Your BMS already knows how to drive. But it still needs a mechanic.

There's a pitch gaining ground in our industry: that the next step for a Building Management System is predictive control. Pull in Bureau of Meteorology forecasts, add occupancy sensors, and let the system shift operating hours and adjust setpoints before conditions change. The message to owners is that with smart enough control, you won't need analytics on top. The building will look after itself.

I want to challenge that. Not on AI, and not on the BMS teams who keep buildings running every day, but on where owners are being asked to spend their money.

Most buildings still don't get the basics right. Predictive control doesn't fix that. It schedules on top of it. What owners need first is a BMS that responds well to change as it happens and holds its specified control bands. Then they need analytics to expose what the BMS is quietly compensating for, so it gets fixed rather than masked.

The predictive control pitch

On paper, it makes sense. If the BOM says tomorrow hits 38°C, pre-cool the building overnight. If the occupancy sensors show level 12 is empty on Fridays, trim its start time. If the system learns how the building responds, it can move setpoints ahead of the load instead of chasing it.

Some of that has value, in the right building. The problem is the assumption underneath it: that the plant will do what it's told. Predictive control decides when and how hard to run the plant. It doesn't check whether the plant is healthy enough to follow through.

Pre-cool a building through a chilled water valve that leaks by, and you don't save energy. You just start wasting it earlier. Trim a start time on a floor where heating is fighting cooling, and the fight runs a little shorter, but it still runs every day. Prediction layered over an untuned building delivers a more sophisticated version of the same waste.

That's why I think it's the wrong place to start spending. A predictive overlay is a line item on a quote, and it looks like progress in a board paper. Tuning a building is slower and less glamorous: walking the trends, finding the loops that hunt, the valves that leak by, the schedules that drift. But that's where the money is.

Cruise control

Think about cruise control in a car. You set 100 km/h. The system doesn't predict the road. It doesn't model traffic or learn your habits. It measures current speed, compares it to the setpoint, and modulates throttle to close the gap. On a flat road it barely touches the accelerator. On an incline it adds fuel. Cresting a hill, it coasts. Intelligence isn't in prediction. It's in knowing when not to act.

A BMS controlling HVAC works the same way. Zone setpoint is 22°C. The system reads the space, compares it to setpoint, and modulates chilled water valves, VAV dampers, fan speeds and compressor staging to close the gap. When the building sits comfortably inside its deadband, a well-designed BMS does nothing. That “nothing” is where efficiency lives.

A good BMS is very good at this. It reacts to the conditions in front of it and brings the building back to target, quickly and safely. That's its job, and it's why almost every larger building has one. Like any system, it performs best when a skilled BMS team keeps it tuned as tenants, equipment and uses change. Planes with autopilot still have pilots, engineers and ground control.

Most buildings don't need prediction. They need the basics.

Most underperforming buildings I walk into aren't short on predictive intelligence. They're over-acting. Simultaneous heating and cooling. Economy cycles disabled because of a complaint two summers ago. Static pressure resets that never reset. Chilled water setpoints locked at design conditions year-round.

The car has its foot on the accelerator and the brake, and we're asking whether it should also forecast the weather.

Energy efficiency in HVAC comes down to two levers: how often the plant runs, and how hard it works when it does. The strategies that move them are well known:

• Optimised start/stop delays plant start until space conditions actually need it. The building coasts longer.
• Supply air temperature reset lifts the SAT when zones don't need cold air. The plant works less hard.
• Condenser water reset lets the cooling tower do more of the work when ambient conditions allow, so the chiller does less.
• Static pressure reset trims fan energy by responding to the most-open VAV damper. The fan works just hard enough.
• Economy cycles let outside air do the cooling when it's free. The chillers coast.
• Demand-based ventilation ties outside air to actual occupancy via CO₂. Empty rooms stop being ventilated.

None of these are predictive. All of them are reactive, in the proper sense of the word. A properly tuned PID loop (the feedback loop that modulates each valve and damper to hold its setpoint) will outperform a sophisticated overlay sitting on a poorly commissioned plant every time.

A 5-star NABERS rating isn't waiting on predictive control. It's waiting on these strategies being commissioned, tuned and kept that way.

