The shift report says the changeover took 25 minutes. It says that because the standard is 25 minutes and 25 is the number printed on the sheet. The line produced nothing for 41.
Every FMCG plant that has run a Single-Minute Exchange of Die (SMED) program, Shigeo Shingo’s method for cutting changeovers toward single-digit minutes, has a changeover standard. Very few have a changeover measurement. The standard was set on a good day, by a good crew, with the stopwatch running from the moment the line stopped to the moment it restarted. The measurement, when a plant finally takes one, runs from the last good unit to the first good unit at rated speed, across every crew, every format pair and every shift. The gap between the two is the loss, and on many packing lines it is the largest controllable loss there is.
This article is about closing that gap with data: measuring changeovers by stage, reading the patterns the measurement reveals, running SMED on the stages that actually carry the loss, and then doing the half of the work most SMED programs skip, which is removing changeovers from the schedule rather than only shortening them.
Changeovers are the one major loss the plan creates on purpose. Breakdowns happen to the plant; changeovers are decided by it, in the planning meeting, usually without anyone pricing them.
The decision has become more expensive. Over the last decade, SKU proliferation, smaller batches and retailer demands for private-label variants and promotional packs have multiplied changeover frequency on many FMCG lines. The 25-minute standard was set when the line changed over four times a week. It now changes over twelve times, and the standard has not been revisited.
The loss is the product of three factors, and most SMED programs address only the second:
A plant that halves changeover duration and leaves the count and the ramp alone has recovered perhaps a third of the loss. The rest sits with the planner and the ramp, and the data has to reach both of them.
Four gaps between what the sheet records and what the line did.
The recorded time is the standard, not the clock. On many lines the operator writes the planned duration because it is the number on the schedule, and the system accepts it. On others the start and end are rounded to the nearest quarter hour. Either way, the sheet records what should have happened.
The end is logged at restart, not at rated speed. The changeover “finishes” when the first unit moves. The line then runs well below speed for most of the next hour while the operator tunes guides and clears the first jams. That hour is logged as production.
Changeover is one block. The log has a single reason code. Nobody knows whether the 41 minutes went into running down the previous SKU, cleaning, swapping format parts, or restarting. A single block cannot be improved; only its stages can.
Changeovers are excluded from OEE as planned. As the OEE audit article covers, the largest controllable loss in FMCG is routinely moved out of the denominator by definition. A loss that is not in the number has no owner, and a loss with no owner does not get fixed.
The effect is a plant that believes its changeovers take 25 minutes, measures nothing that could contradict that, and wonders why the schedule slips every Tuesday.
Before any SMED work, measure every changeover on the line for four weeks. Four weeks covers every format pair and every crew at least a few times. Three numbers per event:
All three are captured by the system. None is written down by a person. The reason is not distrust of operators; it is that a crew doing a changeover has no spare hands for a stopwatch, and a number entered after the fact is the standard, not the measurement.
Record the crew and the format pair against every event. Those two fields are what the next section is built on.
On the packing line used across this series, the standard is 25 minutes. Four weeks of measured changeovers showed a median of 38, a best-crew median of 26 and a worst-crew median of 52. That spread is typical, and it reveals four things in order.
The distribution is wide, not shifted. If every changeover took 38 minutes, the plant would have a capability problem: the standard is wrong and the line needs a different method or different equipment. A spread from 26 to 52 is a variation problem: the 25-minute changeover already exists in the building, roughly once a day, and the work is to make it happen every time. This is the single most useful fact the measurement gives you, and the sheet can never show it.
One or two stages carry the spread. On this line, run-down and clean were nearly identical across all crews. Format change and start-up carried 70% of the difference between the fast crew and the slow one. Everything in the SMED program should be aimed at those two stages; time spent on the others is wasted.
Spread by crew means a method problem. When the same format pair takes 26 minutes on nights and 52 on days, the difference is sequence, preparation and habit. The fast crew stages change parts before the line stops; the slow crew walks to the store. The written standard, if one exists, does not describe what the fast crew does. This is fixable in weeks.
Spread by format pair means a sequencing problem. When format A to format B takes 20 minutes with every crew and B to A takes 35, the loss is in the order the planner chose, not in the crew. The changeover matrix is never symmetric, and the planner has usually never seen it. This is the planner’s fix, and it is covered below.
