What counts as changeover time?

The strict definition is wall to wall: from the last good unit of the previous product at rate to the first good unit of the next product at rate. That includes teardown, setup, dial-in, trial runs, quality approval waits, and ramp. Softer definitions hide real losses in other buckets.

Why do recorded changeover times differ from actual times?

Because crews log the span afterward from memory, definitions vary between shifts, and the expensive tail of a changeover, dial-in, trial scrap, and waiting on quality, tends to get booked as rate loss, scrap, or idle time instead. Wall-to-wall measurement with step timestamps closes the gap.

How do you track changeovers in real time?

Break each changeover into eight to twelve steps, run them as a digital checklist on a tablet at the line so each step timestamps itself, and use machine run-state signals to mark last unit and first unit automatically. Show the live clock against the standard for that product pair on the same board as running lines.

Does changeover tracking replace SMED?

No, it feeds it. SMED's classic first step is filming and timing one changeover; continuous step-level tracking gives you that analysis for every changeover, across all crews and shifts. The improvement work, externalizing steps, staging tooling, parallel tasks, is still SMED. Tracking tells it where to aim and proves whether fixes held.

What is a reasonable number of changeover steps to track?

Eight to twelve per product family works for most lines. Fewer than that and the big losses hide inside broad steps; many more and the capture burden grows without adding much insight. Define the steps with the crews who run the changeover, not for them.

What is the fastest payoff from changeover visibility?

Clearing blocks in the moment. The first thing live tracking exposes is how much of a changeover is waiting, for a quality tech, a missing tool, a material delivery. Those waits get cleared within minutes once they are visible to a supervisor, which typically recovers time from day one before any SMED work starts.

Visibility Into Changeovers: The Invisible Hour

Between the last good unit and the first good unit, most plants go dark. That hour deserves a record.

Changeover visibility means knowing, live and in detail, what happens between the last good unit of one product and the first good unit of the next. On most floors that span is a black box: the clock says an hour passed, and nobody can say where the hour went.

Production tracking has a strange blind spot. While the line runs, there are counts, rates, and reasons. Then a changeover starts and the record goes dark. Forty, sixty, ninety minutes later the line comes back and the record resumes, and the whole span shows up in the log as one number, if it shows up at all. Yet on many lines changeover is the single largest category of planned downtime, and the least examined. This post is about turning that invisible hour into a visible one: what actually happens inside a changeover, why the recorded time is usually fiction, and how to instrument the span so it can finally be improved.

What happens inside the invisible hour?

A changeover is not one activity. It is a chain of a dozen, and each link can stretch without anyone noticing, because no link is measured on its own. A typical chain: run out the last order, clear and clean, tear down tooling, walk to fetch the next tooling, discover a missing part, find the part, set up, load material, dial in, run trial units, wait for quality approval, ramp to rate. The wall clock records the total. It records nothing about the links.

That matters because the links fail differently. The teardown might be tight while the dial-in bleeds twenty minutes every time. The setup might be fast while the crew loses fifteen minutes waiting for a quality check because the tech is on another line. Without link-level visibility, improvement efforts guess, and they usually guess wrong, because the loudest step is rarely the longest one. This is the exact problem SMED was built to attack, and SMED's first step, filming and timing the changeover, is a manual, one-time version of what live tracking does continuously.

The chain inside a changeover. The longest links are usually waits and hunts, not wrench time, and the log never shows them.

Why is the recorded changeover time usually wrong?

Three reasons, all structural. First, the definition drifts. Some crews log changeover from machine stop to machine start; others stop the clock at first unit, good or not. Unless everyone measures last good unit to first good unit at rate, the numbers are not comparable across shifts, and the friendliest definition tends to win. Our post on changeover time measurement covers the definitions in detail.

Second, the record is written afterward, from memory, by a crew that just spent an hour under pressure. A 68-minute changeover becomes 45 because the last 23 minutes of dial-in felt like running. Nobody is lying; the span is genuinely hard to reconstruct, which is exactly why it should not be reconstructed.

Third, the losses around the changeover get booked elsewhere. The slow ramp after first unit gets logged as a rate loss. Scrap from trial units gets logged as quality. The wait for a quality tech becomes idle time. The changeover looks better than it is because its costs are scattered across three other buckets. Plants that consolidate those costs, and the changeover savings calculator is built to help do the math, are usually surprised by the total.

What does live changeover visibility look like?

