Every maintenance conference in the last five years has been dominated by predictive maintenance. AI, IoT sensors, machine learning—the promise is compelling: stop doing maintenance on a schedule and only do it when the machine actually needs it.
The reality is more nuanced. Predictive maintenance works brilliantly for specific assets in specific contexts. For others, calendar-based preventive maintenance is cheaper, simpler, and just as effective. The companies that get this right don't pick one strategy and apply it everywhere. They match the strategy to the asset.
Let's be honest about what each approach actually costs and delivers.
What Preventive Maintenance Actually Is—And Isn't
Preventive maintenance (PM) is maintenance performed on a fixed schedule—calendar-based, meter-based, or cycle-based—regardless of the current condition of the asset. Change the oil every 3,000 hours. Inspect the belt every 90 days. Replace the filter every 6 months.
The advantage is simplicity. Schedules are easy to build, compliance is easy to measure, and technicians know what to do without complex diagnostics. The disadvantage is waste. You're replacing components that might have 40% of their life remaining, and occasionally missing failures that occur between scheduled maintenance intervals.
Most facilities run their PM programs with 70-85% effectiveness—meaning 15-30% of preventable failures still happen on assets with active PM schedules. Usually because the interval was wrong, the task didn't address the actual failure mode, or compliance slipped.
What Predictive Maintenance Actually Costs
Predictive maintenance (PdM) uses condition monitoring data—vibration, temperature, acoustic emissions, oil analysis—to identify when a specific asset is approaching failure. Instead of a fixed schedule, you intervene when the data says the asset needs it.
The hardware cost is real. Industrial vibration sensors run $500-$5,000 per installation point depending on the technology. A motor-compressor-pump system might need 4-6 monitoring points. Add data acquisition hardware, connectivity infrastructure, and software platforms—and a fully instrumented critical asset costs $8,000-$25,000 to set up.
Then there's the expertise cost. PdM data without skilled analysts is noise. Vibration analysis requires trained personnel who can distinguish bearing defect frequencies from imbalance signatures from resonance issues. You either develop that expertise in-house (expensive and slow) or outsource it (ongoing cost).
The break-even calculation: if an asset's failure costs $50,000 in downtime and lost production, and you prevent one failure every 18 months with $15,000 in PdM infrastructure plus $5,000/year in monitoring costs, the math works strongly. If the same asset fails for $8,000 and your PM program already catches 85% of failures, PM at $2,000/year in labor is the right answer.
The Asset Criticality Framework for Choosing Your Strategy
The most practical way to allocate maintenance strategy across your asset base is a two-axis criticality matrix: failure consequence (what happens when it fails?) on one axis, failure pattern predictability (does it give warning before it fails?) on the other.
High consequence, predictable failure pattern: this is the ideal case for condition monitoring and predictive maintenance. Large rotating machinery, compressors, critical pumps—these give warning signs that vibration analysis, thermography, or oil analysis can detect weeks before failure.
High consequence, unpredictable failure pattern: these assets fail without warning regardless of their condition. Electronic components, safety devices, and certain wear items fall here. The right strategy is redundancy plus run-to-failure, not expensive predictive monitoring that can't detect a sudden failure mode.
Low consequence, predictable failure: calendar-based PM works fine. The cost of failure is low enough that optimizing the maintenance interval isn't worth significant analytical investment.
Low consequence, unpredictable failure: run-to-failure with quick restoral. Don't invest in PM programs or monitoring for assets whose failure costs $200 to fix and causes no production impact.
Building a Hybrid Program That Works
The most successful maintenance programs combine all three strategies: predictive for critical assets where the data economics work, preventive for important assets where scheduled maintenance catches the majority of failures, and run-to-failure for low-consequence assets.
A common starting point: identify your top 10-15 critical assets with failure costs above $30,000 per event. Assess whether they exhibit detectable degradation patterns (most rotating equipment does). For those that do, build a business case for condition monitoring investment. For those that don't, build better PM procedures.
Start the predictive program with vibration analysis on your top 3-5 rotating machines before investing in permanent sensors. Portable vibration meters run $3,000-$8,000 and allow quarterly route-based monitoring on multiple assets. This builds the evidence base for your program—and for the capital budget request for permanent sensors.
The goal isn't to maximize predictive maintenance. It's to allocate maintenance investment where it delivers the highest return on prevention.
Technology Tools for Predictive Maintenance
CMMS platforms are integrating predictive capabilities at different levels. Limble CMMS and UpKeep both offer IoT sensor integrations that can trigger work orders from threshold breaches. Fiix has sensor-triggered PM capabilities built into its enterprise tier. IBM Maximo Application Suite and HxGN EAM have mature PdM modules with built-in analytics for large enterprise environments.
Dedicated PdM platforms like Augury, SparkCognition, and SKF Axios add specialized machine learning capabilities—but require integration with your CMMS to close the loop from condition alert to work order to repair completion.
For most facilities, the right starting technology is portable condition monitoring tools plus a CMMS that can ingest the data and trigger alerts. Full sensor-to-work-order automation is the end state to work toward, but it doesn't need to be the starting point.