Predictive vs Preventive Maintenance: When the Math Actually Works
Predictive maintenance pays on some assets and loses money on others. Here is the asset-level breakeven calculation, with worked examples, a criticality score, and a 90-day plan to prove it on your own plant.
Predictive maintenance is not better than preventive maintenance in general. It is better on specific assets, and it loses money on others. The difference comes down to two questions you can answer from your own records: what does an unplanned failure of this asset really cost, and how much of that cost would warning time let you avoid?
Programs that monitor everything with the same rigor tend to discover this the hard way. The sensors on a few critical assets pay for the program, while monitoring on dozens of cheap, redundant assets quietly erodes the return. This article gives you the asset-level calculation, a criticality score to rank candidates, and a 90-day plan to test both on your plant.
What each strategy is, and what the research says
PNNL's O&M guidance defines preventive maintenance as actions performed on a time- or machine-run-based schedule that detect, preclude, or mitigate degradation, and predictive maintenance as measurements that detect the onset of degradation so causal stressors can be eliminated or controlled before significant deterioration [1]. In short, preventive acts on the calendar or the hour meter. Predictive acts on measured condition.
The U.S. Department of Energy's O&M Best Practices Guide (Release 3.0) estimates 12% to 18% cost savings from preventive maintenance over a reactive program, and a further 8% to 12% from a properly functioning predictive program over preventive alone [2]. It also cites independent surveys reporting average reductions of 25% to 30% in maintenance costs and 35% to 45% in downtime from a functional predictive program [2].
NIST's study of U.S. manufacturers associated predictive maintenance with 15% less downtime, an 87% lower defect rate, and 66% fewer inventory increases due to maintenance issues [3].
Key Statistics
12% to 18% [2]
Estimated savings from preventive over reactive maintenance, per the DOE O&M guide
8% to 12% [2]
Further savings from predictive over preventive maintenance alone, per the DOE O&M guide
15% [3]
Less downtime associated with predictive maintenance in NIST's study of U.S. manufacturers
45% to 55% [1]
Predictive share of the maintenance mix in the top-performing benchmark PNNL describes
Those are averages across programs. They do not tell you which of your assets to monitor. Notice too that the top-performing mix PNNL describes still keeps 25% to 35% preventive [1]. Nobody serious argues for monitoring everything.
Where preventive maintenance still wins
Preventive maintenance is the better choice when an asset's wear is predictable from run time, its failure is cheap, or a spare or redundant unit keeps production running. Typical examples are small motors, standard conveyors, building HVAC, and banks of identical equipment where one failure does not stop the line.
Two practical reasons favor it on those assets. First, sensors, connectivity, installation, and someone to review alerts all cost money every year, and on a cheap, redundant asset that annual cost can exceed the cost of the failures you would avoid. Second, OEM schedules and your own failure history already describe the wear pattern well.
Regulated plants have a further consideration. In U.S. pharmaceutical manufacturing, 21 CFR 211.67 requires equipment to be cleaned and maintained at appropriate intervals, with written procedures established and followed for cleaning and maintenance [4]. Condition monitoring can inform those procedures, but it does not remove the obligation to have and follow them, so price predictive monitoring on GMP equipment as an addition, not a replacement.
Where predictive maintenance wins
Predictive maintenance earns its keep where three things are true at once:
- The failure is expensive. It stops a constraint, damages other components, or triggers customer penalties.
- The failure is not well predicted by hours alone. Load, environment, or process conditions vary, so a fixed interval either replaces healthy parts or misses degradation.
- The failure mode is detectable early. Vibration, temperature, current, oil condition, or ultrasound changes before the asset fails, giving enough warning to plan the work.
When all three hold, warning time turns an emergency into a planned job. PNNL describes the costs that warning time avoids: repair labor that runs higher than normal because failures need more extensive repairs, overtime to bring critical equipment back quickly, secondary damage, and shortened equipment life [1]. The DOE guide adds that predictive programs let teams schedule work to minimize overtime and order parts well ahead of time [2].
Rank by consequence, not by asset price
A modest motor on the main line can outrank an expensive standby generator with full redundancy. The motor stops revenue. The generator has a backup. Score what a failure does to production, not just what the asset costs to replace.
The criticality score
Rank every candidate asset on three 1 to 10 scales and multiply them:
| Factor | Score 1 | Score 10 | Source |
|---|---|---|---|
| Production impact | Redundant or non-production asset | Stops the constraint or a whole line | Line layout, operations |
| Failure likelihood | No failures in recent history | Fails repeatedly | CMMS work order history |
| Cost of failure | Cheap, in-stock repair | Major repair, long lead time, secondary damage | CMMS costs, purchasing |
The product ranges from 1 to 1,000. Sort descending and treat the top of the list as predictive candidates. The cutoff is a judgment call your team should agree on, then test with the breakeven calculation below rather than fixing a threshold in advance.
The breakeven calculation, asset by asset
For each candidate, compare the failure cost warning time would avoid each year with what monitoring costs each year:
avoided_cost_per_year = failures_per_year * share_caught_in_time * (unplanned_cost - planned_cost)
monitoring_cost_per_year = monthly_fee * 12 + (sensors + install + integration) / years_of_use
net_per_year = avoided_cost_per_year - monitoring_cost_per_year
payback_months = (sensors + install + integration) / ((avoided_cost_per_year - monthly_fee * 12) / 12)The share caught in time is the assumption that deserves the most scrutiny. Use a conservative figure until your own pilot gives you a measured one.
