Key takeaways
See our guide to the system that helps track changeover data.
SMED has been well known for decades and is still under-deployed in mid-market manufacturing. The reason is not that the methodology is wrong; it is that the standard rollout looks expensive. Consultants come in, video record changeovers, run workshops, produce wall-sized analysis documents. The plant manager looks at the engagement cost and decides this quarter is not the quarter.
The unromantic version skips most of that overhead. It does not produce wall-sized documents. It produces a substantial reduction in the changeover time of one specific transition, often 30-50% from the separation step alone, achieved by one cross-functional pair over three weeks of part-time work. Repeat across the top five changeovers and the plant has captured the bulk of the available capacity.
The temptation is to pick the longest changeover, the one that takes four hours and feels worth fixing. That is usually wrong. A four-hour changeover that runs twice a month is 8 hours a month of available savings. A 25-minute changeover that runs 40 times a month is over 16 hours a month, and is usually easier to compress.
The right pick is the changeover with the highest total monthly time: duration × frequency. Pull the OEE data, sort, pick the top of the list. For most plants this is a within-family or family-to-family change, not the rare big changeover. The piece on manufacturing KPIs covers the underlying time-attribution.
The conceptual heart of SMED is this distinction:
For one specific changeover, walk through it once with a stopwatch. Not video, a stopwatch. Write down every task and tag it I (internal) or E (external). Most plants find that 30-50% of the tasks labelled as "changeover" are actually external work being done internally because nobody thought to move it.
The whole gain in this step is rescheduling. Tasks that were being done in the internal window because that is when they had always been done get moved to the external window before the changeover even starts. The line stops, the team executes only the actually-internal work, the line restarts faster.
This single step typically captures 30-50% of the changeover time. No tooling change, no spend, no consultant. The article on the preventive maintenance schedule covers a similar pattern in PM execution where the same separation applies.
Some tasks tagged as internal in step 2 are genuinely internal, the line has to be stopped for the task to happen. But often these can be partially converted: pre-heating a tool to its operating temperature while the previous batch is still running, calibrating a fixture off-line on a duplicate fixture, pre-loading the next recipe into the controller so the changeover only needs to activate it.
This is where small spend can pay back. A duplicate fixture, a tool warming station, a parameter pre-load script, each costs a few hundred or thousand currency-units, each cuts minutes from every changeover for years. The math is usually obvious once the changeover frequency is known. The article on work order management systems covers how these standing changes get captured as work-order procedures rather than informal practices that decay.
For one changeover, three weeks of part-time work:
The output is a new standard changeover with a measured time, a written procedure, and an owner. The piece on root cause analysis covers how to handle the inevitable regression, within three months, parts of the new procedure will have drifted, and the team needs a way to catch it.
For a typical mid-market plant, the unromantic SMED applied to the top five changeovers in one quarter produces:
None of these numbers come from a consultant. They come from one cross-functional pair, three weeks per changeover, applied through the quarter. Plants that do this consistently end up with all their high-frequency changeovers in the 12-18 minute range, fast enough that the changeover line on the OEE dashboard stops being a top loss category.
The unromantic SMED works on any line where the changeover events are captured in the OEE stream.
Where a unified OEE + CMMS platform helps is in two places: identifying the right changeover to start with (top by total monthly time, queried from data rather than guessed), and capturing the new procedure as a standing work-order template so the gain does not regress in month four.
Fabrico is built for that workflow. To see how the changeover-time picture would look against your top five changeovers, book a demo .
For 80% of changeovers, no. A stopwatch and a clipboard are enough to identify the rescheduling opportunities. Video helps for the most complex changeovers where the sequence is fast and contested between operators; for the typical mid-market changeover, video adds overhead without changing the conclusions.
That variability is itself a finding. It usually means the procedure has not been standardised, and the unromantic SMED's last deliverable, a written standard with an owner, is what fixes it. Plants with high operator-to-operator variance often gain even more than the typical 30-50% just from standardisation, before any sequencing change.
The written standard goes into the CMMS as a procedure attached to the changeover work-order type. The OEE event stream measures actual changeover time per execution. A monthly review of changeover time against the standard catches drift early.
The method applies, but sanitation changeovers have hard biological constraints that limit the internal-to-external conversion. Expect a more modest gain than the 30-50% typical elsewhere. Still worth doing because sanitation changeovers are usually the longest single events.
Starting with the longest changeover instead of the most-frequent one. The longest changeover has the most theoretical gain but the smallest cumulative impact. The most-frequent has the biggest cumulative payoff and is usually easier to compress.
Login
Blog