What unplanned downtime really costs a plant — and how to see it coming

Ask a plant manager what an hour of downtime costs and you will usually get one of two answers: a confident figure that turns out to cover only lost output, or an honest shrug. Both are understandable. The cost of a stopped machine is spread across production, maintenance, purchasing and sales, and no single report adds it up. But it is worth adding up, because the number is almost always larger than people expect — and because a large share of it is avoidable.

The number at the largest plants

The best-known public estimate comes from Siemens' annual survey of large manufacturers, The True Cost of Downtime 2024. Its headline: unplanned downtime costs the world's 500 biggest companies about $1.4 trillion a year — roughly 11% of their revenues. An hour of lost production ranges from about $36,000 in fast-moving consumer goods to $2.3 million in a large automotive plant.

The same survey found that an average large plant still suffers around 25 downtime incidents and 27 lost hours a month — fewer than in 2019, but still more than a full day's production every month. The reduction, the report notes, came largely from plants that started monitoring the condition of their machines instead of waiting for them to fail.

Those figures describe Fortune Global 500 operations. A mid-sized plant will see smaller numbers per hour, but not a smaller share of the damage: one stopped pump or compressor can still stop the whole line, and a smaller company has less slack to absorb it.

What an hour of downtime is actually made of

To estimate the cost at your own site, list the components rather than reaching for a single figure. Most of them are easy to find in data you already have:

• Lost output — the margin on what the line would have produced, not its sales value. If the lost volume can be made up later on overtime, count the overtime instead.
• Idle labour — operators and shift staff who are paid while the line stands.
• The repair itself — technician hours, contractor call-outs, and overtime because breakdowns rarely happen at a convenient time.
• Emergency parts — express freight and premium prices for a part bought in a hurry, or the full lead time if nobody stocks it.
• Scrap and restart losses — material ruined when the line stopped mid-batch, and the off-spec production while it comes back up to speed.
• Contract penalties and lost customers — late-delivery clauses, and the harder-to-count cost of a buyer who quietly moves to a more reliable supplier.
• Secondary damage — a bearing that fails outright can take a shaft, a seal or a motor with it, turning a cheap repair into an expensive one.

A simple way to put it together: annual cost = hours of unplanned stoppage × cost per hour, where cost per hour is the sum of the first two items, and the remaining items are added per incident. For example — purely as an illustration — a plant with 200 hours of unplanned stoppage a year at 30,000 an hour in margin and idle labour loses 6 million a year before a single part or penalty is counted. Your own figures will differ; the point is that they are worth writing down.

The most expensive part of a breakdown is often not the repair. It is the part that takes three weeks to arrive.

Why most failures are predictable

Rotating and thermal equipment — pumps, fans, compressors, motors, gearboxes, conveyors, furnaces — rarely fails without warning. Reliability engineers describe the gap between the first detectable sign of a problem and the actual breakdown as the P–F interval: from potential failure to functional failure. For a degrading bearing or a misaligned shaft that interval is typically measured in weeks, not minutes.

During those weeks the machine is telling you something. Vibration rises as a bearing wears or a rotor goes out of balance. Temperature climbs as friction increases or cooling degrades. Motor current drifts as the load changes or a pump starts to cavitate. Pressure and flow sag as an impeller erodes. None of these signals is exotic, and on most modern equipment several of them are already measured — by the PLC, the variable-frequency drive or the SCADA system — and then never looked at again.

That is the real gap. The data exists; what is missing is someone, or something, watching the trend every hour of every day, across every machine, and acting on it.

From a warning to a part on the shelf

Seeing a trend is only half the job. A warning that ends up in an inbox prevents nothing. To turn early signals into avoided stoppages, a plant needs four steps to happen without anyone having to remember them:

• Connect — read what controllers and drives already measure over standard industrial protocols, and add sensors only where a critical machine has none.
• Forecast — project each trend forward to estimate when it will cross its alarm limit, with an honest range rather than a single date.
• Plan — compare that date with the supplier's lead time for the part and with the next window when the line is not producing anyway.
• Act — open a work order with a checklist, send it to the right technician's phone, and order the part early enough that it arrives before the failure does.

The third step is the one most monitoring tools skip, and it is where much of the money is. If a bearing will cross its limit in nine days and the supplier needs five, the order has to go out now. If the line already stops for planned maintenance next Tuesday, that is when the job should be done — not during Thursday's peak shift.

Where to start

Predictive maintenance does not need to begin with the whole plant. A practical first step is to pick the three to five machines whose failure stops a line, list what is already measured on them, and agree in advance what success would look like — fewer unplanned stops, earlier warnings, parts ordered before the failure rather than after. A few weeks of observation on those machines will tell you more about your real cost of downtime than any industry average.

This is the problem Nodal Dynamics was built for. It connects to existing PLCs, drives and SCADA over open protocols such as OPC UA, Modbus and MQTT, builds a live model of the equipment, forecasts when each trend will cross its threshold, and turns the warning into a work order and a spare part ordered in time — with purchasing drafts for 1C and SAP PM. It can run on the plant's own server, including fully isolated networks, and every number it shows can be traced back to the readings behind it.

We are early, and we would rather show than claim: Nodal Dynamics has a public demo plant and offers pilots of five to seven weeks on three to five critical machines, with success criteria agreed before the start. If you know what an hour of downtime costs your plant — or want to find out — talk to us.