It looks like the changeover having the same status as production. Concretely, four things are visible while the changeover is happening:

State. The line is in changeover, which step it is on, and who is on it, shown on the same live view as running lines. A supervisor glancing at the board sees "Line 3: changeover, step 4 of 9, 12 minutes ahead of standard."
Clock against standard. Elapsed time versus the standard for this specific product pair, because an A-to-B changeover and a B-to-A changeover are rarely the same. This is where changeover sequencing gets its data.
Blocks. The moment the crew is waiting on something external, a quality tech, a missing tool, a material delivery, that wait is flagged as a block, live, so someone can unblock it now instead of reading about it Thursday.
The record building itself. Step timestamps accumulate as the crew works through a digital checklist, so the post-changeover data entry step does not exist. Digitized changeover checklists are the capture mechanism that makes this honest.

The first payoff is boring and immediate: blocks get cleared faster because they are visible. The second payoff compounds: after thirty changeovers you have step-level times across crews, shifts, and product pairs, and the improvement targets pick themselves.

How do you make changeovers visible?

The sequence that works, in order:

Fix the definition first. Last good unit at rate to first good unit at rate, wall to wall, no exceptions. Write it down and put it on the board. Every later number depends on this.
Break the changeover into steps. Eight to twelve steps per product family, defined with the crews who run them. Too few hides the losses; too many burdens the capture.
Digitize the checklist. The crew works through the steps on a tablet at the line. Each tap timestamps a step. Capture cost per changeover: a few seconds.
Add machine state where it helps. Run-state signals from the PLC mark last unit and first unit automatically, bracketing the human record with machine truth. This is the same wiring that powers real-time production tracking generally.
Show the changeover live. Put it on the same board as running lines, with clock-versus-standard and any active blocks. A changeover that is visibly 15 minutes over standard at minute 40 gets help; one discovered tomorrow gets a comment.
Review step data weekly, then improve. Pareto the steps across the last month of changeovers and hand the top loss to the improvement process. This is where SMED restarts, now with continuous data instead of one filmed changeover.

Recorded changeover time flatters the changeover because its costs scatter into other buckets. Wall-to-wall measurement collects them.

What can you do with changeover data once you have it?

Three things, in rising order of value. First, clear blocks in the moment, which typically claws back minutes per changeover from day one. Second, feed improvement: step-level Paretos aim changeover loss reduction at the actual biggest loss, and before-and-after data proves whether a fix held across all crews rather than just the one that was filmed. Third, feed the schedule: once changeover times per product pair are real numbers instead of folklore, changeover-aware scheduling can sequence the week to minimize total changeover time, which is often worth more than making any single changeover faster.

The stakes behind this are set by the wider environment, and the primary sources frame it well:

U.S. manufacturing capacity utilization has generally run in the mid-70s percent range in recent years, per the Federal Reserve's G.17 Industrial Production and Capacity Utilization release. Time recovered from changeovers is capacity that requires no capital.
Manufacturing labor productivity growth has been modest for over a decade, per the BLS productivity program, which puts the improvement burden on running existing lines better.
The BLS Job Openings and Labor Turnover Survey has shown persistently high manufacturing job openings in recent years, in the hundreds of thousands monthly. Crews are scarce; an hour of skilled crew time spent hunting tooling is the most expensive kind of waste.

There is an honest caveat. Visibility does not shorten a changeover by itself; it shows where the hour goes and keeps everyone honest about the total. The shortening still comes from SMED work, staged tooling, parallel tasks, and practiced crews. Plants that instrument changeovers and then never review the step data have bought a very accurate record of a problem they are not working on. The record is the map, not the journey, and a standing weekly review habit is the discipline that turns one into the other.

How does Harmony AI make changeovers visible?

Harmony AI, an AI-native MES layer, treats changeover as a first-class production state. Crews run digitized changeover checklists on tablets at the line, machine signals bracket the span automatically, and the live plant view shows every changeover's step, clock against standard, and active blocks alongside the running lines. Because Harmony AI also connects the ERP schedule, changeover standards per product pair feed sequencing decisions, and its AI agents can flag a changeover trending over standard and notify the right person while there is still time to help, every action cited and approvable. Deployment starts with our team on-site walking your changeovers with the crews, and it layers onto the systems you already run. No rip-and-replace. See how the broader visibility layer came together at CLS in Chattanooga, and if you want to know what your changeovers are worth, start with the changeover and SMED savings calculator. For the wider argument this series makes, read closing the visibility gap.