Predictive vs. Preventive Maintenance: Where the Cost Calculation Actually Breaks

Most cost comparisons between predictive and preventive maintenance read like vendor literature. PdM wins. PM is outdated. Buy sensors. The reality on the floor is more complicated—and more useful to understand.

According to the U.S. Department of Energy's Operations and Maintenance Best Practices guide, a properly functioning predictive maintenance programme saves 8% to 12% over preventive maintenance alone.

That's a real number from a credible source. But it's also an average across a wide range of facilities and asset types, which means some assets do significantly better and some don't justify the monitoring cost at all. The calculation breaks at the asset level, not the programme level.

Why Scheduled Maintenance Wastes Money on Some Equipment

Preventive maintenance operates on a fixed schedule—replace the bearing every 6,000 hours, inspect the seal every quarter, lubricate on the first Monday of the month. The logic is sound: intervene before failure rather than after. The problem is that roughly 30% to 40% of scheduled preventive maintenance tasks are performed on equipment that doesn't need attention, according to industry estimates. That's technician time, parts consumption, and production interruption spent on equipment that was running fine.

For low-value, easily replaced components with predictable wear curves, that waste is manageable. For high-value rotating equipment running continuously, it compounds quickly. A pump that runs 8,000 hours a year on a schedule built around a worst-case failure scenario gets serviced more often than its actual condition warrants. Multiply that across a large asset fleet and the over-maintenance cost becomes significant.

PM also can't catch failures that develop between intervals. A bearing that picks up a defect three weeks after a scheduled inspection will run to failure before the next one. On a continuous process line—paper, chemicals, food processing—that unplanned failure is more expensive than the interval schedule was designed to prevent.

Where Predictive Maintenance Actually Earns Its Cost

PdM works best on assets where three things are true: the failure is detectable in advance through condition monitoring, the cost of failure significantly exceeds the cost of monitoring, and the failure mode doesn't follow a predictable wear curve that PM could capture just as effectively.

High-value rotating equipment fits that profile well. Vibration analysis on motors, pumps, compressors, and gearboxes can identify bearing wear, shaft misalignment, and rotor imbalance weeks or months before failure occurs—detectable shifts in frequency and amplitude that manifest well before a breakdown. For a compressor on a continuous process line, that early warning is worth a significant monitoring investment because the alternative is an unplanned shutdown with cascading production losses.

The cost asymmetry is stark. A planned bearing replacement during a scheduled maintenance window runs a few hundred to a few thousand dollars in labour and parts. The same repair as an emergency call at 2 a.m., with parts sourced urgently and production stopped, costs several times more—and that's before accounting for the downtime itself.

On high-throughput lines, unplanned downtime in process industries can run $10,000 to $250,000 per hour depending on the operation. The monitoring cost doesn't need to be justified against the repair cost. It needs to be justified against the avoided failure cost.

Long lead-time spare parts change this calculation further. If a critical component takes 12 to 16 weeks to source, the option to run to failure doesn't exist in any practical sense—the cost isn't just the repair, it's three or four months of production loss. PdM on those assets isn't an efficiency question; it's a continuity question.

Where Preventive Maintenance Still Wins

Vibration sensors and condition monitoring platforms cost money to deploy, calibrate, and maintain. For a small operation with fewer than 20 assets, or for assets where failure cost is low and replacement is fast, the sensor infrastructure often can't be justified on ROI alone.

Standard auxiliary equipment—smaller pumps, fans, HVAC components, conveyors serving non-critical lines—tends to have predictable wear patterns and low failure consequences. A well-designed PM schedule with regular inspections handles these assets reliably at lower cost than deploying continuous monitoring. The monitoring investment should be proportional to what's at stake if the equipment fails.

There's also a failure mode consideration that gets overlooked. Some assets don't give useful advance warning through vibration or thermal imaging. Electrical faults in certain motor configurations, for instance, may not produce detectable vibration signatures until failure is imminent—which is too late to be useful.

Running condition monitoring on those assets produces data without actionable lead time. PM on a defined interval, combined with basic visual inspection, often performs better for equipment in that category.

