Can predictive maintenance work for commercial property management?

Commercial buildings fail expensively. A chiller unit that breaks during a July heatwave costs $15,000–$40,000 to repair on an emergency basis, plus days of unhappy tenants and potential lease penalties. The same failure caught three weeks earlier, when the unit was showing early stress signals, costs $2,000–$5,000 to fix. That gap is what predictive maintenance is designed to close.

Predictive maintenance is not a new idea in manufacturing. Industrial plants have used it for two decades. What changed in the last few years is that commercial buildings now generate enough digital data from their own systems to run the same approach without buying expensive new hardware. The sensors are mostly already there. The question is whether anyone is reading them intelligently.

Not every building system fails the same way or gives the same warning signals. The three that produce the clearest data patterns, and therefore the strongest candidates for predictive models, are HVAC, plumbing, and elevators.

HVAC represents 40% of a commercial building's energy bill and generates a dense stream of data: supply air temperature, return air temperature, refrigerant pressure, compressor amperage, fan speed, and filter differential pressure readings. When a compressor is starting to fail, it draws more current than normal while delivering less cooling. That pattern can be visible in the data 30–45 days before the unit stops working (Jones Lang LaSalle, 2021).

Plumbing faults are subtler but often more damaging. A slow pipe leak inside a wall can go undetected for months and cause tens of thousands in water damage before anyone notices. Flow meters and pressure sensors at the building's main risers can detect the kind of consistent low-level pressure drops that indicate an active leak. Buildings with sub-metered water monitoring catch leaks on average 11 days earlier than those relying on visual inspection alone (McKinsey, 2021).

Elevators are among the most sensor-rich systems in any building. Motor current, door cycle counts, vibration on the drive shaft, and leveling accuracy are all logged electronically already in modern elevator controllers. Most of that data goes unused. Predictive models that read elevator controller logs directly have shown 70% fewer unplanned outages in commercial deployments (Otis Digital, 2020).

The mechanism is pattern recognition on time-series data, which sounds technical but works like this: the model learns what normal looks like for each piece of equipment in your specific building, then watches for deviations from that baseline.

For HVAC, the system collects 12 to 18 months of sensor readings to build a normal operating profile. It learns that the chiller runs at 85% compressor load on a 90-degree afternoon, that refrigerant pressure sits between certain bounds at different outdoor temperatures, and that the supply fan draws a predictable amount of current at each speed setting. Once that baseline is established, the model flags anomalies: a compressor drawing 12% more current than it should at a given load, or refrigerant pressure trending down by 0.5 PSI per week.

Those numbers alone do not tell you the failure is imminent. The model combines multiple signals, compressor current, supply temperature differential, and refrigerant pressure trend, and calculates a risk score. When the score crosses a threshold, a work order is automatically created and assigned to a technician. The technician shows up with the right parts because the model has already identified the most likely failure mode.

For plumbing, the approach is simpler. The model sets a baseline water consumption pattern by hour of day and day of week. A consistent 3% deviation from that pattern at 2 AM, when the building is empty, is almost always a leak. The model alerts the facilities team before the water reaches drywall.

A real estate technology study by Deloitte in 2021 found that buildings using this type of anomaly detection resolved 60% of flagged issues before they caused any service disruption to tenants. The model is not magic. It is statistics applied consistently to data that property managers have been collecting but not using.

More than most property managers realize. A standard building management system installed in any Class A or Class B commercial property built after 2010 already logs dozens of data points per hour:

The gaps are usually in older Class C properties or buildings that never connected their BMS to a network. For those, adding the necessary sensors costs $8,000–$15,000 per building, depending on floor count and system age. That is a one-time infrastructure cost, not an ongoing expense.

The bigger issue is that data from different systems often lives in separate, disconnected software. The BMS logs to one vendor's platform. The elevator controller uploads to another. The utility meter feeds a third dashboard. A predictive maintenance model needs all of this data in one place. Building a data pipeline to connect these sources typically takes 6–8 weeks and is the single largest time investment in any property management predictive project.

Five buildings. Twenty buildings. Fifty. The answer is different at each scale, and most of the guidance you find in the market is written for REITs with 200 properties, not a regional portfolio.

For a portfolio of 5–15 properties, the economics work if you treat the model as a shared service across all buildings rather than a separate system for each one. A single predictive model trained on data from 10 buildings is more accurate than 10 separate models trained on one building each, because the training data is richer. HVAC units fail in similar ways regardless of which building they sit in, and the model learns faster when it has seen more failure examples.

Building this for a mid-size portfolio typically costs $20,000–$30,000 in initial development: data pipeline integration, model training, the dashboard where facilities managers see alerts, and the work-order connection to your existing maintenance software. A Western engineering firm would quote $80,000–$120,000 for the same scope. A cost-effective global engineering team with experience in property data delivers the same production-grade system for $20,000–$28,000.

The payback period shortens as the portfolio grows because the maintenance savings scale with the number of buildings while the development cost does not.

One honest constraint: the model needs 12–18 months of historical data to produce reliable predictions. If your BMS has been logging data but you have never exported it, that data is recoverable and can accelerate the timeline. If you are starting from scratch with new sensors, expect the first 6 months to be a calibration period where the model is learning your buildings rather than producing actionable alerts.

The numbers from commercial real estate are consistent enough to be useful for planning.

The U.S. Department of Energy published a building systems study showing predictive maintenance reduces maintenance costs by 25–30% compared to reactive repair schedules. The savings come from three sources: fewer emergency callout fees (emergency labor rates run 1.5–2x normal), parts replaced at planned prices rather than expedited-shipping prices, and equipment that lasts longer because failures are caught before they cause cascading damage.

For a portfolio generating $2 million in annual maintenance spend, a 25% reduction is $500,000 per year. Even at the conservative end of that range, 15% savings on $2 million is $300,000, which pays back a $30,000 development investment in about five weeks of savings.

Energy consumption is the less obvious savings driver. HVAC systems running with degraded refrigerant charge or dirty coils use 10–15% more electricity than they should. A predictive model that keeps equipment in optimal condition acts as a passive energy optimization layer. JLL's 2021 Decarbonization Report found that buildings with active equipment monitoring reduced energy costs by an average of 12% within the first year.

The risk that is harder to quantify is tenant satisfaction. A major HVAC failure in a leased office building triggers lease review clauses in most commercial leases. One avoidable equipment failure that forces tenants out of their space for two days costs far more in concessions and relationship damage than any maintenance model would cost to build.

Predictive maintenance does not eliminate all failures. Pipes burst suddenly. Electrical faults happen without warning. Converting the failures that do give advance signals, roughly 40–50% of all equipment failures according to a 2020 Honeywell Building Technologies study, into planned work orders instead of emergency callouts.

For a mid-size commercial portfolio, the combination of lower repair costs, reduced energy spend, and fewer tenant disruptions makes predictive maintenance one of the higher-return technology investments available in property management today. The data is already there. The question is whether you have a system reading it.

Book a free discovery call