Most plants find all four patterns at once. The point of separating them is that each one has a different owner.
SMED has not changed since Shingo wrote it down. What changes with measured stage data is that each step is aimed at a known loss rather than at the whole changeover.
Separate internal from external. Internal steps happen with the line stopped; external steps could happen while it runs. The classic method does this by observation and argument. The stage data does it by showing which steps actually occur inside the clock time on the slow changeovers and not on the fast ones. On the series line, the slow crew’s format-change stage included an eight-minute wait for change parts to arrive from the store. That is an external step being done internally, and the data found it without a workshop.
Convert internal to external. Stage the format parts, tooling, labels and film at the line before the run-down starts. Pre-set what can be pre-set. Write the readiness checklist and give it an owner, and make it a task in the maintenance system with a due time before each scheduled changeover, so that “the parts were not there” becomes a missed task with a name on it rather than a shrug.
Streamline what remains. The fastest crew’s sequence becomes the written standard. Not a standard written by an engineer; the sequence the night crew actually follows, captured, documented and put on a tablet at the line with the stage timer visible. The target is the spread collapsing toward the fast crew’s time. Once the spread is gone, the average has moved by itself.
Fix the ramp. This is the step the stopwatch version misses, because the stopwatch stopped at restart. Start-up losses and the post-changeover climb in micro-stops, which the micro-stops article describes, are often half the real cost of a changeover. Centerline the settings for each format so the first good unit is also the first unit at speed, and measure ramp loss as its own number with its own target.
Re-measure after each change. The stage data shows within a week whether the readiness checklist moved the format-change stage, and by how much.
Everything above shortens changeovers. The other half of the loss is how many of them the plan creates, and that belongs to the planner.
Sequence by measured changeover cost. Build the matrix: every format pair, the measured median time in each direction. It is never symmetric; light-to-dark, small-to-large and allergen-free-to-allergen cost less in one direction than the other. A planner sequencing by the matrix rather than by due date alone can often remove a quarter of the weekly changeover time without changing what gets made.
Group formats. Running two orders of the same format back to back costs one changeover, not two. The trade-off is inventory and service, and it is a real trade-off; the point is that it should be made with the cost of a changeover in hours and dollars on the table, not with the assumption that changeovers are free because the standard says 25 minutes.
Place the hard changeovers where they go well. If the data says a particular format pair takes 26 minutes on nights and 52 on days, schedule it on nights until the method is fixed. The schedule can route around the variation the SMED work has not yet removed.
Replan on overrun. When a changeover runs 20 minutes over, the shift does not need to slide. The remaining orders can be re-sequenced, a short order moved to another line, a low-priority run pushed, so the overrun costs one order a delay rather than the whole shift an hour. That requires a schedule that knows the changeover has overrun, which means the plan and the measurement have to be in the same place.
On the series line, resequencing by the measured matrix alone took the weekly changeover count from twelve to nine. No crew did anything differently; the planner simply saw the cost.
The series line: 120 planned hours a week, 58% OEE, changeovers carrying 11 OEE points, about 13 hours a week, of which the hidden capacity method judged 5 recoverable.
| Before | After SMED on two stages | After resequencing | After ramp work | |
|---|---|---|---|---|
| Changeovers per week | 12 | 12 | 9 | 9 |
| Median clock time | 38 min | 30 min | 30 min | 30 min |
| Median ramp loss (minutes at rated speed) | 28 min | 28 min | 28 min | 18 min |
| Clock time per week | 7.6 h | 6.0 h | 4.5 h | 4.5 h |
| Ramp loss per week | 5.6 h | 5.6 h | 4.2 h | 2.7 h |
| Total changeover loss | 13.2 h | 11.6 h | 8.7 h | 7.2 h |
What each column took. The SMED work was aimed only at format change and start-up, the two stages carrying the spread; the readiness checklist and a staged change-part cart removed the walk to the store, and the fast crew’s sequence became the written standard. The resequencing was the planner using the measured matrix, with no change on the floor. The ramp work was centerlining the settings per format and fixing the two micro-stop causes that climbed after every changeover, which the micro-stops article covers.
