Preventive vs Predictive vs Reactive Maintenance

Ask ten maintenance teams to describe their strategy and you will get ten different vocabularies for roughly the same four ideas. Reactive, corrective, breakdown, run-to-failure. Preventive, planned, scheduled, time-based. Predictive, condition-based, prognostic, data-driven. Proactive, which sometimes means predictive and sometimes means something entirely different. The confusion is not academic: it drives budget decisions, CMMS configuration, contract wording and headcount plans. Across two decades of CMMS, CAFM and EAM implementations I have watched organisations spend seriously large sums moving assets onto the wrong strategy because nobody stopped to define the terms and compare them honestly. This guide does that, then gives you a framework to decide per asset.

The message up front: two beliefs cause most of the damage in this space. The first is that run-to-failure is a sign of an immature organisation. It is not; for low-consequence assets it is the cheapest correct answer, and deliberately choosing it is a mark of maturity. The second is that predictive maintenance is universally superior. It is not; it is selectively superior, on assets whose failures are both consequential and detectable with enough warning to act. Everything else in this article follows from those two corrections.

1. Why the vocabulary actually matters

I have sat in workshops where the maintenance manager and the operations director spent forty minutes disagreeing before discovering they were using the word "corrective" to mean opposite things. One meant unplanned repair after a breakdown. The other meant planned rectification work raised from an inspection finding. Both are legitimate industry usages. Neither is wrong. But the budget line they were arguing about depended on which definition you applied.

The practical consequence shows up in three places. First, in the CMMS: work order types and maintenance-type flags are how you measure the mix between planned and unplanned work, and if the definitions are loose the reporting is meaningless. Second, in contracts: a service provider committed to "preventive maintenance" with no definition of trigger basis can bill you for calendar visits while the assets degrade on meter hours nobody tracks. Third, in strategy conversations: you cannot compare options you have not defined.

So this article uses one consistent set of definitions throughout, aligned with how the reliability standards and the major platforms use them. If your organisation uses different words, that is fine, but write yours down and make them exclusive. For the mechanics of encoding these distinctions in your system, see the work order types in CMMS pillar.

2. Reactive maintenance (run-to-failure): the strategy with the worst reputation

Definition: no intervention until the asset fails functionally. On failure, you repair or replace it. No inspections, no scheduled service, no monitoring.

There is a crucial distinction inside this family that the industry frequently blurs, and getting it right changes how you think about the whole topic:

• Deliberate run-to-failure: you assessed the asset, concluded its failure consequence is low and easily absorbed, and made a documented decision not to maintain it proactively. You may still stock a spare and have a response process. This is a strategy.
• Default reactive: nobody assessed the asset. It simply never made it into a maintenance plan, so it runs until it breaks and everyone is surprised when it does. This is an absence of strategy, and it is what gives reactive maintenance its bad name.

The two look identical from the outside, which is exactly the problem. When an auditor sees seventy percent of your work orders coded reactive, they cannot tell whether you have a disciplined run-to-failure policy on low-criticality assets or no policy at all. The only thing that distinguishes them is documentation: a criticality assessment, a recorded decision, and a defined response standard.

Where reactive genuinely hurts is when it is applied by accident to consequential assets. The cost profile then turns brutal: emergency labour at premium rates, expedited parts freight, secondary damage from a small failure cascading into a large one, unplanned downtime, and the organisational chaos of firefighting. The classic industry observation is that an emergency repair typically costs a multiple of the same work performed planned, and while I will not quote a precise ratio because it varies enormously by sector, the direction is not in dispute.

Cost profile: near-zero prevention cost, high and volatile failure cost. Data requirement: almost none, though you still want the failure captured properly afterwards. Labour model: responsive, requires surge capacity and out-of-hours cover, difficult to plan or level-load. Spares implication: you either hold the spare or you accept the lead time, and for consequential assets that decision is the whole strategy.

