Executive summary: SMED (Single-Minute Exchange of Die) cuts changeover time from hours to minutes by separating work that must happen while the machine is stopped from work that can happen while it runs. On a constraint, every minute of changeover is throughput you never get back, so a changeover cut there converts almost directly into system output. Most manufacturers reach a 50 to 90 percent reduction with existing people and near zero capital. This guide shows the four stages, where to point them, and why the constraint is the only changeover that pays.
- What is SMED and why does it matter?
- What is the difference between internal and external setup?
- What are the four stages of SMED?
- Where should you apply SMED first?
- How do you quantify the payback on a changeover cut?
- What are the most common SMED mistakes?
- SMED: operator FAQ
- About the Stagnation Assassin
What is SMED and why does it matter?
SMED is a Lean method that cuts changeover time by separating setup work that must happen while a machine is stopped from work that can be done while it still runs, then converting as much of the first kind into the second as possible. Applied at a constraint, it turns lost setup time into system throughput without buying equipment.
The method was developed by Shigeo Shingo inside the Toyota Production System, where he drove a press changeover that once took hours down to under ten minutes. The name is literal: single-minute means the changeover fits inside a single digit of minutes, under ten. It became a foundational tool in Lean because it attacks one of the quietest thieves of capacity in any plant, the time a machine spends stopped and producing nothing while someone swaps a die, a fixture, or a tool.
Here is what the textbooks bury. Changeover reduction is not about tidiness or discipline for its own sake. It is about capacity you already own but cannot see, because it is hiding inside stopped time. I have led transformations at Berkshire Hathaway, Illinois Tool Works, and Whirlpool, and in every stalled plant I walked into, there was a machine everyone swore was maxed out that was actually stopped for changeovers a third of the day.
The reason this matters more than most improvement work is placement. A changeover cut on a random machine feels productive and moves nothing. A changeover cut on the process that limits your output converts almost one for one into system throughput. Same tool, same technique, wildly different return, depending entirely on where you point it.
What is the difference between internal and external setup?
Internal setup is work that can only happen while the machine is stopped, such as removing and mounting a die. External setup is work that can happen while the machine is still running, such as staging the next die, gathering tools, and pre-heating. The entire leverage of SMED comes from moving work out of the internal bucket and into the external one.
Let me make this concrete, because the distinction sounds trivial and is anything but. Walk up to any changeover and watch it honestly. You will see an operator stop the machine, then walk to the tool crib to find a wrench, then hunt for the next fixture, then call maintenance, then wait. Every one of those steps is happening on stopped time, and almost none of them needs to. Finding the wrench, staging the fixture, and calling ahead to maintenance are all external work being done internally because nobody ever separated the two.
The first pass through any changeover is pure observation and sorting. You are not improving anything yet. You are just labeling every action as internal or external. In most plants that have never done this, forty to sixty percent of what looks like unavoidable stopped time is external work in disguise. I once timed a changeover the plant called a twenty minute job. Stopwatch said seventy-three minutes once you counted the walking, the searching, and the phone calls. Sorting internal from external on that single line, before we changed one physical thing, took the real number back under thirty.
What are the four stages of SMED?
SMED runs in four stages: observe and separate internal from external setup, convert internal work into external where possible, streamline whatever internal work remains, and standardize the new changeover as documented standard work. Each stage compounds on the last, and most of the early gain comes from the first two before a single dollar of tooling is spent.
Stage 1: separate internal from external
Record the changeover and classify every action as internal or external. This is diagnosis, not improvement. Just sorting the work and doing all external tasks off the stopped clock typically cuts changeover time thirty to fifty percent on its own, because so much of the current stoppage is external work happening at the worst possible time.
Stage 2: convert internal to external
Attack the remaining internal work and ask what would let each task move outside the stoppage. Pre-heating a die, pre-assembling a fixture, or staging tools on a cart converts internal minutes into external ones. This is where the biggest structural wins live, and none of it requires new equipment.
