What breakdowns cost, and what the evidence actually supports.

The headline numbers

The most quoted figure comes from Siemens, which estimates that the world's 500 biggest companies lose almost $1.4 trillion a year to unplanned downtime, about 11% of their revenues.^⁠[1] It is a vendor's estimate, extrapolated from 181 online interviews with maintenance, engineering and IT staff at large manufacturers in four sectors between 2019 and 2023. It tells you downtime is expensive for very large plants. It does not tell you what yours costs, and it was not designed to.

Most downtime statistics in circulation are like this: surveys run by vendors, figures repeated until their origin is lost. Below are the ones we could trace to an original with a stated method, and what they actually support.

The strongest number we found

In 2020 the US National Institute of Standards and Technology published an estimate of what maintenance costs American discrete manufacturers, and what poor maintenance costs them on top.

US$57.3 bnMaintenance spend

Machinery maintenance spending by US discrete manufacturers in 2016.^⁠[2]

US$119.1 bnLosses from preventable maintenance issues

Of which $18.1 billion was downtime and $100.2 billion lost sales from delays and defects. Same population, same year.^⁠[2]

3.3×Downtime, most reactive vs least

Plants in the top quarter for reliance on reactive maintenance were associated with 3.3 times the downtime of those in the bottom quarter.^⁠[2]

16×Defects, same comparison

The same plants were associated with 16 times more product defects.^⁠[2]

Read these with their caveats. The survey returned 85 responses, of which 71 were used. The downtime estimate carries a 90% confidence interval of $9.4 billion to $29.5 billion, and NIST notes that many of its sub-group comparisons are not statistically significant. The fair reading is an association: plants that run mostly reactively also report much more downtime and more defects. It does not prove that changing one changes the other by the same amount.

Where the cost of a pump really sits

The pump industry's own life-cycle guidance, produced with the US Department of Energy, is plain that 'the initial purchase price is a small part of the life cycle cost for high usage pumps', and that 'depending upon the process, down time costs can be more significant than the energy or maintenance elements'.^⁠[3] It also says something every maintenance manager should take personally:

The Department of Energy's pumping-system guide lists 'frequent replacements of bearings and seals' among the five common signs that a pump is oversized for its duty, and warns that 'frequently replacing pump seals and bearings can keep a maintenance crew so busy that they overlook the system operating conditions that are causing most (or all) of the problems'.^⁠[4] That is the cost of a repeat failure in one sentence: not the seal, but the cause nobody had time to find. The fault that keeps coming back shows what that looks like on a machine's record.

Deferred work does not stay the same size

The US Government Accountability Office reported in 2025 that federal deferred maintenance and repair backlogs had more than doubled, from $171 billion to $370 billion, between 2017 and 2024, and warned that assets left to deteriorate 'need premature replacement, which can be significantly more expensive than if maintenance and repairs were done when originally scheduled'.^⁠[5]

The UK's offshore regulator found the same pattern at the level of individual installations. Inspecting nearly 100 offshore installations, it reported 'high levels of both safety- and non-safety-critical backlog and poor backlog management' across the industry, and maintenance records whose quality was 'often inconsistent, particularly in recording the "as found" condition rather than the "maintained" condition and in reporting the cause of failure'.^⁠[6] A backlog without causes is a list of jobs. A backlog with causes is a reliability plan.

When the record itself is wrong

In 2010 a gas transmission pipeline ruptured in San Bruno, California. Eight people were killed and 38 homes destroyed. The company's records described the failed section as pipe with no welded seam; investigators found it had seam welds. The US National Transportation Safety Board concluded the operator's integrity management program was deficient in part because it 'was based on incomplete and inaccurate pipeline information'.^⁠[7] An asset register is not paperwork. Every decision about an asset inherits its errors.

The first shift after a service

Maintenance itself carries risk, and some of it is measurable. The NSW Resources Regulator reviewed 18 fires on mobile plant at open-cut coal mines between April and October 2021. Nine happened within the first shift after the plant was returned to service, and ten came from workmanship errors 'likely to have been detected by visual inspection by a supervisor or a short period of workshop commissioning'.^⁠[8] A return to service is a decision with a cost attached either way, which is why it should be someone's decision, on record. See From a failed check to a safe return to service.

Figures we looked for and could not stand behind

These are widely repeated. We could not trace any of them to an original source with a stated method, so we do not use them:

• That reactive maintenance costs three to five times as much as planned maintenance.
• Percentage savings from moving off reactive maintenance, often quoted at 12–18%, which trace back to unnamed 'studies'.
• A fixed share of pump failures caused by seals.
• The share of a technician's day spent searching for information. We found no maintenance-specific measurement with a stated method.

What to measure on your own plant instead

The most useful downtime figure is your own, and the questions that produce it are simple to ask and hard to answer without a connected record:

• Which ten machines took the most repair jobs last year, and how many of those jobs were the same fault coming back?
• How old is your oldest open defect, and what is the machine doing in the meantime?
• What does an hour of downtime on your most critical asset cost, in your own production numbers?
• When a machine comes back from a service, who checks it before the first shift, and where is that recorded?
• Could you tell twenty new seals on one pump from twenty new seals across several pumps? The UK regulator's ageing-plant guidance says a maintenance log should be able to.^⁠[9]