Building analytics alerts do not optimise buildings. People do. A platform that fires hundreds of automated alerts a month does not make a building perform better. It just moves the analysis work into a facilities manager’s inbox and waits. The result is alert fatigue: so much noise that teams stop reading the list. What actually lifts a NABERS rating or cuts energy waste is a tighter workflow: AI that detects anomalies, a filter that strips out the false positives, a human expert who validates what is real, and a continuous commissioning programme that turns confirmed issues into action. For buildings across Australia and New Zealand, that human-in-the-loop model is the difference between data and decisions.
In this article
- What is alert fatigue in building analytics?
- Why do automated building alerts create noise instead of action?
- Isn’t AI supposed to solve this?
- What actually optimises a building?
- Raw alerts vs human-reviewed AI: a side by side
- How BraveGen does building optimisation differently
- Frequently asked questions
What is alert fatigue in building analytics?
Alert fatigue is what happens when a monitoring system produces so many low-value notifications that the people receiving them stop trusting it and tune it out. The concept started in hospitals, where staff became desensitised to constant device alarms, and it now shows up everywhere from cybersecurity to industrial maintenance. In buildings, it is the quiet failure mode of fault detection and diagnostics (FDD).
A recent example makes it concrete. A facilities contractor forwarded us a run of around 20 automated alerts that had landed in his inbox overnight from a building analytics platform. His request was not for better analysis. It was simply to be taken off the email list, because the volume was clogging his inbox and he could pick things up through the portal anyway. That is alert fatigue in a single email, and it carries a hidden risk: when alerts fire straight to a contractor, work can get raised and actioned without the property manager ever knowing it happened.
Process industries have a benchmark for this. The EEMUA 191 guideline suggests no more than one alarm per ten minutes per operator during steady operation. Most building platforms blow past that within weeks of going live.
Why do automated building alerts create noise instead of action?
The mechanics are well understood. A single physical fault, say a stuck valve or a failed damper, often trips several rules at once across every sensor sitting downstream of it. Ten related alarms for one root cause read as ten problems on a screen. The operator ends up doing the correlation work the system should have done, every time, until they stop reading the list.
Then there are false positives. A pump that spikes for a moment on startup does not need an alarm. Weather swings, seasonal shifts, and normal operational variation all generate deviations that look like faults but are not. When a platform flags all of them, teams learn that most alerts are safe to ignore, which is exactly when the one that mattered slips through. The cost is real and specific: a $1,000 contractor callout to chase a $20-a-year problem, while a genuine baseload creep quietly adds tens of thousands of dollars to the annual energy bill.
The root cause is not too many alerts. It is unqualified alerts. A detection layer without a diagnostic and validation layer pushes the interpretation work onto a team that does not have time for it. You can read more on how fault detection and diagnostics is meant to work in the smart buildings community’s buyer guides.
From: XXX
Date: Monday, July, 2026 at 08:14
To: XXX
Subject: FW: XXX assigned to you a new action at XXX
Hey,
I’ve had 25 of these just overnight. Too much for me to look at. Can you remove me from the notifications?
Thanks,
XXX
Isn’t AI supposed to solve this?
Detection is the easy part. Modern analytics can scan millions of data points a day and surface anomalies a rules-based BMS would never see: baseload shifts, performance drift, erratic cycling. But detection on its own is what produces the flood. AI that fires every anomaly straight to a human is not reducing the workload, it is generating it faster.
The missing layers are qualification and validation. Qualification means contextualising each anomaly against weather, season, occupancy, and normal operation so only genuine issues surface. Validation means a person who knows the building and the portfolio checking that the issue is real, assessing its actual impact, and deciding whether it warrants action. In an ANZ context that expertise is not generic. It understands NABERS and NABERS NZ scoring, Green Star Performance, and GRESB submission windows, so the alerts that reach you are ranked by the outcomes you are actually measured on.
What actually optimises a building?
Four things working together, in this order: data, AI, human expertise, and continuous commissioning.
Data gives you the picture. AI finds the anomalies at a scale no person could. Human expertise decides what is real and what matters. And continuous commissioning (CCx) is the programme that keeps the loop running year-round rather than at quarterly review. That last piece is the one most platforms leave out. Buildings do not wait for a scheduled review to drift. Consumption patterns shift after tenant changes, sensors drift, schedules fall out of sync with occupancy, and one system quietly compensates for a fault in another. By the time a quarterly BMS review catches it, weeks of waste have already accumulated and a NABERS rating has already slipped.
