud AI — it’s a relatively simple anomaly detection model running on the edge gateway, comparing real-time sensor readings against the system’s normal operating curve. The compute requirements are modest; a Raspberry Pi 5 or industrial-grade equivalent handles it comfortably.

The outputs go to the facilities management team as prioritised work orders, not to a cloud dashboard that nobody looks at. That integration with existing CAFM (computer-aided facilities management) systems, via email alerts or API, is what makes FDD actionable rather than just informative.

The Business Case

For a typical 5,000m² commercial office building:

  • Occupancy-based HVAC control: 25–35% HVAC energy saving = £8,000–15,000/year (at current UK commercial electricity rates)
  • Fault detection: reduced reactive maintenance costs, 15–25% reduction in unplanned HVAC downtime
  • Edge gateway hardware: £2,000–8,000 depending on scale and integration complexity
  • Commissioning and integration: £5,000–20,000 depending on existing BMS complexity

Payback periods of 18–36 months are realistic for occupied commercial buildings with variable occupancy. For buildings with high energy costs or cooling loads (data centres, laboratory spaces), the ROI compresses considerably.

The smart building opportunity isn’t about replacing your existing BMS — it’s about adding an edge intelligence layer that makes the existing infrastructure perform better, using the sensor data the building is already generating.