Stage 3: streamline internal setup
Whatever internal work genuinely must stay is now compressed. Quick-release clamps replace bolts, standardized die heights kill shimming, and parallel work by two people replaces one person walking back and forth. These are the classic SMED mechanics, and they belong here, not first. Streamlining before you have separated and converted just makes waste faster.
Stage 4: standardize and document
Lock the new sequence as documented standard work so the gain survives shift change and turnover. An improved changeover that lives only in one operator’s head evaporates the day that operator is out. Standardization is what turns a one-time win into permanent capacity, and it is the stage most teams skip in their rush to celebrate.
Here is the discipline most people get backward. They jump straight to Stage 3, buying quick-release clamps and fancy tooling, because that is the part that feels like real engineering. Then they wonder why the changeover only dropped a little. You cannot streamline your way out of a changeover that is sixty percent external work done at the wrong time. Separate first. Convert second. Only then streamline. Order is not a suggestion here.
A constraint that changes over four times a shift, losing 73 minutes each time, hands back nearly five hours of stopped time a day. Cut that changeover to 11 minutes and you recover more than four hours of pure constraint capacity daily, with no new machine and no added labor, just work moved off the stopped clock.
Where should you apply SMED first?
Apply SMED to the constraint first, and only the constraint, until it is fully exploited. Changeover time on a non-constraint is free, because that process has spare capacity to absorb the stoppage. Changeover time on the constraint is throughput the whole system loses forever. Same reduction, radically different value, so placement decides your return.
This is where I watch good plants waste months of effort. A Lean team runs a SMED event on whatever line the loudest supervisor complains about, cuts a forty-five minute changeover to twelve, celebrates in the monthly review, and system throughput does not move an inch. Why? Because that line was never the constraint. It had spare capacity. Cutting its changeover just gave it more idle time it did not need.
The rule is brutal and simple. Find the process that sets your system output, and point SMED there before you touch anything else. If your constraint changes over frequently, changeover reduction is often the single highest-return move available to you, because it directly buys constraint hours. Pair it with tight loss tracking so you know the changeover was actually the loss worth attacking, and confirm the constraint really is where you think it is before you commit the event. I have seen plants run flawless SMED events on the wrong machine for years.
Once the constraint changeover is genuinely minimized, and only then, does it make sense to push SMED into non-constraints, and even there only where faster changeovers let you run smaller batches that protect the constraint from starving. Everywhere else, changeover reduction is a solution to a problem the system does not have.
How do you quantify the payback on a changeover cut?
You quantify it in constraint hours recovered, then convert those hours to throughput value. Multiply changeovers per day by minutes saved per changeover to get daily minutes recovered, translate that into additional units at the constraint rate, and multiply by throughput value per unit. On a constraint, the annual number is almost always large enough to embarrass the capital request it replaces.
Run the arithmetic on a real case. A packaging constraint changed over three times per shift across two shifts, six changeovers a day, at forty-five minutes each. That is 270 minutes, four and a half hours, of daily stopped time on the exact process that capped the plant. We took the changeover to twelve minutes through separation and conversion, which recovered thirty-three minutes per changeover, or 198 minutes a day, more than three hours of constraint time returned.
At that plant, a constraint hour was worth roughly ten thousand dollars of throughput, a common figure once you count price minus truly variable cost. Three hours a day, across 250 working days, is 750 hours a year at ten thousand dollars, which is 7.5 million dollars of annual throughput recovered. The event cost about ninety thousand dollars in engineering time and a few quick-release fixtures. That is the whole reason SMED belongs near the top of any constraint operator’s list.
Trading 45 minute changeovers for 12 minute changeovers on a constraint running six changeovers a day returns roughly three hours of constraint capacity daily. At about $10,000 of throughput per constraint hour over 250 days, that recovers close to $7.5M a year for under $100,000 of implementation. The capital alternative, a second line, was quoted near $15M.