A critical asset. Here is an illustrative modeled estimate for a reactor mixing system: about $135,000 a year of avoided cost against about $4,100 a year in monitoring fees, paying back roughly $20,400 of setup cost in under two months. Model inputs: 1.8 unplanned failures a year from site history, half of them caught in time, $187,000 per unplanned failure versus $37,000 for the same repair planned, $8,400 for sensors, $12,000 for CMMS integration, and a $340 monthly fee.
avoided = 1.8 * 0.5 * (187000 - 37000) = 135000
fees = 340 * 12 = 4080
payback = 20400 / ((135000 - 4080) / 12) = 1.9 monthsA non-critical asset. Here is an illustrative modeled estimate for a small cooling pump: about $1,200 a year of avoided cost against $2,160 a year in monitoring fees, so it never pays back. Model inputs: 0.4 failures a year, half caught in time, $8,200 per unplanned failure versus $2,200 planned, $2,800 for a sensor, and a $180 monthly fee.
avoided = 0.4 * 0.5 * (8200 - 2200) = 1200
fees = 180 * 12 = 2160
net = avoided - fees = -960 per year, before the sensor costThe reactor is an obvious candidate. The pump stays preventive. Most real plants have a long middle band of assets where the answer depends on the share caught in time, which is why the pilot matters.
Two cautions. Use your own failure history, not OEM averages, because your load, feedstock, and operating hours differ. And count planned repair cost honestly; predictive maintenance does not make the repair free, it makes it cheaper and scheduled.
The costs that quietly erode the return
Even on the right assets, four costs can push payback out:
- Installation and integration. Mounting, connectivity, and getting alerts into the CMMS as work orders often cost more than the sensor hardware. A legacy CMMS without an API adds middleware to build and maintain.
- Alert review time. Someone has to triage every alert. Poorly tuned thresholds create alert fatigue, and once technicians stop investigating, the program stops catching failures.
- Diagnostic disagreement. When the system names the wrong failure mode and an experienced technician finds the real cause, trust drops. Capture those corrections in the work order so the model and the team improve together.
- Data readiness. Missing asset hierarchies, inconsistent failure codes, and unmapped sensor tags delay value. See our guide to why predictive alerts point to the wrong asset.
PNNL's KPI guidance gives targets that show whether the program is shifting the mix: corrective maintenance below 10% to 20% of maintenance hours, predictive above 45% to 55%, and total maintenance cost below 3% of replacement asset value, each trending the right way [5].
A hybrid matrix for the whole asset base
Plot every asset on criticality and on how predictable its wear is from run time:
| Quadrant | Strategy | Example assets |
|---|---|---|
| High criticality, low predictability | Predictive monitoring | Presses, turbines, reactors, critical compressors |
| High criticality, high predictability | Preventive, with condition checks to adjust intervals | Main drive motors, pumps on critical lines |
| Low criticality, low predictability | Basic preventive or run to failure | Small fans, non-critical conveyors |
| Low criticality, high predictability | Time-based preventive | Standard motors, belts, filters |
A 90-day plan to prove it
Days 1 to 30: score the assets. Build the criticality list from CMMS history, production impact, and repair costs. Pick the top three to five candidates and run the breakeven calculation for each with conservative inputs.
Days 31 to 60: pilot. Instrument the candidates, connect alerts to the CMMS, and define success before you start: failures caught in time, a false-positive limit, and the payback you expect. Log every alert, the investigation result, and the action taken.
Days 61 to 90: decide from data. Recalculate the breakeven with the measured share caught in time and the real alert workload. Expand to the next tier only if the measured numbers hold. If they do not, fix data and workflow gaps before adding sensors.
For the finance side, our predictive maintenance ROI guide covers total cost of ownership, and the AI maintenance business case covers presenting payback, NPV, and IRR to a capital committee.
Monitory helps manufacturing teams make this transition with AI-powered predictive maintenance that integrates with your existing CMMS and automates the asset selection, monitoring, and work order workflow. See how Monitory reduces unplanned downtime across automotive, food processing, and chemical manufacturing.
Frequently asked questions
Is predictive maintenance always better than preventive maintenance?
No. It wins on assets where failures are expensive, not well predicted by run time, and detectable early. On cheap, redundant assets with predictable wear, preventive maintenance usually costs less.
How do I calculate predictive maintenance ROI for one asset?
Multiply failures per year by the share you expect to catch in time and by the difference between unplanned and planned repair cost. Subtract annual monitoring cost. Divide setup cost by the monthly net benefit to get payback.
What savings does the research report?
The DOE O&M guide estimates 12% to 18% savings from preventive over reactive maintenance and 8% to 12% more from predictive over preventive [2]. NIST associated predictive maintenance with 15% less downtime [3]. Your results depend on your asset mix.
Can condition monitoring replace preventive maintenance in regulated plants?
Not entirely. In U.S. pharmaceutical manufacturing, written cleaning and maintenance procedures must still be established and followed [4]. Condition data can inform the intervals.
References
[1] Pacific Northwest National Laboratory, "O&M Best Practice Issue Discussion: Maintenance Approaches." https://www.pnnl.gov/projects/om-best-practices/maintenance-approaches
[2] U.S. Department of Energy, Federal Energy Management Program, "Operations and Maintenance Best Practices Guide, Release 3.0," 2010. https://www.energy.gov/sites/prod/files/2020/04/f74/omguide_complete_w-eo-disclaimer.pdf
[3] NIST, "Research Suggests Significant Benefits to Investing in Advanced Machinery Maintenance," 2020. https://www.nist.gov/news-events/news/2020/06/research-suggests-significant-benefits-investing-advanced-machinery
[4] Legal Information Institute, "21 CFR 211.67, Equipment cleaning and maintenance." https://www.law.cornell.edu/cfr/text/21/211.67
[5] Pacific Northwest National Laboratory, "Applying Key Performance Indicators." https://www.pnnl.gov/projects/om-best-practices/applying-key-performance-indicators
Related articles
Ready to put this into practice?
See how Monitory helps manufacturing teams implement these strategies.