SMED vs Sequencing: Why Reducing Setup Time and Optimizing Setup Order Are Two Different Projects
A plant runs SMED workshops for three years. Average changeover time drops by nearly half. Everyone can point to the improvement on the board in the corridor. Then someone pulls the total number of hours the plant spent in changeover last quarter, compares it to the same quarter two years earlier, and the curve has barely moved.
This is one of the most common plateaus in operational excellence, and it is not a failure of the SMED work. It is the result of two distinct problems being treated as one.
SMED reduces the cost of an individual changeover. Sequence optimization reduces how many expensive changeovers a schedule contains in the first place. They act on the same production reality, they are owned by different functions, they require different data, and neither one substitutes for the other.
What Is SMED and What Does It Actually Change?
SMED, or Single-Minute Exchange of Die, is a lean method formalized by Shigeo Shingo at Toyota for reducing the time required to switch a resource from one product to the next. The name is often misread. It does not mean a changeover of one minute. It means a single-digit number of minutes, so under ten.
The mechanism at the core of the method is the separation between two categories of work. Internal setup can only be performed while the machine is stopped, such as removing tooling or purging a line. External setup can be performed while the machine is still running, such as staging the next tool, pre-heating, or bringing components to the line.
Most of the gain comes from converting internal work into external work, then streamlining what genuinely has to remain internal. The operators do not work faster. Work is moved out of the stopped window.
The important point for scheduling is what SMED leaves untouched. It shortens each transition taken on its own. It says nothing about which transition happens next.
What Is Sequence Optimization and Why Is It a Different Problem?
Sequence optimization starts from a different observation: the cost of a changeover is not a property of the job you are starting. It is a property of the pair. Switching from product A to product B and switching from product A to product C are two different events with two different costs.
This is what a changeover matrix captures, and it is worth reading how a changeover matrix is built and why most ERP systems never give you one before going further, because the rest of this article uses that structure.
The consequence is that two schedules containing exactly the same set of jobs, on the same resource, in the same week, can consume very different amounts of non-productive time. The difference comes only from the order.
Note that this is a combinatorial problem, not an arithmetic one. It cannot be solved by applying a flat average setup time, which is precisely what a spreadsheet-based schedule is forced to do.
Why Do Both Levers Act on the Same Matrix?
Here is the clearest way to see the relationship. Take the three-product changeover matrix used in our matrix article, with its asymmetric transitions, and run the same four jobs through it: A, B, C, and a second batch of A.
The worst ordering, A to C to B to A, consumes 160 minutes of changeover. The best ordering, A to A to B to C, consumes 60 minutes. That is a spread of 100 minutes on four jobs, purely from sequencing.
Now assume a successful SMED programme that cuts every transition in the matrix by 45 percent. This is an arithmetic hypothesis chosen to make the mechanism visible, not a measured MangoGem result.
The absolute spread is halved, which is real value. But the ratio is identical. If every cell in the matrix falls by the same proportion, the relative advantage of good sequencing is mathematically unchanged. SMED reduced the magnitude of sequence dependency. It did not reduce the dependency itself.
In practice, of course, SMED does not compress uniformly. It bites harder on long, complex, multi-step changeovers than on short ones, so the dispersion across the matrix does narrow, and the return on sequencing does erode somewhat. What it will not do is remove structural asymmetry. A transition from a dark pigment to a light one will still require a deeper clean than the reverse, as chemical and process plants know from CIP sequencing, no matter how efficient the cleaning procedure becomes.
Which Lever Should You Invest in First?
Both eventually. The order depends on what your matrix looks like today, and five checks are enough to tell you.
- Measure total changeover hours, not average changeover duration. The average is the SMED metric. The total is the capacity metric, and it is the one that moves your throughput.
- Separate frequency from unit cost. Multiply the number of transitions per period by their average cost. If the count is high and the unit cost is moderate, your problem is sequencing.
- Compare your most expensive transition to your cheapest. This dispersion is the single best predictor of how much sequencing has to give.
- Check for asymmetry. Confirm whether A to B genuinely differs from B to A across your main product families. Asymmetry is sequencing value that SMED cannot reach.
- Check how constrained your order is. If due dates and material availability leave almost no freedom in sequencing, setup reduction is the only lever actually available to you.
The reading rule is simple. Wide dispersion in transition costs means sequencing pays first. Uniformly high transition costs mean SMED pays first. Most plants that have already run SMED programmes for several years fall into the first category without knowing it, because the work of the last three years compressed the values and never touched the path.
How Do SMED and Sequencing Compound?
The two levers multiply rather than add, but only if the scheduling model can represent what SMED has produced on the floor.
This is a concrete technical requirement. Once a changeover has been split into internal and external work, part of the preparation happens while the previous operation is still running. A scheduling engine that treats every setup as a monolithic block of stopped time will simply not see that gain. It will keep planning against the old, pre-SMED reality and quietly give back what the improvement teams earned.
MangoGem APS Optimizer models setups as sequence-dependent transitions and supports preparation work that starts before the preceding operation ends on the resource. It reads the full pairwise matrix rather than a flat average, groups changeover families where lateness constraints allow it, and weighs setup savings against tardiness risk across the whole horizon instead of the next job alone.
In high-variety environments, plants applying this approach typically target a 20 to 30 percent reduction in changeover time.
The compounding effect is straightforward. SMED lowers the floor of what each transition costs. Sequencing ensures you take the cheap path through the matrix as often as your due dates permit. A plant that has done one and not the other is running at roughly half of the available benefit.
Conclusion
Reducing setup time and optimizing setup order are complementary, not competing. One works on the shop floor and changes what a transition costs. The other works in the schedule and changes which transitions you pay for.
Plants that treat them as a single initiative usually end up doing only the first, because it is visible, physical, and owned by a team that already exists. The second is invisible until someone builds the matrix and looks at the spread.
If you want to know which of the two your plant should fund next quarter, start with the five checks above using your own changeover data. The answer is usually clear within a day.
Frequently Asked Questions
1. Does a successful SMED programme make sequence optimization unnecessary?
No. If setup reduction is applied evenly across transitions, the relative gap between a good and a bad sequence stays the same. The absolute stakes shrink, but the decision remains just as valuable per changeover. Asymmetric transitions in particular survive SMED almost intact.
2. Should we build a changeover matrix before or after a SMED project?
Before, if possible. The matrix tells you which transitions are worth a SMED workshop, so it directs the improvement effort instead of spreading it evenly. It also gives you a measurable baseline against which the SMED gain can be verified pair by pair.
3. What if all our setups are already short and similar in duration?
Then sequencing has little to give on that resource, and you should look elsewhere for capacity. Low dispersion across the matrix is a legitimate reason not to invest in sequence optimization on that line, and it is exactly what step 3 of the diagnostic is designed to reveal.
4. Can optimizing for setup time damage on-time delivery?
It can, if setup is optimized in isolation. Grouping every similar product together minimizes changeover time and will push some orders late. This is why setup reduction has to be one objective among several, weighed against tardiness, rather than a standalone rule.
To know more about reducing setup time: www.mangogem.com.