3. Preventive maintenance: the workhorse and its blind spot

Definition: intervention at a fixed interval, regardless of the asset's actual condition. The interval is set by calendar time, by meter reading (running hours, cycles, kilometres, throughput), or by statutory requirement.

Preventive maintenance is the backbone of almost every real maintenance programme, and for good reason. It is plannable, so you can level-load a crew twelve months ahead. It is auditable, which matters enormously in regulated environments and in facilities management contracts where the schedule is the deliverable. It requires no sensors, no analytics and no data-science capability. And for failure modes that are genuinely age-related, replacing a component before it wears out is exactly the right move.

The blind spot is in the definition: fixed interval regardless of condition. That cuts both ways. You service healthy assets that did not need servicing, which wastes labour and parts. And you can still be surprised by an asset that develops a fault two weeks after its annual service, because the schedule has no mechanism to notice.

There is a second, less-discussed cost. Intrusive maintenance introduces its own failure risk. Every time you open a machine you risk contamination, misassembly, incorrect torque, damaged seals and infant mortality on newly fitted parts. Reliability literature has documented this for decades: a meaningful proportion of failures occur shortly after maintenance intervention. Over-maintaining an asset is not neutral. It actively reduces reliability.

Cost profile: moderate, predictable and continuous prevention cost; reduced but non-zero failure cost. Data requirement: an accurate asset register, manufacturer recommendations or a task library, and meter readings if you are using meter triggers. Labour model: the most plannable of all strategies, which is why it dominates outsourced FM. Spares implication: highly predictable consumption, so kitting and min/max stocking work well.

A note on scope: this article compares strategy families, so it treats preventive as one family. Within it, the choice between calendar triggers, meter triggers and condition triggers is a distinct and important design decision covered in the preventive maintenance strategies pillar, with the full programme build-out in the complete guide to preventive maintenance.

4. Predictive maintenance: powerful, conditional, and frequently mis-sold

Definition: you measure the actual condition of the asset, detect the early signature of a developing failure, estimate how much useful life remains, and intervene only when the evidence says intervention is warranted.

The mechanism that makes this possible is the P-F interval: the window between the point at which a developing failure first becomes detectable and the point at which the asset fails functionally. Predictive maintenance lives or dies on two questions about that window. Does it exist for this failure mode at all? And is your monitoring frequent enough to catch it inside the window?

Some failure modes give generous warning. A bearing degrading through the classic stages produces detectable vibration change over weeks. Oil analysis can flag wear metals months out. Insulation degradation, corrosion and erosion are all gradual. Other failure modes give effectively none. A control board fails, a shaft fractures, a fuse blows. There is no trajectory to trend, so no amount of monitoring will predict them. Sensors on assets with sudden failure modes generate data and no value, which is one of the most expensive mistakes in this field.

The techniques that carry most real-world predictive programmes are unexciting and long-established: vibration analysis on rotating equipment, infrared thermography on electrical distribution and mechanical plant, oil and lubricant analysis on gearboxes and engines, ultrasonic inspection for leaks and early bearing faults, motor current signature analysis, and simple process-parameter trending from data your BMS or SCADA historian already holds. The detailed treatment, including remaining useful life and where machine learning genuinely helps, is in the predictive maintenance and failure prediction pillar.

Cost profile: high upfront investment, ongoing analytics and skill cost, sharply reduced failure cost on the right assets. Data requirement: the heaviest of any strategy. Condition data at sufficient frequency, a clean asset register, and ideally several years of reliable failure history to know what you are looking for. Labour model: fewer routine intrusive tasks, more analysis and more just-in-time intervention, which is harder to level-load than a fixed schedule. Spares implication: this is where predictive quietly pays. Advance warning lets you order rather than expedite, and stage parts instead of stocking them speculatively.

5. Corrective and proactive: the two terms people keep confusing

These two words cause more cross-talk than any others in maintenance vocabulary, so they deserve their own section rather than a footnote.

