Most FMCG production schedules are a due-date list. Each order is placed where it fits on the line that can run it, and the changeovers are whatever that produces. On a line that changes over twelve times a week at a measured median of 38 minutes, the sequence is deciding nearly eight hours of capacity, and nobody is deciding it on purpose.
The changeover article was about making each changeover shorter. This one is about choosing fewer and cheaper ones. The two are different levers with different owners: SMED belongs to the line, sequencing belongs to the planner, and on most lines the planner’s lever is the larger of the two.
A changeover matrix is a table with every format or SKU on both axes and, in each cell, the measured time to change from the row to the column. Twelve SKUs give a 12 × 12 matrix with 132 real transitions. It is the single most useful document a planner can own, and most planners have never seen one, because no system they use produces it.
Three things about the matrix that decide how it is used.
It is asymmetric. Light to dark costs less than dark to light, because the clean is shorter. Small to large format costs less than large to small on many fillers, because of how the change parts stack. Allergen-free to allergen is a format change; allergen to allergen-free is a format change plus a validated clean. A matrix built on the assumption that A→B equals B→A is wrong in roughly half its cells, and the planner sequencing by it will put the expensive direction in the plan as often as the cheap one.
It must be measured, not estimated. The standard says 25 minutes for every pair. The measurement says 20 for some and 45 for others. An estimated matrix is a uniform matrix, and a uniform matrix tells the planner nothing about which sequence to choose. The measurement comes from the same stage-level changeover data the SMED work uses: last good unit to first good unit, per format pair, over enough weeks to see each transition several times.
It changes. As the SMED work lands and the spread between crews collapses, cells shrink. Until it lands, some cells differ by crew, and the matrix should carry that: a transition that takes 26 minutes on nights and 52 on days is a scheduling constraint as much as a training problem. A matrix refreshed monthly from measured data stays true; one built once in a workshop is out of date by the second quarter.
Once the matrix exists, four rules capture most of the saving, and none of them needs software.
Group by format family. Orders that share a format cost no changeover between them. Two orders of the same SKU due Tuesday and Thursday, run back to back, cost one changeover rather than two. This rule alone often removes a quarter of weekly changeovers on a high-mix line.
Within a family, run the cheap direction. Order the formats from the transition the matrix says is cheapest to the one it says is most expensive: light to dark, small to large, plain to allergen. The expensive reverse transition then happens once, at the end of the cycle, rather than several times through the week. This is the idea behind the product wheel, and it is just the matrix read in one direction.
Put the unavoidable expensive changeover where it costs least. Every cycle has one long transition, usually the one that includes a validated clean. Place it at a shift boundary, against a planned maintenance window, or at the point where the crew that handles it best comes on. The changeover takes the same time; the capacity it costs is lower.
Insert exceptions by the matrix, not by the calendar. A promotional order or a rush order has to go somewhere. Placed where its due date lands, it typically costs two expensive changeovers, in and out. Placed next to its own format family, or at the point in the wheel where its transitions are cheapest, it costs one cheap one. The rule is to find the cheapest insertion point that still meets the date, not the earliest.
A planner with the matrix on the desk and these four rules often removes 25 to 40% of weekly changeover time in the first week, before any SMED project starts.
Sequencing by changeover cost means some orders run earlier or later than their due dates would place them. That has a cost, and a planner who pretends otherwise will lose the argument with the supply chain team.
Running early costs inventory: finished goods made Tuesday for a Thursday ship date sit for two days. Running late costs service risk: an order moved to the end of its family group may finish closer to its deadline than anyone is comfortable with. Both are real, and the right answer is to make the trade-off explicit rather than to pick a side.
Put three numbers next to every candidate sequence: changeover hours saved, inventory days added, and orders whose completion moves within one shift of the due date. Then look at the curve. On most lines the first part of the savings, often a third or more of the changeover time, costs almost nothing in service or inventory; it comes from grouping orders that were due on the same day anyway and from choosing the cheap direction. The last part costs a lot; it comes from moving orders days from their due dates to complete a wheel. Most plants should take the first half of the curve and stop, and the planner should be able to show where on the curve the plan sits.
The point is not that changeovers always win. It is that the trade-off is currently being made by default, in favor of the due-date list, by a planner who has never been shown what it costs.
A product wheel is a fixed repeating sequence: the same formats in the same order every cycle, with the cycle length set by demand. Wheels have two advantages that are easy to underrate. Changeovers become predictable, so crews get fast at them and the readiness checklist is the same every time. And the planner’s job shifts from inventing a sequence each week to fitting volumes into a known one.
They have one weakness, and every FMCG planner has met it: the first promotion breaks the wheel. A retailer brings a display order forward, a new variant launches, a competitor’s stockout doubles demand for one SKU, and the fixed sequence cannot absorb it. Plants then either abandon the wheel entirely and go back to the due-date list, or run the promo order wherever it lands and pay two expensive changeovers for it.
The practical answer is a hybrid. A wheel for the base load, which is usually 70 to 80% of volume and is stable enough to repeat. Exceptions sequenced by the matrix, inserted at the cheapest point that meets the date. And a rule for when the exceptions are large enough to re-cut the wheel rather than bolt onto it. The wheel gives the predictability; the matrix handles what the wheel cannot.
The packing line used across this series, one week, twelve orders across six formats: three A, two each of B, C, D and E, and one F. The previous week ended on a D run. The matrix below is illustrative.
| From → To | A | B | C | D | E | F |
|---|---|---|---|---|---|---|
| A | n/a | 20 | 25 | 30 | 35 | 40 |
| B | 35 | n/a | 25 | 30 | 35 | 40 |
| C | 40 | 35 | n/a | 20 | 35 | 35 |
| D | 45 | 40 | 35 | n/a | 25 | 30 |
| E | 95 | 95 | 90 | 85 | n/a | 40 |
| F | 100 | 100 | 95 | 90 | 35 | n/a |
A to D are standard formats; E and F are the allergen-containing formats: 25 to 40 minutes to enter, 85 to 100 to leave because of the validated clean. The matrix is asymmetric throughout: going “down” the standard family (A→B→C→D) costs 20 to 25 minutes a step; coming back up costs 35 to 45.
