If you pulled your mid-year OEE numbers and found changeover losses parked in the same spot they were in January, you’re not alone. Most plants have a changeover time metric. Very few plants have a changeover process that’s actually engineered to shrink that number. The checklist lives on a laminated sheet taped to the machine, the sequence lives in the head of whichever setup tech has been there fifteen years, and the “SMED program” from three years ago produced a nice binder that nobody opens anymore.
The fix isn’t a new SMED workshop. It’s putting the checklist somewhere it can enforce sequence, timestamp every step, and force reason coding at the moment the delay happens — which means putting it in the MES, not in a document control system. Here’s how to actually build that.
Start by separating internal and external work — for real, not on paper
SMED’s core move, going back to Shigeo Shingo’s original framework, is splitting changeover tasks into internal setup (line must be stopped) and external setup (can happen while the line is still running the outgoing job). Almost every paper checklist claims to do this. Almost none actually enforce it, because nothing stops an operator from starting the internal steps five minutes early out of habit, or from realizing mid-changeover that a fixture wasn’t staged and now has to be fetched — with the line down.
When you rebuild the checklist in MES, the first job is an honest audit of every line item on the current sheet:
- True external tasks — staging tooling, pulling the next work order’s material, pre-heating, gathering fixtures, verifying the incoming recipe or BOM — anything that doesn’t require the machine to be down.
- True internal tasks — the physical swap, adjustment, or calibration that can only happen with the machine stopped.
- Fake internal tasks — steps that are currently done with the line stopped purely because that’s how it’s always been sequenced, but that could be converted to external with a small process or tooling change (duplicate fixtures, quick-disconnects, pre-set gauge blocks).
That third category is where most of the real time savings live, and it’s the whole point of doing this exercise before you digitize anything. Digitizing a bad sequence just gets you a bad sequence with a timestamp on it.
Build the workflow so the MES won’t let internal work jump the queue
Once you know which tasks are genuinely external, the MES workflow should structure the job so those tasks are pushed to the operator or setup crew before the current run ends — triggered off the schedule, not off the machine going idle. A well-built changeover workflow in most MES platforms (whether that’s a native workflow engine, a task-list module, or an integration with a paperless work instruction tool) should:
- Fire the external-task checklist automatically at a configurable lead time before the scheduled changeover (based on the current job’s remaining cycle count or scheduled end time, not a guess).
- Block the “start internal changeover” step from being marked complete — or in stricter setups, block the internal task list from even opening — until the external prerequisites are confirmed done.
- Require positive confirmation (checkbox, scan, or in higher-risk setups, a photo or sensor confirmation) rather than a blanket “select all” shortcut. If your checklist can be closed out in one tap, it will be, and you’ve bought yourself a compliance record with no actual compliance.
- Sequence internal steps in a fixed order when order matters (e.g., de-energize before tool removal, changeover of dies before alignment checks), rather than presenting them as a flat list an experienced tech can reorder from memory — because the whole point is designing the process so it doesn’t depend on that one tech’s memory.
Don’t skip the validation step
The step plants love to cut for speed — first-piece or first-article verification — is the one that causes the changeover to look “fast” on paper and then bleed time back out as scrap or micro-stops during ramp-up. Build the verification/first-piece check into the workflow as the last gated step of the internal sequence, not as a separate quality process that happens whenever QA gets around to it. If verification fails, the workflow should route back into a defined internal-adjustment step, and that time should still code to changeover, not silently disappear into an unplanned stop bucket.
Capture time by reason code, not just as one lump number
This is the piece most existing setups get wrong even when they’ve already digitized. A single “changeover” duration field tells you nothing about where the time actually went. Structure the reason coding around the task groups you already defined:
- External prep time (and specifically, external prep that ran late or wasn’t ready — this is a scheduling and materials problem, not an operator problem)
- Internal setup — mechanical/tooling change
- Internal setup — parameter/recipe load
- Adjustment and calibration
- First-piece/verification and rework of the first piece
- Waiting on external resources (crane, forklift, second operator, quality sign-off)
Tie each checklist step to its reason code on the back end so the timestamp between “step started” and “step completed” rolls up automatically. This is what makes the changeover event legible for root-cause work later, and it’s what feeds a real availability breakdown into OEE instead of a lump changeover number that everyone already ignores because it never changes.
Converting the paper sheet: a working template
For each line item on your current paper checklist, capture five fields before you build anything in the MES:
- Task description — as specific as the current sheet allows, ideally more specific.
- Internal or external — assign honestly, and flag any “fake internal” candidates for a follow-up kaizen or tooling fix.
- Trigger condition — what causes this task to appear in the workflow (schedule lead time, prior step completion, changeover start event).
- Confirmation method — checkbox, barcode/RFID scan, sensor tag confirmation, photo, or supervisor sign-off. Match the rigor to the risk; not every step needs a scan.
- Reason code mapping — which downtime/availability bucket this step’s elapsed time rolls into.
Run this exercise machine by machine, not as one generic template across the plant. A press changeover and a filler changeover have almost nothing in common structurally, and forcing them into an identical workflow is how digital changeover projects lose the floor’s trust in week one.
Set the KPIs up so the improvement is actually visible
The reason most plants can’t see changeover improvement in their OEE is that changeover time is buried inside the availability calculation as an aggregate, alongside unplanned downtime, with no separate trend line. Once your MES is capturing changeover by reason code, build reporting that tracks, at minimum:
- Changeover time by product-to-product transition pair — some changeovers are inherently harder than others, and averaging them together hides where the real waste is.
- Internal vs. external time split, tracked over time, so you can see the ratio actually shift as fake-internal tasks get converted.
- Adherence to sequence — how often the workflow gets overridden or steps get skipped, which tells you whether the process is actually being followed or just checked off.
- Changeover contribution to availability loss, reported as its own OEE sub-metric alongside unplanned downtime, not folded into it.
Review this monthly with the same rigor plants apply to scrap and unplanned downtime. Changeover improvement is slow and incremental by nature — it moves in minutes per event, not in one dramatic before-and-after — so if you’re not tracking it as its own trend line, the improvement work will look like it’s producing nothing even when it’s working.
What “done right” actually looks like
A mature MES-driven changeover workflow means the operator gets prompted for staging tasks before the line stops, can’t advance into internal steps without confirming prerequisites, follows a fixed sequence for the internal work instead of relying on memory, and every minute of the event gets attributed to a specific cause automatically. None of that requires exotic technology — it requires actually finishing the SMED analysis before you digitize, and resisting the urge to let the checklist be closeable with one tap. The plants that get real availability gains out of this aren’t the ones with the fanciest workflow engine. They’re the ones that did the unglamorous work of classifying every task honestly before they ever touched the MES configuration screen.
This article was written with the assistance of artificial intelligence. While we aim for accuracy, the information may be incomplete, out of date, or incorrect, and should be independently verified before you rely on it for any decision. It is provided for general information only and does not constitute professional advice.
