Predictive Maintenance for Buildings: What Works, What's Hype and What It Costs

Predictive maintenance for buildings uses equipment data to spot problems before they cause breakdowns, so repairs can be planned instead of rushed. The proven part today is fault detection, which finds problems already happening and typically saves energy. Genuinely predicting failures is newer and patchier, and depends on good data, integration and people to act.

This guide explains the maintenance strategies, the difference between fault detection and failure prediction, what the independent evidence shows on savings and costs, where lifts and HVAC stand, the risks — including cyber — and what Dubai's rules require of owners and management companies.

Key takeaways

• Fault detection is the proven part. A US national lab campaign across 6,500 buildings measured median energy savings of 9% from fault detection tools, with about a two-year payback.
• Buildings generate a lot of faults. In the largest study of its kind, fault detection tools reported an average of 245 faults per building per month, and on any given day 40% of air handling units had one.
• Failure prediction is still early. A 2023 review found HVAC fault prognosis "still in its infancy"; the most promising regional result, from one Riyadh office building, is a back-test, not a live trial.
• Lifts are the most connected equipment. KONE had 44% of its service base connected by June 2026 and Otis about 1.1 million units — but their outcome figures are their own.
• Dubai is formalising maintenance. Law No. 3 of 2026 requires quality and safety certification for buildings 20 or more years old, and workplace lifts must be certified by an accredited third party every year.

What is predictive maintenance in buildings?

It's one of four ways to look after building equipment, and the best-run buildings mix them rather than choosing one.

Strategy	How it works	Trade-off
Reactive	Fix equipment after it fails	Cheap until it isn't: outages, overtime and knock-on damage
Preventive	Service on a fixed schedule or run hours	Reliable, but includes work that wasn't needed
Predictive (condition-based)	Act when measured condition shows early signs of trouble	Needs sensors, data, analysis and people to act
Reliability-centred	Choose the strategy per asset, based on how critical it is	Some trivial assets are best run to failure

Pacific Northwest National Laboratory, drawing on US federal guidance, puts the typical industry mix at 40–60% reactive, 30–50% preventive and 10–15% predictive work, against a best-in-class mix of under 10% reactive, 25–35% preventive and 45–55% predictive. The standards are maturing too: ISO published the third edition of its prognostics standard, ISO 13381-1, in 2025.

Definition

Fault detection and diagnostics (FDD) — software that analyses building management system data to find faults that already exist, such as a stuck valve, a drifting sensor or a schedule running out of hours, and suggests the likely cause. Prognostics go a step further and estimate when a component will fail. Most "predictive maintenance" in buildings today is FDD.

What's the difference between fault detection and failure prediction?

Fault detection tells you something is wrong now; prediction tells you something will go wrong later. The first is common and measurable. The second is where most of the hype sits.

From building data to a planned repair

1. 01Collect
   • BMS points
   • Meters and sensors
   • Work-order history

   Data quality decides everything

2. 02Detect
   • Rule-based checks
   • Machine-learning models
   • Alert thresholds

   Finds faults that exist

3. 03Diagnose
   • Likely cause
   • Energy or comfort impact
   • Priority

   Turns alerts into tasks

4. 04Predict
   • Remaining useful life
   • Failure risk
   • Timing of repair

   Still early for HVAC

5. 05Act
   • Planned work order
   • Parts and access
   • Verify the fix

   Where savings are realised

The largest empirical study of building faults shows why detection alone is a big job. Lawrence Berkeley National Laboratory analysed multi-year data from more than 60,000 pieces of HVAC equipment covering over 90 fault types. It found an average of 245 reported faults per building per month; on any given day, 40% of air handling units and 30% of terminal units had a reported fault. Sensors were among the components most commonly affected, followed by valves and dampers — which is why calibrating sensors comes before any advanced analytics.

Prediction is less mature. A 2023 review of data-driven fault detection hosted by the same lab concluded that fault prognosis for building HVAC systems is still in its infancy, and that real-building deployment, benchmarking and transferability remain major challenges.

Does predictive maintenance work, and what does it cost?

The strongest independent evidence covers energy savings from fault detection; evidence on avoided breakdowns is thinner.

What the data shows

What the evidence shows

• Energy savings: Lawrence Berkeley National Laboratory's campaign with 104 organisations and 6,500 buildings documented median annual energy savings of 9% from fault detection tools and 3% from energy information systems, with about a two-year simple payback.
• Cost: the median cost of fault detection tools was about US$0.06 per square foot (US$0.65 per square metre) to set up, plus US$0.02 per square foot (US$0.22 per square metre) a year, in US data.
• Failure prediction, Gulf evidence: a study of one large Riyadh office building, using two years of one-minute data from chillers, pumps and air handling fans, found its model would have cut unplanned outages by 47.6%, downtime by 41.3%, HVAC electricity by 10.6% and total operating cost by 9.7% — in a counterfactual back-test, not a live deployment.
• Pilot reality: a 2026 Norwegian pilot on an air handling unit could run only 18 of 28 standard diagnostic rules because the building management system lacked key data points, and one model missed most cases of a rare fault until its training data was rebalanced.