Corrective maintenance is work done to restore an asset to its required function after a fault has been identified. The key point, and the source of all the confusion, is that corrective work is not automatically unplanned. It splits cleanly:

• Immediate corrective (breakdown, emergency): the asset has already failed and work starts now. This is the overlap with reactive maintenance, and it is why the two words get used interchangeably.
• Deferred corrective (planned corrective): a fault or defect was found during an inspection, a PM visit or a condition reading, but the asset is still functioning. The rectification is raised, scheduled and executed in a planned window. This is proactive work even though it is coded corrective.

The ratio between those two sub-types is one of the most useful and most underused metrics in maintenance. A high deferred-corrective proportion means your inspection and monitoring regime is actually finding things before they break, which is precisely what you want. If your corrective work is almost entirely immediate, your proactive regime is not detecting anything and you are being told about failures by the asset rather than by your own programme.

Proactive maintenance is used two ways. Loosely, it is the umbrella for anything that is not reactive: preventive plus predictive plus planned corrective. That usage is common and harmless. More precisely, and more usefully, proactive maintenance means acting on the root cause of failure rather than its symptoms: precision alignment and balancing, contamination and lubrication control, correcting operating practices, root cause failure analysis feeding design changes, and specification improvements at procurement.

That narrow definition matters because it is a genuinely different lever. Preventive and predictive both manage failures. Proactive maintenance in the precise sense eliminates them. If a pump family keeps failing on seals, no monitoring frequency fixes that; changing the seal specification, the alignment standard or the operating regime does. The discipline that formally decides which lever each failure mode deserves is reliability-centred maintenance, covered in the RCM introduction pillar.

6. The full comparison across all strategies

Here is the honest side-by-side. Read it as a menu, not a ranking.

Two observations from that table. First, no column wins on every row. Preventive is unbeatable on plannability, reactive on prevention cost, predictive on avoiding unnecessary intervention, proactive on eliminating the problem entirely. Second, the bottom row is the only one that matters when you are making a decision. Everything above it is input.

7. The cost comparison nobody runs properly

The most common analytical error I see is comparing prevention cost across strategies and stopping there. Reactive looks cheapest because it has no prevention line. Predictive looks most expensive because its capital cost is visible. Both conclusions are wrong because the comparison is incomplete.

The only defensible comparison is total cost of the failure consequence over the asset's life, which has four components for each strategy:

• Prevention cost: inspections, services, parts consumed proactively, monitoring hardware and analysis.
• Repair cost: labour and parts to fix failures that occur under that strategy, at whatever rate the urgency demands.
• Consequence cost: lost production or service availability, contractual penalties, safety and compliance exposure, energy waste from degraded operation, reputational damage.
• Secondary damage cost: the difference between fixing a developing fault and fixing the wider machine after the fault destroyed it.

That fourth component is where reactive maintenance quietly bankrupts a maintenance budget on consequential assets, and where it is genuinely harmless on non-consequential ones. A bearing caught early is a bearing. The same bearing run to destruction can take the shaft, the seal, the coupling and the alignment with it. If the asset is a corridor extract fan, the cascade does not matter. If it is a chilled water pump serving a data hall, it matters a great deal.

When I build this comparison with clients I do it on one asset class at a time, never on the portfolio average, because the averages hide exactly the variation the decision depends on. If you want the reporting structure that makes these numbers visible in the first place, the FM KPI framework pillar covers the metric set, and the failure codes pillar covers the coding discipline without which none of it is measurable.

8. A decision framework you can apply per asset

This is the part you can lift and use. The framework rests on three questions, asked in order, per asset or per asset class. Not per site, not per portfolio.

Question 1: Consequence. If this asset fails unexpectedly right now, what happens? Score it across safety, statutory compliance, service or production availability, environmental impact, cost of repair and reputation. Keep it to three bands to stay usable: high, medium, low. Do not let this become a twenty-factor weighted matrix nobody completes; the useful precision is coarse. The ranking method is set out in the asset criticality classification pillar.

Question 2: Detectability. For the dominant failure modes of this asset, is there a detectable warning, and is the P-F interval long enough that you could realistically monitor inside it and still have time to act? Three outcomes: detectable with a comfortable interval, detectable but the interval is too short to be useful, or not detectable at all.