The comparison that ends the debate every time is cost of delay. Every day that constraint ran with the old changeover burned roughly thirty thousand dollars in lost throughput. Every week executives spent debating the ninety thousand dollar event cost the plant a hundred and fifty thousand in capacity it would never recover. Framed that way, approval stops being a budget question and starts being an emergency.
What are the most common SMED mistakes?
The failures are behavioral, not technical: buying tooling before separating internal from external, running events on non-constraints because they are politically easier, treating SMED as a one-time project instead of standard work, and letting the improved changeover decay after the consultant leaves. None of these is about the method. All of them are about discipline and placement.
Mistake 1: streamlining before separating
Teams buy quick-release clamps and hydraulic tooling before they have sorted internal from external work. They spend capital making the wrong minutes faster while forty percent of the stoppage is still external work done on the stopped clock. Separate and convert first. Those two stages are free and deliver most of the gain. Spend on tooling only for the internal work that genuinely remains.
Mistake 2: pointing it at the wrong machine
The single most expensive SMED mistake is running a flawless event on a non-constraint. You cut a real changeover, you document a real gain, and system throughput does not move because that process had spare capacity. All that effort produced idle time. Confirm the constraint first, apply SMED there, and expand outward only once it is exploited.
Mistake 3: treating it as a project
Teams run a kaizen event, cut the changeover, disband, and move on. Six months later turnover and drift have crept the changeover back up, and nobody noticed because nobody was measuring it. SMED gains survive only as documented standard work with an owner and a tracked metric. Without Stage 4, you are renting the improvement, not owning it.
Mistake 4: ignoring the human resistance
My own worst SMED mistake was assuming that showing people the data would change behavior. Operators had run a changeover the same way for fifteen years. Handing them a faster sequence on paper did nothing until we rebuilt it with them on the floor, timed it together, and let them own the new standard. Spend more time on the change management than on the stopwatch. The technique is the easy part.
SMED: operator FAQ
What does SMED stand for?
Single-Minute Exchange of Die. It means a changeover fast enough to complete in single-digit minutes, under ten. The method separates setup work that must happen while the machine is stopped from work that can happen while it runs, then converts as much stopped work as possible into running work, cutting changeover time by 50 to 90 percent with existing resources.
How much can SMED reduce changeover time?
Typically 50 to 90 percent, and most of it comes free. Just separating internal from external work and doing all external tasks off the stopped clock cuts changeover 30 to 50 percent before any tooling. Converting internal work to external adds more. Streamlining and standardization capture the rest and make the gain permanent.
Where should SMED be applied first?
The constraint, and only the constraint, until it is fully exploited. Changeover time on a non-constraint is free because that process has spare capacity to absorb the stoppage. Changeover time on the constraint is throughput the whole system loses forever, so a changeover cut there converts almost directly into system output.
Is SMED the same as Lean?
No. SMED is one tool within Lean, focused specifically on changeover reduction. It works best when constraint analysis tells you which changeover actually limits system output, then SMED provides the method to attack it. Applied without that targeting, SMED often produces faster changeovers on processes that were never the bottleneck.
About the Stagnation Assassin
Todd Hagopian is a Fortune 500 transformation executive who has generated $3B+ in shareholder value across Berkshire Hathaway, Illinois Tool Works, Whirlpool, and JBT Marel, where he serves as VP of Global Product Strategy. Known as The Stagnation Assassin, he is the author of two published books: The Unfair Advantage: Weaponizing the Hypomanic Toolbox and Stagnation Assassin: The Anti-Consultant Manifesto. His blog is published in 15+ languages and read by operators worldwide. Bring him to your stage via the speaking page or connect with him on LinkedIn.
Next step: a constraint changeover audit
Your constraint is probably stopped for changeovers a third of the day, and you are calling it maxed out. Book a 20 minute changeover audit and I will help you find how many constraint hours are hiding inside your stopped time, then show you how to recover them without buying a single machine. Start the audit here.