The pattern is consistent: detection pays off when faults get fixed, and prediction works best where data is clean, equipment is critical and failures are costly. The older headline figures that circulate online — ten-times returns, 70% fewer breakdowns — come from industrial surveys summarised in 2010 US federal guidance, not from buildings.

Where do lifts stand?

Lifts are the most connected equipment in most buildings, because manufacturers have added remote monitoring to their service contracts.

• KONE reached more than 40% connectivity in its service base in 2025 and 44% of elevators by the end of June 2026. It claims 55% fewer entrapments, 80% of equipment faults identified proactively and 25% of issues solved remotely within two minutes, while noting that results vary and not all faults can be prevented.
• Otis ended 2025 with about 1.1 million connected units, including units under warranty, and says its Otis ONE service isn't compatible with all elevator controllers.

Those outcomes are the companies' own figures; no independent audit has been published. In Dubai, remote monitoring doesn't change the statutory duty: Dubai Municipality's guideline requires lifts and other lifting equipment used in workplaces, including hotels, to be tested and certified by an EIAC-accredited third party every 12 months.

What are the risks of predictive maintenance?

Five practical risks recur.

• Alert overload. Hundreds of faults a month per building is more than most teams can triage without prioritisation.
• Bad data. Faulty or uncalibrated sensors produce false alarms and missed faults.
• Overstated accuracy. Models can score highly overall while missing rare, expensive faults, and single-building studies don't transfer automatically.
• Lock-in. Monitoring tied to one manufacturer's controllers can make switching providers harder; ask who owns the data and how to export it.
• Cyber exposure. The US cybersecurity agency CISA has issued advisories on building systems, including Siemens OZW web servers rated 10.0 on its severity scale and Johnson Controls supervisory controllers rated 8.4, both exploitable remotely. Connecting equipment to analytics platforms widens that attack surface.

Common misconception

"Connected means predictive." A connected lift or chiller sends data; it doesn't automatically predict anything. Most value today comes from detecting and fixing existing faults. Ask a vendor what it detects, what it predicts, how it was validated and who acts on the alerts.

What do Dubai's rules require?

Dubai's recent laws are pushing buildings toward documented, data-backed maintenance.

Rule	What it requires	Why it matters for maintenance
Law No. 3 of 2026 (building quality and safety)	A Quality and Safety Certificate, based on a technical assessment, for buildings completed 20 or more years earlier; Dubai Municipality to build a digital management and maintenance system and a unified building database; fines from AED 100 to AED 1 million	Maintenance records and condition data become evidence for certification
Law No. 3 of 2026, jointly owned buildings	The management entity appointed under Law No. 6 of 2019 carries the owner's duties	Management companies own the compliance task
Law No. 4 of 2025 (Civil Defence)	Civil Defence regulates early-detection fire alarms and ensures buildings are connected to its approved electronic systems	Fire-system monitoring is mandated, not optional
Dubai Municipality lifting-equipment guideline	Third-party testing and certification by an EIAC-accredited body every 12 months for lifting equipment used in workplaces, including hotels	Monitoring supports, but doesn't replace, certification
RERA service charges	Management companies' budgets need RERA approval, and late approval means owners get instalment options	Maintenance technology in shared buildings goes through an approved budget

The UAE's Hassantuk programme, run by the Ministry of Interior, adds a national layer for fire safety, connecting buildings and homes to central alarm monitoring; it targets more than 500,000 buildings and private homes.

What this means

For Dubai owners and management companies, the case for predictive maintenance is shifting from "nice to have" to evidence: when certification and budgets depend on documented condition, the data that predicts faults also proves the building is being looked after.

How should owners and facility managers get started?

1. Start with fault detection on HVAC, where energy savings can be measured against a baseline.
2. Fix sensors and data points first. Calibrate the sensors that fault studies show fail most often, and fill gaps in building management system points.
3. Prioritise critical assets. Chillers, pumps, lifts, fire systems and main electrical boards justify monitoring; trivial equipment may not.
4. Budget for people. Someone has to triage alerts and close work orders, or savings never materialise.
5. Contract for data and security. Secure data access and export rights, network isolation and patching responsibilities.
6. Keep statutory inspections. Remote monitoring complements Dubai's certification rules; it doesn't replace them.
7. Measure outcomes: energy, unplanned downtime, work-order backlog and, in shared buildings, service charges.

For the wider technology picture, see smart buildings, autonomous smart buildings and digital twins in real estate, which covers linking equipment history to a building model.

Final takeaway

Predictive maintenance for buildings is real, but most of its proven value comes from detecting and fixing faults that already exist, not from forecasting failures. The independent evidence supports modest, measurable energy savings with short paybacks; bigger claims about avoided breakdowns mostly come from vendors or single-building studies. In Dubai, new certification rules and regulated budgets make good maintenance data valuable in its own right. Owners who fix their data, focus on critical assets and staff the response will get the benefits; those who buy dashboards alone will mostly get alerts.