Question 3: Cost of the strategy versus the consequence avoided. Would the annual cost of monitoring or servicing this asset be a small fraction of the consequence cost you are avoiding, a comparable amount, or more than it? This is the question that stops enthusiasm from outrunning economics.

Combine them and you get a small set of defensible outcomes:

Applied across a typical asset register, this framework usually lands predictive on a minority of assets, the critical rotating equipment and key electrical infrastructure, preventive on the bulk, and deliberate run-to-failure on a larger population than most managers expect. That distribution is not a lack of ambition. It is the correct answer, and it frees budget for the assets that deserve it.

9. Working the framework on three real asset types

Abstract frameworks are easy to agree with and hard to apply, so here is the framework worked through on three assets that sit in almost every commercial or industrial estate. These are illustrative reasoning patterns, not case studies with invented numbers.

Asset A: primary chilled water pump serving a data hall

• Consequence: high. Loss of cooling to a data hall escalates within minutes. Contractual availability exposure, potential equipment damage, reputational impact. Even with an installed standby, losing the duty pump removes your resilience margin, which is itself a high-consequence event.
• Detectability: strong. The dominant failure modes are bearing degradation, seal wear, impeller erosion and coupling misalignment. Bearing and alignment faults produce characteristic vibration signatures with weeks of warning. Discharge pressure trending at constant speed reveals hydraulic wear. There is a genuine P-F interval.
• Cost versus consequence: overwhelmingly favourable. Permanently mounted vibration and temperature sensors on a pump set cost a small fraction of a single cooling interruption to a data hall.
• Decision: predictive, with a preventive backbone for statutory and lubrication tasks, and the condition alerts integrated into the CMMS as real work orders rather than emails. If the prediction does not become a scheduled, closed-out work order in the system where technicians actually work, it will be ignored within a couple of months.

Asset B: diesel generator in a hospital standby plant room

• Consequence: high, and partly statutory. Life-safety dependency plus regulatory inspection and testing obligations. This asset spends almost all its life idle, which is a specific reliability problem: the failure risk is concentrated at the moment of demand.
• Detectability: mixed. Engine wear shows well in oil analysis. Cooling and electrical connection issues show in thermography. Fuel degradation is testable. But the failure modes that actually cause a generator to fail to start on demand, battery collapse, control board fault, fuel system blockage, starter failure, are largely sudden or only revealed by operating it.
• Cost versus consequence: favourable, but continuous monitoring is the wrong tool. The asset is not running, so there is little condition to monitor for most of its life.
• Decision: preventive-led, with targeted condition techniques. The dominant strategy is scheduled functional testing under load, which is a preventive failure-finding task and the only way to detect hidden failures in a standby system. Layer on oil analysis, battery condition testing and annual thermography. Do not build a continuous-monitoring programme here; build a rigorous testing regime and never let a test get deferred.

Asset C: toilet extract fan in a low-rise office floor

• Consequence: low. Failure causes an odour complaint and a helpdesk ticket. No safety, no compliance breach, no production impact. Resolution within a working day is acceptable to everyone.
• Detectability: irrelevant to the decision. You could technically detect bearing wear on it. The point is that it is not worth detecting.
• Cost versus consequence: unfavourable for anything proactive. A quarterly inspection across a few hundred of these consumes real technician hours every year to prevent a consequence that costs almost nothing.
• Decision: deliberate run-to-failure, with a stocked spare unit and a documented response target. Record the criticality assessment so that when an auditor or an incoming manager asks why these assets have no PM, there is a reasoned answer rather than an embarrassed silence.

10. Running a mixed portfolio without losing control

Once you accept that different assets get different strategies, you inherit a management problem: how do you keep a mixed portfolio coherent? Four practices make the difference.