Run-to-failure is also a legitimate strategy for redundant, non-critical equipment where a spare is on the shelf and switchover takes minutes. Spending on monitoring for equipment in that category is waste, not prudence.

The Asset Criticality Framework That Makes the Decision Cleaner

The facilities that get this right don't choose between PdM and PM as programmes. They map their asset population by criticality and match the maintenance strategy to the failure consequence.

Category A assets are those where unplanned failure causes a line shutdown, has high repair or replacement cost, involves long lead-time parts, or presents a safety or regulatory exposure. These get condition monitoring—vibration analysis, oil sampling, thermal imaging, or some combination depending on the failure mode. The monitoring cost is justified by the avoided failure cost, not by a general ROI argument.

Category B assets are important but have shorter recovery times, lower replacement costs, or some level of redundancy. These typically run on PM schedules with inspection intervals set by failure history rather than manufacturer defaults. Adjusting PM intervals based on actual failure data—rather than leaving the OEM-recommended schedule in place indefinitely—recovers significant maintenance cost over time.

Category C assets are low-consequence, easily replaced, or redundant. Run-to-failure or minimal PM. No monitoring investment.

A petrochemical facility with 1,000 assets might have roughly 120 in Category A, 230 in Category B, and 450 in Category C, with the remainder genuinely suited to run-to-failure, according to analysis from Vista Projects. Recommending condition monitoring for all 1,000 assets overlooks the economics entirely.

The Sensor Infrastructure Cost Nobody Puts in the Proposal

PdM vendors tend to lead with ROI projections and follow with implementation costs. The implementation costs deserve more scrutiny.

Basic temperature sensors start around $100 per monitoring point. Vibration sensors capable of the frequency resolution needed for bearing fault detection run $500 to $1,000 each or more, depending on specification. Add installation, network infrastructure, integration with your CMMS or EAM system, and the software platform to interpret the data, and a serious PdM deployment on a mid-sized asset fleet carries a capital cost that needs to be recovered over time.

The payback period matters. A PdM programme on the right assets typically achieves positive ROI within 12 to 18 months, according to industry research. But the models need 6 to 12 months of baseline data before they start generating reliable predictions—which means the benefit lag is real. Budgeting for a 12-month ramp-up period where the sensors are collecting data but not yet driving decisions is part of an honest implementation plan.

There's also an ongoing capability requirement. Vibration data produces actionable insights only when someone can interpret the frequency spectra and distinguish between normal operational variation and developing faults. ISO 18436-2 certification covers this competency for vibration analysts. Facilities that deploy sensors without building that internal capability—or contracting it externally—end up with expensive data they can't use.

What a Realistic Hybrid Programme Looks Like

Most facilities don't run a pure PdM or pure PM programme. They run a combination, and the mix should reflect asset criticality and failure history, not a blanket policy.

Category A assets get continuous condition monitoring where the failure mode supports it. Vibration analysis on rotating equipment, oil sampling on gearboxes and hydraulic systems, thermal imaging on electrical panels and high-load bearing surfaces. Inspection intervals and monitoring thresholds are reviewed annually and adjusted as failure data accumulates.

Category B assets run on PM schedules with intervals adjusted by failure history rather than defaults. If a component has never failed before its scheduled replacement in five years of data, the interval can often be extended. If it's failed twice between intervals, the interval needs tightening—or the asset needs to move up to Category A.

Category C assets get run-to-failure with spare parts stocked. The stock level is driven by lead time and failure frequency, not by anxiety about unplanned downtime on a $400 pump.

The total cost of this hybrid approach is lower than a blanket PdM programme and more reliable than a blanket PM programme applied uniformly across a mixed asset fleet. The decisions are harder because they require asset-level analysis rather than a programme-wide policy, but the economics are significantly better.

If you're evaluating your maintenance strategy and want to work through the asset criticality mapping for your facility, our team can help structure the analysis. Contact us to set up a working session.

For more on how maintenance strategy intersects with multi-site operations, see our piece on process standardization across facilities and how PM interval decisions feed into cross-plant benchmarking.