The BMS is the driver. It isn't the mechanic.

Cruise control will hold 100 km/h with the handbrake half on and never tell you. It will just burn far more fuel doing it, and the dashboard won't show a thing. The cruise control is doing its job. The car has a problem that needs to be fixed.

A BMS does exactly the same. Every scan cycle, it reads inputs, compares them against setpoints and writes outputs. When something disturbs the building, it works around the fault to get back to setpoint, which is exactly what a control system should do. So it holds 22°C while quietly compensating for a leaking valve, a stuck damper or heating fighting cooling. From its perspective, the setpoint is held. Mission accomplished.

Predictive control doesn't change this. A smarter driver still can't feel the handbrake.

I was working on a site where the chilled water system was underperforming: branches running warm, zones complaining, energy bills climbing. Working from what the head-end showed, the logical diagnosis was that the system needed chilled water re-balancing. Anyone looking at the same screen would have reached the same conclusion, and quotes were prepared on that basis.

The trends told a different story. A significant number of chilled water valve actuators on the air handling units have failed or reversed; 100% Open meant they were closed. The BMS had no way of knowing. As far as it could tell, the control loop was commanding the actuator correctly.

Re-balancing would have redistributed the imbalance, masked the fault and locked in inefficient operation for years. The real fix was to fix or replace the failed actuators, then letting the existing controls do their job.

That diagnosis didn't come from BMS. It came from AI-driven analytics across the BMS data, asking questions the BMS was never designed to ask:

• Is this hot water valve commanded to 0% while the supply air temperature sensor downstream is still gaining temperature? It's leaking by.
• Is this AHU heating and cooling at the same time a couple of times a day? The control sequence needs tuning.
• Is this VAV box pinned at maximum airflow with the zone still off setpoint? Something upstream is starving it.
• Is this chiller's kW/ton drifting up over weeks? Tubes are fouling, or the refrigerant charge is off.
• Are two identical Chillers on identical loads showing 15% energy variance? One of them has a problem.

None of these are control problems. They're diagnostic problems, and they're exactly what stops a well-designed control strategy from delivering. Every strategy assumes the plant does what it's commanded to do. Take that away and the whole control hierarchy, predictive or not, is operating on a false assumption.

Motorsport makes the split obvious. A race car's management system adjusts in real time to hold its design settings. It doesn't work out why the car was slower in the wet. That's the data team's job. Nobody in the garage would argue that one makes the other unnecessary. A building is the same, except its conditions keep changing for twenty years, not a one race weekend.

What this means in practice

The control strategy matters more than the platform. Two BMS from different vendors will both control a building well if the sequences of operation are tight, and both will struggle if they're loose. Strategy design comes first.

Commissioning doesn't end at handover. Every tenancy change, fit-out, equipment replacement and schedule override moves the building a little further from how it was commissioned to run. In practice, continuous commissioning is analytics: watching the building's data every day and flagging the moment something drifts.

Historical data isn't optional. Many faults are intermittent: a valve that only leaks by on hot afternoons, or a sequence that only fights itself during morning start-up. They look healthy whenever someone happens to check the head-end. A good analytics platform collects data directly from the BMS controllers and builds the record itself, rather than relying on whatever trends were set up at handover.

A finding only counts once it's fixed. A fault flagged with no one to own it is just a more expensive dashboard. Each issue needs someone accountable for closing it out, and a check that the fix actually held. That's what stops a fault being masked a second time.

A well-controlled building looks boring. Plant off, or coasting. Setpoints held. Energy not being spent. That's the win.

The point

Cruise control isn't impressive because it predicts the road and conditions ahead. It's impressive because it holds a setpoint without you thinking about it, and lets the car coast when the road allows. A BMS is the same machine, scaled up to a building. It already knows how to drive.

What it needs isn't a predictive co-pilot. It needs a mechanic: analytics watching the trends and catching the leaking valves, stuck dampers and drifting setpoints the BMS was never designed to notice, so the right people can fix them.

So the case is simple. Get the basics right. Let the BMS control to its bands. Use analytics to expose the inefficiencies underneath.

The energy savings and ratings follow from that.

‍