• Record the strategy on the asset. Every asset record should carry its assigned maintenance strategy and criticality band as a field, not as tribal knowledge. Every major platform supports this, whether that is IBM Maximo, SAP PM, Hexagon EAM, Infor EAM, Planon or a mid-market tool like Fiix, Limble, eMaint or MaintainX. If the strategy is not queryable, it is not managed.
• Code work orders so the mix is measurable. You need to be able to report planned versus unplanned, and within corrective, immediate versus deferred. Without that split you cannot tell whether your proactive regime is working.
• Review the assignments on a cycle. Strategy assignments age. An asset's criticality changes when the business it serves changes; a failure history accumulates and reveals modes you did not anticipate. An annual review of criticality and strategy assignment is enough for most estates.
• Let failure history override the original judgement. If an asset you put on run-to-failure has failed four times this year and each failure caused more disruption than you predicted, the assessment was wrong. Move it. The framework is a starting position, not a verdict.

On tooling: the ability to run differentiated strategies cleanly is one of the practical dividing lines between system categories. If you are still choosing a platform, the CAFM vs CMMS vs EAM vs IWMS comparison sets out what each category actually gives you.

For the standards underpinning all of this, the asset management framework in ISO 55001 gives you the governance language for documenting strategy decisions, and SFG20 remains the reference maintenance task library for building schedules in a UK and Gulf facilities context.

11. The mistakes I see repeatedly

• Treating the strategies as a maturity ladder to climb. The goal is correct assignment, not universal predictive coverage. An organisation that has thoughtfully assigned every asset is more mature than one that has monitored everything.
• Confusing default reactive with deliberate run-to-failure. The fix is documentation, and it is cheap. Assess, decide, record.
• Buying the monitoring platform before defining the use case. The technology then arrives looking for a problem, and the programme never recovers the initiative.
• Applying one strategy across a whole asset class regardless of context. Two identical pumps in different duties and different criticalities deserve different strategies. The asset type does not determine the answer; the consequence does.
• Letting the PM schedule grow without ever pruning it. Preventive programmes accumulate tasks and almost never shed them, so over-maintenance builds up quietly and nobody owns the decision to stop.
• Measuring compliance instead of reliability. Ninety-eight percent PM completion tells you the schedule ran. It says nothing about whether the assets are more reliable.
• Leaving predictive insight outside the CMMS. A prediction that does not become a work order in the system of record is an observation, not maintenance.

None of these is a technology problem. They are all strategy and discipline problems, which is encouraging, because the fix sits with maintenance leadership rather than with a vendor roadmap.

The idea to walk away with

Reactive, preventive and predictive are three different economic bets about the same question: how much do you spend before a failure to reduce what it costs you when it happens? For a low-consequence asset the right bet is to spend nothing and absorb the failure. For an age-related wear-out mode the right bet is a fixed-interval intervention. For a high-consequence asset with a detectable degradation path the right bet is to watch it and act on the evidence. And when the same failure keeps returning, the right bet is none of the three: go and remove the cause.

The two ideas worth carrying out of this article are the ones from the top. Deliberate run-to-failure is a legitimate, documented, mature strategy, and organisations that use it properly free up the budget to do predictive properly where it counts. And predictive maintenance is selectively, not universally, superior; its value is entirely conditional on consequence being high and warning being detectable. Get those two straight and most of the strategy argument resolves itself.

Final thoughts

If you want a concrete first step, it is not a platform evaluation. Take your asset register, add two columns, criticality band and assigned strategy, and fill them in for your top hundred assets by consequence. You will find assets with elaborate PM schedules that do not warrant them, and assets with no maintenance plan at all that absolutely do. That reallocation typically improves reliability without increasing spend, which is the rarest kind of win in maintenance.

Then look at your work order history and check the split between immediate and deferred corrective work. That single ratio will tell you more about whether your proactive regime is genuinely working than any dashboard a vendor will show you.

Related reading: Preventive maintenance: the complete guide, Preventive maintenance strategies (time vs meter vs condition), Predictive maintenance and failure prediction, Reliability-centred maintenance: an introduction, Asset criticality classification, Work order types in CMMS.