Changeover Time: The Hours That Appear in No Costing

The symptom

Almost every manufacturer has a number for changeover time. It sits in the ERP, it feeds the costing, it determines batch sizes and delivery dates. And in almost every business, nobody can say precisely where it came from.

Usually it was estimated or measured once, years ago, under conditions that no longer exist: different tooling, different parts, different people. It has been carried forward since, because it looks plausible and because changing it starts an argument with sales.

Running alongside it is a second, unwritten number: what the people at the machine consider realistic. Everyone on the shop floor knows that figure, and it is recorded nowhere. The gap between the two numbers is the interesting part.

The mechanism

Changeover isn't a single operation, it's a chain. Fetch the tooling, strip the old fixture, clean, mount the new fixture, set the datum, load the program, run a first article, measure, correct, obtain release. The documented changeover time generally measures only the middle block, the actual physical conversion at the machine.

What is systematically missing are the edges. Hunting for tooling that wasn't returned to its place after last use. Waiting for the setter, who is currently standing at another machine. Waiting for measurement release, because the inspection room is occupied. The second and third first article, because the first came in outside tolerance.

These edge pieces share a property: they aren't the setter's fault, and so they don't get recorded as changeover time in the confirmation. They vanish into "other" or into nothing. The machine is standing still regardless.

An uncomfortable chain follows. Understated changeover times make small batches look attractive on paper. Small batches mean more changeovers. More changeovers mean more unrecorded edge time. The business plans itself, with entirely correct arithmetic, into a utilisation that doesn't exist, then wonders why delivery dates slip while the capacity calculation balances. The worked example on how to calculate capacity utilization shows how to recalculate the figure from booked machine hours, and how changeovers and breakdowns shift it.

The cost

An illustrative example with invented figures. Take a contract machining business producing turned parts, just under 60 employees, that struggles with chronic delivery delays alongside apparently free capacity. The costing assumes 45 minutes of changeover.

For three weeks the business measures machine standstill instead of analysing the confirmations, from last good part of the old order to first good part of the new. In the example the mean is 97 minutes. The median is 84, and the upper quartile runs over two hours.

The largest individual items aren't spectacular: 14 minutes average waiting for the setter, 11 minutes hunting for tooling, 19 minutes for first-article inspection and release. Together, almost as much as the entire costed changeover time.

Extrapolated across roughly 1,100 changeovers a year on four machines, the difference amounts to about 950 unplanned machine hours. At an internal rate of €78, that's some €74,000 of capacity that exists in the plan and not on the shop floor. A business in that position soon starts considering a fifth machine.

The fix

The first step is unglamorous: the business corrects the costing to the measured values. Internally this is the hardest decision, because in the short term it makes prices and delivery dates look worse. It also makes them honest, and everything else builds on that.

Then come the three largest edge items. They sit around the machine, not in it. Tooling gets fixed locations with shadow boards and a mandatory return before end of shift. First-article inspection for defined part families moves from the inspection room to a gauge at the machine. And changeovers are spread across the day so that one setter is never serving two machines simultaneously.

All three measures follow the same logic. Work that can be done while the machine is still running has no place inside the standstill. That idea is the starting point of Shigeo Shingo's SMED method.

The example assumes that measured changeover time then falls to an average of 61 minutes. More important than the mean is the spread: the upper quartile falls from over 120 minutes to 78. Predictability comes from less variation, not from better averages.

That separation is the core of a process analysis. It measures what actually happens instead of carrying forward what was once true. Fixed scope, fixed duration, fixed price.

The next step

Stand at a machine and time a single changeover: last good part to first good part. Compare the result with the number in your ERP. That one measurement costs you two hours and tells you whether a closer look is worth it. Our worked OEE calculation for one shift shows how the measured changeover feeds into the machine's overall score.