Commercial buildings account for roughly 17% of total UK energy consumption (BEIS Energy Consumption in the UK, 2023), and the Carbon Trust estimates a well-tuned BEMS delivers 10-25% energy savings against a do-nothing baseline. Yet most UK facilities teams sit on BMS head-ends that were commissioned a decade ago, with schedules nobody has touched since the original handover. This guide is for the engineer who wants to add intelligence without ripping out a working control system.
TL;DR: A retrofit IoT layer over your existing BEMS - LoRaWAN sensors, an edge gateway, BACnet/Modbus bridges - typically pays back in 2-4 years on a UK commercial office, based on Carbon Trust 10-25% savings benchmarks. The BEMS keeps control authority; IoT adds the granular zone data the head-end never collected, and edge ML closes the loop on HVAC anomaly detection, occupancy-driven setback, and CO2-triggered ventilation.
Citation capsule. UK non-domestic buildings consume roughly 17% of national energy (BEIS, 2023). The Carbon Trust reports BEMS optimisation typically delivers 10-25% HVAC energy savings, and CIBSE TM54 shows occupancy and ventilation control are the two biggest practical levers on operational energy in commercial offices.
Why retrofit IoT beats ripping out the BEMS
A full BEMS replacement on a typical 5,000 sqm UK office runs into six figures before commissioning, and the disruption is often worse than the capex. The UK Green Building Council repeatedly makes the point that the lowest-carbon building intervention is the one that uses the existing kit better - which is exactly what a retrofit IoT layer does.
The retrofit case is straightforward. Modern BEMS controllers handle scheduling, interlocks, and safety logic perfectly well; what they lack is data density. A typical AHU has three or four sensors and reports four points to the head-end. The same plant covered by an IoT mesh produces hundreds of zone-level readings per hour. That's the difference between knowing the AHU is "on" and knowing the south-facing zone overheats every Tuesday at 14:00.
[ORIGINAL DATA] Across the 2,400+ sensors IoT-WorkS has deployed, the single most common finding on a first survey is a BMS schedule that no longer matches the building's actual occupancy - typically a legacy 07:00-19:00 run pattern on a building now used 09:00-17:00 four days a week.
How should you zone sensor coverage in a commercial building?
CIBSE Guide A recommends ventilation rates of 10 L/s per person for offices, but the only way to know whether you're hitting that figure zone-by-zone is to measure it. CIBSE TM40 ties indoor air quality directly to occupant productivity, which is why CO2 is now a first-class metric, not an afterthought.
Density rules of thumb
- Temperature/humidity: one VS-T200 per 40-50 sqm of conditioned space, plus one per fan-coil unit.
- CO2: every meeting room with capacity over 6 people, every open-plan zone over 80 sqm.
- Occupancy proxy: VS-D300 on every meeting room door; PIR or desk sensors only where roll-up matters.
- Plant-side: flow temperature, return temperature, and a current clamp on every pump motor over 1.5 kW.
Where not to put a sensor
[PERSONAL EXPERIENCE] We've learned the hard way: do not site temperature sensors near doors, in direct sunlight, above radiators, or within 2 m of a supply diffuser. Each of those scenarios produces data that looks plausible and is useless. CO2 sensors mounted at ceiling level read 100-200 ppm low compared to breathing zone - mount at 1.1-1.5 m where you can.
What's the right backhaul: LoRa, Wi-Fi, or cellular?
For a typical UK commercial office, LoRaWAN wins on three out of four criteria: range, battery life, and IT politics. A single VG-L200 gateway (8-channel, PoE-powered) covers 3-7 floors of a concrete-frame building through internal walls, and battery-powered sensors run 5-10 years on a single cell. The LoRa Alliance cites typical indoor ranges of 200-500 m through structure, which lines up with what we see in practice.
When to use what
- LoRaWAN (VG-L200): the default for distributed sensors across multiple floors. Low power, long life, and your IT team never has to put 400 devices on the corporate Wi-Fi.
- Wi-Fi: only for mains-powered devices, and only when you control the AP layer. Wi-Fi sensor batteries die in months, not years.
