The Changeover Process Nobody Times and Everybody Blames

The capacity gap with three suspects

The plant is behind. Orders are shipping late, overtime is up, and the monthly review produces the same three explanations it produced last quarter. Scheduling says the customer order pattern has too many short runs. The supervisors say the new operators are slow. Sales says the plant should be able to run what it sells.

All three may be partly right. None of them can be checked, because the one loss that sits between all three explanations is the one nobody in the building measures: the changeover.

Ask a plant manager how long a changeover takes on their bottleneck machine and you will get an answer, usually confident, usually a single number. Ask for the distribution of changeover times on that machine last quarter, by crew, and the room goes quiet. That silence is the finding.

How a plant removes its own ability to see

Changeover is unusual among the major losses. It sits inside somebody's budget, because labor and machine hours are paid either way. And it sits outside everybody's metric, because of a classification decision made years ago and never revisited.

Many plants class changeovers as planned downtime. Planned downtime is subtracted before availability is calculated. The effect is that a changeover that runs ninety minutes instead of forty does not move the OEE number at all. Symestic, which sells OEE software and states it plainly, notes that changeover is planned downtime in some OEE definitions and unplanned in others, but that in every credible calculation it counts against the machine. The phrase doing the work there is "credible." Plenty of plants report a number that is not.

Once the loss is out of the metric, it is out of the conversation. Nobody is asked to explain it in the morning meeting. Nobody's bonus moves with it. The capacity it consumes still has to come from somewhere, so it shows up as a schedule that never quite holds, and the schedule gets the blame. We see the same move with exceptions in manufacturing: when a loss is reclassified as normal, the organization stops paying attention to it and starts paying for it instead.

Why the SMED workshop is not the fix

The standard answer to long changeovers is a SMED event. Bring in a facilitator, film a changeover, separate the internal steps from the external ones, stage the tooling, standardize the sequence. The method is sound and the history behind it is real. Shigeo Shingo's work at Toyota took a stamping press changeover that ran several hours in the early 1950s down to minutes by the end of the 1960s.

Workshops routinely produce large first-pass reductions. Symestic's own examples include a stamping press going from 120 minutes to 12 and an injection molding changeover from 75 minutes to 18. We have no reason to doubt that numbers like these are achievable in the week of the event.

The problem is the month after. And the six months after that.

iFactory, another vendor with a product to sell in this exact space, describes the failure honestly: the most common root cause of SMED regression is that changeover time stops being measured once the workshop concludes. New operators join without training on the improved sequence, small shortcuts creep back in, and nobody notices the drift until times have crept most of the way back toward the original baseline. Their estimate is that this happens within the first six months.

Read that again from the buyer's side. The workshop fixed a sequence. It did not fix the plant's ability to know whether the sequence is still being followed. Crews turn over, a fixture goes missing, the staging cart becomes a storage cart, and the gain decays quietly because the instrument that would have caught it was never installed. The improvement was real. It just had a shelf life nobody was tracking.

So the plant runs another workshop two years later, gets the same impressive before and after slide, and concludes that SMED works. It does. The plant just keeps buying it twice.

You do not have a changeover problem yet

Here is the uncomfortable test, and it is the one we would put in front of any operations leader before they spend a dollar on changeover improvement.

Can you produce last quarter's changeover times as a distribution, by machine and by crew?

Not an average. Not the standard from the routing. The actual spread: the fastest quartile, the median, the long tail. If you can, you have a changeover problem and you can start working on it, because the distribution will tell you where it is. A tight distribution with a high median is a method problem, which is where SMED genuinely earns its fee. A wide distribution with a reasonable median is a consistency problem, which is about crews, staging and handoffs. A long tail concentrated on a few product transitions is a sequencing problem, and it belongs to scheduling.

Those are three different interventions with three different owners. A plant that cannot tell them apart will pick one based on who argues hardest in the review meeting.

If you cannot produce the distribution, you do not have a changeover problem yet. You have a measurement problem, and every other conversation is premature.

What measurement actually means here

This is not a pitch for a monitoring platform, though some plants will end up buying one and some of those will be right to.

Continuous measurement per transition means every changeover gets a start, an end, a machine, a crew and a from-to product pair, and someone looks at the result on a fixed cadence. In a low-mix plant that can be a log at the machine and a weekly review. In a high-mix job shop running dozens of transitions a day, manual capture tends to fail the same way paper quality data fails, which we covered in why quality data on paper is worth less than the time spent capturing it: operators learn quickly whether anyone reads the numbers, and record accordingly.

The definition questions matter more than the tool. Where does a changeover start: last good part, or machine stop? Where does it end: first part produced, or first good part? Is the first-article wait included? Two plants in the same company will often answer these differently and then compare their averages in a board deck. The numbers will not be comparable, and the plant that chose the friendlier definition will look better for it.

The from-to pair is the field most plants skip and the one that pays most. A changeover is not one event type. Going from a light product to a dark one, from a short run to a long one, from one material family to another, can differ by a factor of several. Without the pair, sequencing decisions get made on folklore. With it, scheduling stops being the default suspect and becomes a lever.

Who should own the number

Measurement without an owner decays at roughly the same rate as an unmeasured SMED improvement. Somebody has to be asked, weekly, why the distribution moved.

In most plants this lands awkwardly. Production owns the crews. Engineering owned the workshop. Scheduling owns the sequence. Finance owns the labor variance and sees none of the detail. The changeover time is everybody's input and nobody's output.

We think the right owner is whoever owns throughput on the constraint, because that is where a changeover minute is worth the most. A minute lost on a machine with spare capacity is a minute of idle labor. A minute lost on the bottleneck is a minute of output the plant never gets back. Measuring every machine equally is a reasonable place to start. Managing every machine equally is not.

This also changes how changeover cost flows into the commercial side. If the plant does not know what a transition costs, the quote does not either, which is the problem we described for co-manufacturers in losing money on changeovers they quoted correctly.

What this does to the investment decision

Plants approach changeover with one of three purchases in mind: a SMED program, a scheduling tool, or a monitoring system. Each vendor will show you a case where their piece was the answer. Sometimes it was.

The order matters more than the choice. Measurement first, because it tells you which of the other two you need. A scheduling optimizer fed with standard changeover times will sequence beautifully against numbers that are not true. A SMED program without ongoing measurement will produce a great slide and a gain that fades. A monitoring system nobody reviews will produce a dashboard that is accurate and ignored.

The cheapest version of the right answer is often an honest definition, a log, an owner and a weekly fifteen minutes. That is not a satisfying thing to buy. It is, however, the thing that makes every later purchase worth what it cost.

Where this stops being an article

We cannot tell from here whether your capacity gap is a method problem, a consistency problem or a sequencing problem. That takes pulling what changeover data exists, walking the transitions on your constraint with the crews who run them, and checking how your plant's definitions compare with the ones in your reports.

If your plant keeps arguing about whether the schedule, the crews or the customer is to blame, and nobody can put a distribution on the table, talk to us. We will help you find out where the changeover time is actually going before you pay to reduce it a second time.