SiteThe board pack lands on the Monday. Output, OEE, scrap, overtime, a red cell for the line that missed its target. There's a line about downtime. Fourteen hours last month on the bottling line.
Nobody asks how long each stop took to fix, or why.
Take one of those stops. The drive on the infeed conveyor tripped at 9pm on a Thursday. The line was back at rate by 1am. Four hours down. The actual repair, from spanner on to guard back on, took 40 minutes.
So where did the other three hours and twenty minutes go? That's the question mean time to repair should make people ask. On most sites, nobody asks it.
Four hours, forty minutes of work
Break that Thursday night down and it looks something like this.
- 9.00pm: drive trips. The operator resets it twice. It trips again.
- 9.20pm: shift manager rings the on-shift fitter, who's on another line.
- 9.45pm: fitter arrives, reads an overcurrent fault, checks the motor and the load. Can't find it.
- 10.30pm: fitter rings the electrician who's off. No answer. Tries the controls engineer. They can come, but they're 50 minutes away.
- 11.30pm: controls engineer arrives and goes online to the drive.
- 11.50pm: finds a parameter changed after last week's swap. Fixes it. Test run.
- 12.30am: line back at rate after restart scrap and a clean-down.
The repair was 40 minutes. The wait for the right person was nearly two hours. That pattern is the one to look for in your own logs. Our view is that on most sites, the biggest chunk of downtime is the time spent finding someone who can do the fix.
Working it out without a consultant
Mean time to repair is total downtime from breakdowns divided by the number of breakdowns. Ten stops, 30 hours down, gives you three hours each.
The sum is easy. The hard part is agreeing when the clock starts and stops. We'd start it when the machine stops making product. We'd stop it when it's back at normal rate, not when the fitter walks away. If you stop the clock at "repair done", you hide the restart, the clean-down and the first-off checks. Pick one rule and stick to it, or the number will drift every month.
Then split it. A single figure for the whole site tells you very little. Split MTTR by:
- shift, because nights are nearly always slower, for reasons we cover in why breakdowns on nights take longer
- trade, so mechanical, electrical and controls faults sit apart
- stage, so you can see wait time, diagnosis time, parts time and repair time
The stage split is the one that changes decisions. You can only do it if the log records times as well as fixes. If yours doesn't, a breakdown log that actually tells you something shows what to capture.
One more thing. Watch the average. One 20-hour gearbox job can make a good month look awful. Look at the middle value too, and list the long ones by name.
Put a price on the wait
An hour on the stage list means nothing to a finance director until it has a pound sign next to it. Work out your true cost of an hour of downtime for your worst lines. Then multiply it by the hours spent waiting for the right engineer last quarter.
That number is usually the one that gets a board's attention. It's money spent on nothing. No parts, no labour, just a stood line and a phone that nobody answers. Put it next to what a second electrician or a standby callout would cost, and the case often makes itself. If you need to take that to the board, making the case for another engineer walks through it.
What the board should be asking
HSE's guidance for directors says the board should receive "both specific (eg incident-led) and routine reports" on how health and safety is going (INDG417). Breakdowns belong in that picture. A machine that keeps failing puts people near moving parts more often. HSE warns that extra hazards can occur "when machinery becomes unreliable and develops faults" (HSE, maintenance).
So we'd want a board to see four things each month:
- MTTR for the worst five machines, split by shift
- The average wait from stop to the right engineer starting work
- How many stops waited on a part
- Repeat failures: the same fault on the same machine inside 30 days
None of those are hard to pull from a decent log. Together they say more than OEE does about whether your maintenance team has the people and the spares it needs. For the wider set of numbers, see maintenance KPIs that mean something.
Don't let the number push the wrong fix
There's a trap here. Once MTTR is on the board pack, people will try to make it smaller. The quickest way is reset and run. A two-minute reset counts as a two-minute repair. Do that 30 times and your MTTR looks brilliant while the machine gets worse. We've written about why reset and run is the most expensive habit on site.
HSE's code of practice for PUWER makes a point worth keeping in mind. "A fault which affects production is normally apparent within a short time." A fault in a safety system "could remain undetected" unless maintenance checks for it (L22, paragraph 71). MTTR only counts the faults that stop the line. And when PUWER asks for kit to be kept "efficient", the code says that is about safety: "It is not concerned with productivity" (L22, paragraph 70).
So treat MTTR as a production number that points at a staffing and spares problem. Never let it become a reason to rush a job or skip a safety check.
What to do on Monday
- Pull last month's breakdowns for your three worst machines. For each one, write down when it stopped, when the right engineer started, and when it was back at rate.
- Add up the hours spent waiting for the right person, and multiply by your cost per hour of downtime.
- Agree one rule for when the MTTR clock starts and stops, and put it in writing.
- Ask for the four numbers above to go in next month's board pack.
* Paragraph numbers refer to L22, fourth edition (2014, amended 2018).