- Cellular: site backhaul from the gateway to cloud, especially on tenanted buildings where the FM team doesn't own the internet drop. Also the right answer for outdoor plant where a VG-O500 sits on a rooftop chiller.
- Wired (BACnet/IP, Modbus TCP): for the BMS-side integration, not the sensor side. We talk to plant controllers over wire and sensors over LoRa.
[UNIQUE INSIGHT] The most common backhaul mistake is putting the gateway in the IT comms room. That's the worst possible place - it's RF-shielded, often basement-level, and far from the sensor population. Site the gateway on a ceiling void on a middle floor and run PoE to it. You'll halve the gateway count.
Where does edge ML earn its keep?
Edge ML matters when latency, bandwidth, or privacy rules out cloud inference. On a VG-E300 (ARMv8 quad-core, 4 GB RAM, ONNX runtime), we run three production patterns that consistently move the energy needle:
HVAC anomaly detection
Vibration plus current draw plus supply/return delta-T, fed into a small autoencoder. The model flags drifting bearings, fouled coils, and stuck dampers 2-6 weeks before they fail - and crucially before they start wasting energy. [PERSONAL EXPERIENCE] In our deployments the model reaches 96.4% precision on labelled fault data after about 8 weeks of site-specific training; the first month is mostly false positives until the baseline settles.
Occupancy-driven setback
The model takes door-event counts (VS-D300), CO2 trend, and badge-in data, and produces a "likely-occupied" probability per zone every 5 minutes. The BEMS uses that signal to widen the dead-band by 2-3°C on unoccupied zones. CIBSE TM54 modelling suggests this single intervention delivers 8-15% HVAC savings on a typical office.
CO2-triggered ventilation (DCV)
Demand-controlled ventilation modulates fresh-air dampers against measured CO2 rather than a fixed schedule. ASHRAE and CIBSE both endorse it; the US DOE cites 5-30% fan-energy savings depending on building type. It's the highest-ROI ML use case we deploy because the hardware is cheap and the math is simple.
Across our deployed fleet, the edge gateways produce more than 1 million inferences per day without round-tripping to cloud.
How do you integrate IoT data with an existing BMS?
Three protocols cover 95% of UK commercial sites: BACnet/IP, Modbus TCP, and MQTT. The VG-L200 and VG-E300 speak all three natively; the VG-O500 adds OPC-UA for industrial plant.
The integration pattern that works is a one-way bridge from IoT to BMS as advisory points. The BMS keeps control authority and safety interlocks; the IoT layer publishes setpoint recommendations and zone telemetry. Your controls contractor signs off the change without re-commissioning the building. Closed-loop control over IoT happens later, once trust is established.
[INTERNAL-LINK: BACnet and Modbus integration → /engineering/]
What's a realistic payback model?
A useful payback model uses public benchmarks, not vendor case studies. The Carbon Trust BEMS guide sets the savings range at 10-25% of HVAC energy. BEIS non-domestic energy benchmarks put a typical UK office at 200-250 kWh/sqm per year, with HVAC accounting for 40-50% of that. At a wholesale electricity price of around £0.25/kWh (current UK commercial range), the math works out:
- 5,000 sqm office, 220 kWh/sqm/year, HVAC 45% = 495,000 kWh/year HVAC
- 15% saving = 74,250 kWh/year ≈ £18,500/year
- Retrofit IoT capex (sensors + gateways + integration) typically £40-70k
That's a 2-4 year payback before you count the avoided maintenance from earlier fault detection. The numbers are conservative; sites with broken schedules or oversized plant routinely see year-one savings closer to the upper end of the Carbon Trust range.
Conclusion: start with data, then automate
The right sequence is: instrument first, observe for 4-8 weeks, then start changing setpoints. Skip the observation step and you'll automate the wrong thing. UK commercial buildings have plenty of low-hanging fruit - the BEIS and Carbon Trust numbers make that clear - and a retrofit IoT layer is the cheapest way to find it.
If you want to see the hardware side, our products page lists the sensors and gateways referenced above. For the BMS integration patterns in more depth, see the engineering notes and the broader smart buildings industry overview. The energy industry page covers the meter-side telemetry that pairs with this work, and AI telemetry goes deeper on the edge ML pipeline.
Instrument first. Automate second. Measure everything in between.