Total: 6.0 hours a week recovered, 5 OEE points, from a loss the shift report had recorded at 25 minutes a time. The remaining 7.2 hours are the 6 points article one accepted as the cost of running a high-mix line at standard, and they are now a known, priced number the planner can trade against inventory. The weekly totals multiply counts by medians for simplicity. The figures are illustrative; the proportions are what we see when all three levers are used.
A changeover touches every part of the plant at once: the machines, the crew, the plan and the margin. That is why it is hard to fix with a stopwatch and a spreadsheet, and why it is the loss where Fabrico’s manufacturing performance platform (MES, OEE, CMMS & AI) does the most work.
Computer vision separates the stages without a stopwatch. The camera at the station recognizes run-down, clean, format change and start-up as distinct states, so every changeover arrives with its stage times already attached. The spread-by-stage analysis exists on day one, for every crew and every format pair, without anyone timing anything.
OEE puts changeovers in the number. Changeovers sit in the denominator, by format pair and by crew, with the distribution visible rather than the average. The standard and the measurement are on the same screen, which is the moment the gap becomes an owned loss.
The scheduler plans with the measured matrix. The production scheduling module sequences orders by the measured changeover cost in each direction, plans each changeover at its demonstrated time rather than the standard, groups formats against the plant’s inventory and service rules, places hard changeovers on the shifts that handle them, and replans the remaining orders when one overruns. The planner is working from the same numbers the floor is producing.
The financial impact module prices each changeover on what follows it. With the selling price and margin entered once per SKU, a changeover into a high-margin format is worth more to fix than one into a low-margin format, and the AI actionable insights rank the proposals accordingly: “fix the start-up on format D” against the rest of the plant’s backlog, in dollars, with an owner and a measured result.
The readiness checklist lives in the maintenance module. Staging change parts, checking tooling and confirming format-part condition are tasks with a due time and a name, raised automatically ahead of each scheduled changeover. A changeover that ran long because parts were missing traces back to a missed task, not an argument.
The shift report said 25 minutes. With the loss measured by stage, priced by SKU, planned by the matrix and sustained by a task with an owner, the plant gets to a number that is both lower and true.
What is a good changeover time in FMCG? There is no universal figure; it depends on the format pair, the equipment and whether a clean is required. The useful benchmark is internal: the best crew’s measured median for each format pair. If one crew does it in 26 minutes, 26 is the target for everyone, and the spread between crews is the first thing to remove.
What is SMED? Single-Minute Exchange of Die, Shigeo Shingo’s method for cutting changeover time toward single-digit minutes by separating steps that need the line stopped (internal) from steps that can be done while it runs (external), moving as much as possible to external, and streamlining what remains. Measured stage data makes each step more precise by showing which stages actually carry the loss.
Should changeovers be included in OEE? Yes, for any plant that wants to reduce them. A changeover consumes planned production time. Excluding it as “planned” removes the largest controllable loss in FMCG from the number and leaves it with no owner.
Why do changeover times vary so much by crew? Sequence, preparation and habit. One crew stages parts before the run-down; another walks to the store during it. One follows a fixed order; another improvises. The variation is a method problem and is usually fixable within weeks by making the fastest crew’s sequence the written standard.
How do you reduce the number of changeovers without hurting service? Sequence by the measured changeover-time matrix rather than by due date alone, group orders of the same format where inventory rules allow, and make the inventory-versus-changeover trade-off with the real cost of a changeover in hours and dollars in front of the planner. Many plants find a quarter of weekly changeover time can be removed by sequencing alone.
If the changeover time in your report is the same every time, it is the standard, not a measurement. Four weeks of measured changeovers, by stage, crew and format pair, will show a spread the plant has never seen, and the spread is the opportunity.
The fixed-scope pilot does exactly that: one line, six weeks, an industrial sensor and hub installed with your team and computer vision at the changeover stations, every changeover captured from last good unit to rated speed. The readout includes the format-pair matrix, the two stages carrying the spread, and the three highest-value interventions, one of which is usually a sequencing change the planner can make the following week. The fee is fixed and credited in full against a first-year subscription if you roll out. The pilot runs on Fabrico’s manufacturing performance platform (MES, OEE, CMMS & AI), which connects machine data, OEE and loss analysis, production scheduling, SKU-level output value and maintenance in one system, so the changeover the data finds is the changeover the planner schedules.
Request a demo or read how the scheduling module sequences by measured changeover cost.
Login
Blog