The maths is blunt. An HVAC fault caught after three days wastes three days of energy. The same fault caught after six weeks wastes 42 days of energy and drags down your rating. The fix costs the same either way. The losses do not.
Raw alerts vs human-reviewed AI: a side by side
The difference between a fully automated feed and a human-reviewed programme is not subtle once you see it on the criteria that matter to a facilities or asset team.
| What matters | Fully automated alerts | Human-reviewed AI + CCx |
|---|---|---|
| Alert volume | Hundreds of raw alerts a month | Triaged to roughly 5 to 10 actionable ones |
| False positives | High, teams chase self-resolving anomalies | Minimal, a proprietary ANZ training library filters noise |
| Who reviews it | No one, before it hits your inbox | Engineers who know your buildings |
| Cost of misreading | $1,000 callouts for $20-a-year problems | Low-value issues closed internally, high-value ones prioritised |
| Time to resolution | Weeks, as alerts queue and get processed manually | Hours, with expert validation and handoff to operations |
| Oversight | Work can be raised without the property manager knowing | Validated, tracked, and reported back to owners monthly |
| Confidence | Uncertainty about which alerts matter | Audit-proven, assurance-grade confidence in every alert |
How BraveGen does building optimisation differently
BraveGen’s approach is deliberately human-led. Clive AI detects anomalies across more than four million data rows a day, then a proprietary qualifier module, trained across hundreds of ANZ sites, strips out the seasonal and operational noise. Only then does a BraveGen engineer review the qualified alert, add site knowledge, and confirm the real impact before it reaches your team. Alerts are issued within 48 hours of an anomaly, already carrying the context, the estimated financial impact, and the recommended next step.
Around that sits Continuous Commissioning: ongoing monitoring, monthly NABERS NZ tracking, quarterly site visits, contractor liaison, and capital planning support. It is the year-round programme that turns a validated alert into a fixed fault and a defended rating, rather than another line in an inbox. This is what moves a team, in Centuria New Zealand’s words, from admin to action.
“Working with BraveGen has improved how we manage these buildings and delivered great results. They’ve helped us shift from reactive to proactive management driven by data, visibility, and accountability.”
Dean Davies, Senior Property Manager, Bayleys
The results follow the model, not the alert count: $3 million in verified energy savings at Bayleys, a 49% energy reduction at ASB, and 11.73 GWh saved across Auckland Council’s portfolio. Continuous commissioning clients see an average 8.3% year-on-year energy reduction. And because BraveGen was built for Australia and New Zealand from day one, every rating, factor, and report is aligned to NABERS, NABERS NZ, Green Star, and GRESB rather than configured for somewhere else and hoped for the best.
Frequently asked questions
What is alert fatigue in building management?
Alert fatigue is when a building analytics or FDD platform generates so many low-value notifications that facilities teams stop trusting and reading them. Genuine faults then get missed among the noise, so problems that should have been caught in days run for weeks, wasting energy and eroding ratings.
Do AI building analytics platforms reduce false alarms?
Only if they include a qualification and human validation layer. AI detection on its own tends to increase alert volume, because it surfaces more anomalies faster. False alarms fall when anomalies are contextualised against weather, season, and normal operation, and when a person confirms the issue is real before it is escalated.
What is continuous commissioning (CCx)?
Continuous commissioning is an ongoing service that monitors, detects, acts, and verifies year-round, rather than tuning a building once or reviewing it quarterly. It catches issues between traditional review cycles, so a NABERS NZ rating is managed every month instead of discovered at submission time.
Is BraveGen’s Clive AI fully automated?
No. Clive AI detects and qualifies anomalies automatically, but every qualified alert is reviewed by a BraveGen engineer who knows the building before it reaches your team. This human-in-the-loop step is what makes the alerts relevant, high confidence, and audit-proven.
How does human-reviewed AI improve a NABERS rating?
By catching performance issues in days rather than months, and by prioritising the faults that actually move the rating. Continuous commissioning clients typically see about a 0.5 star improvement in the first twelve months, because problems are fixed before they compound into a submission-time surprise.