Most FM teams will tell you they are moving toward smarter, more data-driven maintenance. Sensors are going in, systems are being connected, dashboards are being built. Somewhere in that process, predictive maintenance became the goal.
Predictive maintenance is not the end goal. It’s a milestone, and there is a next stage that most organizations in the industry haven’t seriously thought through yet. The gap between where FM currently operates and where it needs to go is wider than most leaders are willing to admit.
Still largely reactive
Maintenance in most buildings starts when something breaks, when a user complains, or when an inspection catches a problem that has already been developing for weeks. The structural conditions that keep this the default, fragmented data, limited budgets, systems that don’t talk to each other, are still very much in place across the sector.
In recent studies on predictive maintenance in building facilities, this is described as the operational reality for a large part of the sector, including organizations that consider themselves ahead of the curve. That finding tends to land differently when you’re sitting in a strategy meeting talking about smart buildings.
Four levels that are not the same thing
Reactive maintenance is run-to-failure. An asset stops functioning, someone responds, the system is restored. Most buildings spend more time here than their managers would like to admit.
Preventive maintenance adds a schedule. Calendar-driven, interval-based. It reduces surprises but creates a different problem: you end up replacing components that still have useful life left simply because the date says so.
Condition-based maintenance changes the trigger. Instead of the calendar, you use the actual condition of the asset. Measurement and monitoring decide when intervention is warranted. This is where a lot of organizations think they’ve arrived when they’ve installed sensors and built a dashboard. Most of the time, they haven’t.
Predictive maintenance is a different thing entirely. According to research on IoT-enabled maintenance business models, the standard definition describes it as condition-based maintenance performed on the basis of a forecast derived from repeated analysis of significant degradation parameters. A sensor telling you the temperature is too high right now is condition monitoring. A system telling you a component is likely to fail in the next three weeks, based on its degradation trajectory, is predictive maintenance. Most organizations conflate the two, and that confusion is expensive.
The groundwork most organizations want to skip
Treating the gap between reactive and predictive as a technology problem is the most common mistake organizations make. The technology is rarely the bottleneck.
Maintenance histories that were never properly recorded. Building automation systems that don’t connect to maintenance management platforms. Technicians who notice things in the field that never make it into any system. Models that work in one building and fall apart in another because the occupancy patterns or climate are different enough to break the assumptions.
Regarding the challenges holding predictive maintenance back in practice, data heterogeneity and the gap between research conditions and real-world deployment consistently come out as the persistent blockers, not algorithmic sophistication. Organizations that fail at this transition usually didn’t pick the wrong software. They skipped the groundwork.
It starts with the assets that hurt when they fail
HVAC is the most defensible place to begin. Energy-intensive, always running, directly tied to comfort and air quality. Based on research into building facility maintenance, faulty HVAC operations can drive 20 to 30 percent excess energy consumption. That’s not a rounding error. That’s a structural cost being paid every month by organizations that believe they’re reasonably in control.
Looking specifically at predictive maintenance algorithms applied to HVAC systems, the potential is real but so are the constraints around data quality, model transferability, and validating predictions in live environments. Pick a small number of assets where failure has real consequences for energy, comfort, or continuity. Build the data infrastructure there first. Get the feedback loops working. Then expand.
Beyond prediction
Predictive maintenance tells you what is likely to happen. Prescriptive maintenance tells you what to do about it. Technician availability, scheduling constraints, cost trade-offs, contract obligations, all factored in. It doesn’t stop at the forecast, but generates a recommendation.
The distinction matters more than it might seem. Referring to research on offshore wind maintenance strategies, prescriptive maintenance incorporates predictions into a wider maintenance plan, while predictive approaches stop at the failure probability. In industrial manufacturing, this has already been demonstrated connecting directly to production planning and resource allocation in real time. Not FM, but the direction is clear.
As for where prescriptive maintenance actually stands today, the most comprehensive recent overview covers a decade of publications and finds that interest is growing fast while deployment remains limited. Interoperability issues, real-time optimization demands, and scalability constraints are all still blocking progress. For buildings specifically, prescriptive maintenance is mostly still in the conceptual and pilot phase. That will change. The question is which organizations are building toward it deliberately, and which will still be catching up on predictive when it does.
A structural lag
Buildings are generating more operational data than ever. The analytical capability to use it keeps improving. What isn’t keeping pace is the organizational capacity to act on what the data reveals.
Grounded in systematic research on predictive maintenance and digital twins, prescriptive maintenance can be understood as predictive maintenance with an action-planning layer on top, and that layer is both technically and organizationally demanding. FM decision structures were designed for scheduled maintenance and reactive repair. They were not built to absorb continuous data streams and turn probabilistic forecasts into coordinated operational decisions.
That’s not a criticism of the people running these organizations. It’s a description of a design that made sense for a different era. The problem is that the era is shifting faster than the structures are.
Three questions for you
For the ten most critical assets under management: do you know when each was last serviced, what its current condition is, and when it will likely need attention next? Not roughly. Specifically enough to make a decision.
Is the maintenance history of your portfolio recorded in a way a system can actually learn from? Or does the knowledge of how your assets behave and fail live mostly in the heads of people who will retire in the next several years?
Who in your organization owns the connection between building data and operational decisions? Not the BMS. Not the CMMS. The actual question of what your buildings know, and what happens with that.
If those answers are vague, the distance to prescriptive maintenance is longer than the roadmap suggests. And it doesn’t close on its own.
Originally published on LinkedIn.