Setup Reduction / SMED: Size the Changeover Gain Before You Commit to the Pilot

Setup Reduction / SMED applies Single-Minute Exchange of Die to reduce setup and changeover time. It does this by separating internal tasks from external tasks. Internal tasks can only happen while the process is stopped. External tasks can happen before or after, while the process is still running or already running again.

The tool sits in the Improve phase of RDMAICS. That placement matters. By the time you open it, you should already have a cause. Measure told you where the time goes. Analyse told you why. Improve is where you turn that finding into a specific change with a number attached to it.

Reach for this tool when your Analyse work points at the gap between runs rather than the run itself. That is the signature of a setup problem. The work is fine when it is moving. The loss sits in the stop. If your diagnosis instead points at defects inside the run, or at demand variation, a different Improve tool will serve you better.

The output is not a plan. The output is a sized opportunity. You use it to decide whether a pilot is worth running at all.

The tool opens with sample data already loaded. The following case is fictional. It exists so you can see the mechanics before you type anything of your own.

The case is a document processing changeover. Five tasks make up the setup. Configure OCR template takes 15 minutes. Load vendor rules takes 10 minutes. Approver assignment takes 8 minutes. Test match rules takes 12 minutes. System warm-up takes 5 minutes.

Four of those five tasks are marked Internal. Only System warm-up is marked External. That single classification column carries most of the analytical weight. It says that almost the whole setup currently happens with the process stopped, which is exactly the condition SMED exists to attack.

Each task also carries an improvement idea and a future time. Configure OCR template moves to 3 minutes by pre-configuring templates externally. Load vendor rules moves to 2 minutes by pre-staging rule sets. Approver assignment moves to 1 minute by auto-assigning by threshold. Test match rules moves to 4 minutes through parallel testing. System warm-up moves to 0 minutes by scheduling it before the shift.

The worksheet is editable. You enter one row per setup task, in the order the tasks actually happen. Order matters more than you expect, because it exposes which tasks are waiting on which.

For each row you record the current time in minutes, the type, the improvement idea and the future time in minutes. The sample data shows all four columns filled for all five tasks. Blanks are excluded from the calculation, so a partially filled row will quietly drop out of the chart.

Timing is the part people rush. Do not estimate current time from memory. Stand at the process and time the changeover. If the sample case had guessed 10 minutes for Configure OCR template rather than observing 15, the whole picture would shift.

Classification is the second judgement call. Ask one question of each task. Could this be done while the process is still running? System warm-up in the sample is already marked External, which means it is already being done outside the stop, or could be. The other four are marked Internal, which means they currently require the stop.

The result is a chart that plots Current Time in minutes against Future Time in minutes, task by task. It is built directly from the worksheet, and the tool states that values are parsed from the editable worksheet with blanks excluded.

Read it left to right by task name. Configure OCR template, Load vendor rules, Approver assignment, Test match rules and System warm-up each appear with two bars. The axis runs from 0 to 16, which comfortably contains the largest current value of 15 minutes.

The shape of the chart is the finding. Configure OCR template has the tallest current bar at 15 minutes and drops to 3 minutes. Test match rules sits at 12 minutes and drops to 4 minutes. Those two tasks carry the biggest visible movement, so they are where a pilot should start.

System warm-up is the quiet one. It is already External and its future time is 0 minutes, because the improvement is simply to schedule it before the shift. That is the cheapest change on the sheet and it requires no new tooling at all.

The tool also offers an AI review that explains the result. Treat it as a second reading rather than a verdict.

The first decision is whether to pilot at all. You compare the effort of the improvement ideas against the movement in the chart. If the bars barely move, the setup is not your constraint and you should go back to Analyse.

The second decision is sequencing. You do not pilot five changes at once. The chart tells you which task to take first. In the sample case, Configure OCR template offers the move from 15 minutes to 3 minutes, so it leads. Approver assignment, moving from 8 minutes to 1 minute through auto-assignment by threshold, is a different kind of candidate because the change is a rule rather than a physical preparation.

The third decision is who owns each change. Pre-staging rule sets is a preparation task that someone has to actually do, before the stop, every time. Scheduling system warm-up before the shift needs a roster change. Parallel testing for Test match rules needs a second environment or a second pair of hands. Each future time in the sheet is a promise that someone has to keep.

The fourth decision belongs to Control and Sustain. The future times become your targets. If the pilot holds, those numbers become the standard you monitor.

The most common mistake is classifying everything as Internal. People assume the stop is required because the stop is habitual. In the sample case only one of five tasks is External, and that proportion is typical of an unexamined setup. Challenge each row.

The second mistake is writing improvement ideas that are really wishes. Look at how the sample case phrases them. Pre-configure templates externally. Pre-stage rule sets. Auto-assign by threshold. Parallel testing. Schedule before shift. Each one names a mechanism, not an aspiration.

The third mistake is setting future times by subtracting a round percentage from current times. Future time should come from reasoning about what is left after the change. Approver assignment drops to 1 minute because automatic assignment leaves only a check. System warm-up drops to 0 minutes because it no longer happens inside the stop at all.

The fourth mistake is leaving cells blank and then trusting the chart. Blanks are excluded. A task with no future time simply will not show the comparison you were hoping to read.

This is a sizing tool, not a verification tool. Every future time in the sheet is a forecast you wrote yourself. The chart renders your forecast faithfully. It does not test it. Nothing in the result confirms that Configure OCR template will actually land at 3 minutes.

It will not tell you the cost of the change. Parallel testing may need a second environment. Pre-staging rule sets may need someone rostered earlier. None of that appears in the worksheet, so you have to carry the cost question separately.

It will not tell you how often the changeover happens. A setup is only worth attacking in proportion to its frequency. The sample data shows times per changeover and nothing about volume. You supply that context from your Measure work.

It will not tell you whether the saved time becomes useful capacity. Time released at a step that is not the constraint changes very little. It will also not account for variation, since each task carries a single current time rather than a range.

Finally, it will not run the pilot. The chart gives you a case worth testing. The test is still yours to design, run and hold.