Field-Service Advisory

The preventive-maintenance paradox: when more maintenance creates more downtime

PM compliance tells you the tasks were done on time. It does not tell you they reduced any risk — and for most failure modes, they can't. Meanwhile the intervention itself is neither free nor safe.

By Paula Navarro · Cosmicalley  |  July 2026  |  8 min read

Almost every maintenance organisation reports PM compliance. It is on the monthly pack, it is green, and it is defended: 96% of scheduled preventive tasks completed on time. It is also, quietly, one of the least informative numbers in the operation — because it measures whether the calendar was obeyed, not whether any failure was prevented.

That gap is not academic. It is where a maintenance function can work harder every year, hit its compliance target every month, and watch unplanned downtime stay exactly where it was.

Compliance measures the calendar, not the risk

A PM-compliance number knows one thing: the task was performed inside its window. It does not know whether the failure mode that task addresses was ever going to happen. It does not know whether the intervention introduced a defect. It does not know what else the technician would have done with those four hours, or what it cost to take the asset out of production. It cannot tell you that the task is worth doing — only that it was done.

PM compliance is easy to measure, so it became the target. Reliability is hard to measure, so it became the assumption.

Most failures don't wait for the calendar

The uncomfortable evidence is nearly fifty years old and still routinely ignored. In 1978, Stanley Nowlan and Howard Heap of United Airlines published the study that founded reliability-centred maintenance, sponsored by the US Department of Defense. They classified failure modes by how failure probability behaves as an item ages, and found six patterns.

Only three of them — 11% of failure modes — showed a clear wear-out pattern: a point in the item's life where failure becomes markedly more likely, which a fixed-age replacement or restoration can be timed against. The other 89% did not. Some of those still change with age (pattern D rises early, then flattens) — the point is not that age is irrelevant everywhere, but that there is no usable wear-out age for a routine calendar task to target. And the largest single pattern, at 68%, was early-life failure: risk is at its highest immediately after an item enters service, then settles to a low, roughly constant level.

Sit with what that means for a calendar. If there is no wear-out age, then replacing or overhauling the item at a fixed age cannot reliably reduce the chance of failure. There is no rising risk to get ahead of and no moment to get ahead of it. The task can only consume capacity — and, if it is intrusive, do something worse.

The intervention is neither free nor risk-free

Here is the mechanism the compliance number is structurally blind to. An intrusive intervention does not literally reset a component's material age — but it does move a stable asset back into a period of elevated post-maintenance risk, because disassembly, replacement, adjustment and reassembly each introduce fresh opportunities to get something wrong.

This is measured, not theorised. Saleh and colleagues, analysing US civil helicopter accidents from 2005–2015 in PLOS ONE, found that flawed maintenance and inspection were causal factors in 14–21% of accidents — and that 31% of those maintenance-related accidents occurred within the first 10 flight-hours after the maintenance was performed. That is an early-life failure signature, arriving not when the part was new, but when the work was. The most common error category was improper or incomplete (re)assembly or installation of a part, at 57% of cases. Hands on a working machine is not a neutral act.

And the intervention has two more costs that never appear next to the compliance percentage: it takes the asset out of production to gain access, and it consumes the scarcest thing the operation has. Skilled-labour shortages have run at historically high levels across advanced economies for most of the past decade, and structural forces are keeping them tight (OECD). Every hour spent stripping a pump that was never going to fail is an hour not spent on the asset that was.

The paradox, stated as a mechanism

So the chain runs: a blanket calendar schedule generates interventions on failure modes that are not age-related → those interventions consume scarce technician hours and production access → some of them introduce defects, because reassembly is where maintenance errors concentrate → and the capacity they consumed is capacity the genuinely high-consequence assets did not get.

More maintenance. Less reliability. Compliance at 96% the entire time — because every one of those tasks was, in fact, completed on schedule.

Challenge the task, not the plan

None of this is new thinking, and it would be dishonest to present it as ours: it is the core of RCM, and it has been available since 1978. What is new is how rarely the questions get asked of an existing PM portfolio. Most schedules were inherited — from an OEM recommendation, a previous plant manager, an incident nobody wants to repeat — and then never re-examined, because the compliance number kept coming back green.

It takes fewer questions than people expect. Start with the one most PM reviews never ask: has this task ever actually caught anything? If nobody can show that it has — or nobody knows — that is your first finding, not a reason to keep going. Then just two more, answered from failure history rather than the manual: if it fails, does it hurt? And does age predict the failure? Those two answers place every task, and the placement carries a policy.

That is the fast version. Underneath it sits the evidence you would actually examine — the consequence, whether there is a wear-out age, whether the failure is detectable, whether the task has ever prevented anything, how intrusive it is, and what it costs in scarce capacity. Lay that out across a portfolio and the policy stops being a matter of opinion.

Notice what falls out. A task can be worth keeping (bearing lubrication: genuine wear, cheap, non-intrusive). It can be worth replacing with a sensor — an annual strip-down of a pump whose failures are not age-related but are detectable by vibration is a task that should become a measurement. It can point at a redesign rather than a task at all: if the consequence is high, there is no warning and no age relation, the honest answer is redundancy or a spare, not a calendar. It can be worth stopping. And sometimes the finding is simply that nobody knows why the task exists — which is itself a result worth having.

A caveat on the tool itself, and it matters: this is first-pass triage. It tells you which tasks deserve a deeper challenge first. It is not a substitute for a failure-mode-level RCM analysis — function, functional failure, failure mode, consequence, task feasibility — which is what you do to the tasks the triage flags.

This is also where technology gets misdirected. Condition monitoring is not automatically the answer; it earns its place only where a developing failure is genuinely detectable far enough ahead to act, and where the sensing costs less than the failure. Sensors on an asset with a random, undetectable failure mode buy you telemetry, not reliability. Let the failure mode choose the instrument.

When preventive maintenance is exactly right

This is not an argument against PM. It is an argument against PM by default. Where failure genuinely is age-related — wear, fatigue, corrosion, consumables — a calendar interval is precisely the right instrument, and the 11% is not a small number when it contains your most consequential assets. Statutory and safety-mandated tasks sit outside the keep-or-remove decision — they are done because they are required. They can still be executed more efficiently: bundled, re-sequenced, or run at the permitted interval rather than a habitual one.

The figures deserve their own caveat. Nowlan and Heap measured proportions of failure modes in civil aviation in 1978 — not failure frequency, not consequence, and not your industry. Later studies find different distributions across sectors. The number to take from it is not 11%. It is the ranking: a usable wear-out age is the minority case, and a schedule built as though it were the majority case will spend most of its effort where it cannot help.

Start with the tasks you cannot justify

You do not need to rebuild the maintenance plan to start. Take the twenty tasks that consume the most technician hours, and ask those three questions of each. A first review typically separates three groups: tasks that clearly earn their place, tasks that should become a measurement rather than an intervention, and a tail whose original justification can no longer be demonstrated. That tail is where the capacity for the assets that actually matter is hiding.