Due-date sequence. Orders placed where their dates fall: A, B, A on Monday; C, D on Tuesday; B, D, A on Wednesday; E, C on Thursday; F, E on Friday. Twelve changeovers, counting the change from last week’s D, with two entries into the allergen formats and one validated-clean exit in the middle of the week. Total: 7.6 hours.
Matrix sequence. The same orders on the same days, reordered within each day: A, A, B on Monday; C, D on Tuesday; D, A, B on Wednesday; C, E on Thursday; E, F on Friday. The family runs down from A to D twice, and the allergen orders run together at the end of the week, entered once. Nine changeovers. Total: 4.6 hours, and the one allergen exit happens during the planned weekend clean instead of mid-week.
The cost. None the supply chain team will feel. Every order still runs on its due day, so no inventory is added and service is unchanged. The plan sits at the cheap end of the trade-off curve, where the saving comes from grouping orders due on the same day and choosing the cheap direction.
The promo variant. A retailer brings forward an E-format display order that ships on Thursday. Placed by habit a day ahead, on Wednesday between the A and B runs, it costs A→E (35) plus E→B (95), less the A→B it replaces (20): 110 extra minutes. Inserted by the matrix, at the front of the allergen block on Thursday, it costs nothing extra; the block runs one order longer and the order still ships on time. Where the date genuinely cannot wait for the block, the cheapest earlier slot is right after Wednesday’s D run: D→E (25) plus the clean back to A (95), less the D→A it replaces (45), 75 extra minutes.
Three hours a week of changeover time recovered by sequencing alone, with no change on the floor and no order moved to another day. That matches the resequencing step in the changeover article, and it is the planner’s share of the changeover loss.
The matrix is the hard part, and it is the part no planning system produces, because planning systems do not measure changeovers. Fabrico’s manufacturing performance platform (MES, OEE, CMMS & AI) builds the matrix from the floor and puts it in the scheduler.
The matrix is measured and stays current. Every changeover is captured by the OEE module from last good unit to first good unit, by format pair and by crew, with stages separated by computer vision. The matrix is assembled from that data and refreshed as the SMED work lands and the cells shrink. Where a transition still differs by crew, the scheduler sees both values.
The scheduler sequences by it. The production scheduling module sequences orders by measured changeover cost in each direction, groups format families, places the expensive transition against a shift boundary or a planned clean, and inserts exceptions at the cheapest point that meets the date. It holds a base wheel for the stable load and sequences the exceptions around it.
The trade-off is shown, not hidden. For each candidate sequence the scheduler shows changeover hours, inventory days added and orders within a shift of their due date, against the plant’s own inventory and service rules, so the planner can show the supply chain team where on the curve the plan sits.
The financial impact module prices the sequence. With the selling price and margin entered once per SKU, a sequence is also shown as margin: a changeover avoided ahead of a high-margin run is worth more than one avoided ahead of a low-margin one, and the scheduler ranks candidates accordingly.
The floor confirms the matrix. When the changeover runs, the OEE module records what it actually took, and the matrix learns. A cell that keeps coming in above its value is flagged, which is how a SMED project on one format pair gets justified in dollars by the AI actionable insights.
Replanning uses the same matrix. When a line goes down mid-week, the reroute options in the replanning ladder are costed by the same measured transitions, so a rescue does not quietly cost two expensive changeovers.
The planner stops inheriting the changeovers the due-date list produces and starts choosing them.
What is sequence-dependent setup? A changeover whose duration depends on what ran before and what runs next, rather than being a fixed time. Almost all FMCG changeovers are sequence-dependent, which is why the order of production decides the changeover hours and why a changeover matrix is needed to schedule well.
What is a product wheel? A fixed, repeating production sequence in which formats run in the same order every cycle, chosen so that each transition is cheap. It makes changeovers predictable and crews fast, and it needs a rule for inserting exceptions, because promotions and rush orders will not fit the wheel.
How do you reduce changeovers in FMCG? Two levers: make each one shorter (SMED, on the floor) and schedule fewer and cheaper ones (sequencing, by the planner). Sequencing by a measured changeover matrix typically removes 25 to 40% of weekly changeover time before any SMED project starts.
Should you sequence by due date or by changeover cost? By changeover cost within the limits of service and inventory, with the trade-off made explicit. The first portion of changeover savings usually costs almost nothing in service; the last portion costs a lot. Most plants should take the first half and show where the plan sits on the curve.
What is campaign planning? Running orders of the same or similar formats together in a block, or campaign, to minimize changeovers between them, usually in a repeating cycle. It is the production-planning name for the grouping rule above, and it trades changeover time for inventory.
If the planning system has one changeover time for every format pair, the schedule is being built on a matrix that is uniform and wrong, and the sequence is costing hours nobody is counting. Four weeks of measured changeovers will produce the real matrix, and the first week sequenced by it will show what the due-date list has been costing.
The fixed-scope pilot does that on one line: six weeks, every changeover measured by format pair and crew with an industrial sensor and hub installed with your team and computer vision at the changeover stations, the measured matrix in the readout, and one week resequenced by it as a before-and-after, with the service and inventory effect shown alongside the hours saved. 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 matrix the planner sequences by is the one the floor measures.
Request a demo or read how the scheduling module sequences by measured changeover cost.
Login
Blog
