Changeover and Setup Time Reduction 2026 — Why the Gains Reverse in Six Months

Most plants that run a changeover reduction program are back at their original numbers six months later. The method is not what is wrong. Because changeover is only ever measured as a single lump, nobody can see which part crept back. This article splits changeover into four segments and models, for a Japanese-owned plant in Thailand, which segment actually pays when you invest in it.

Changeover Reduction Reverses in Six Months Because of the Measurement Unit, Not the Method

The scene on the shop floor is almost always the same. As a kaizen activity, the team films a changeover, separates internal setup from external setup, and converts fasteners to one-touch clamps. A setup that took 48 minutes now takes 42. That is a 12.5% reduction, and the review meeting applauds. Six months later the same machine measures 43.5 minutes.

The conclusion most plants reach at that point is that the standard is not being followed, or that the operators have turned over, and retraining gets scheduled. Retraining is not wasted effort, but it addresses a symptom rather than the cause. The real problem is that the 48 minutes was only ever recorded as a single lump. As long as you measure the lump, a result of 43.5 minutes tells nobody which task got longer. What you cannot see, you cannot fix.

Measuring in lumps has a second side effect. Improvement targets skew toward the visible work. Film a changeover and the sequence where the old die comes off and the new one is clamped down has the most motion and looks like it holds the most room for improvement. So one-touch tooling, intermediate stoppers, and other clamping-side countermeasures are the first things proposed.

What actually eats the setup time, though, is the segment with little motion and no visual drama. After clamping is finished, the machine has to be dialed in to condition, a first article has to be run and measured, and then the line waits on the inspector and the approver. Because the operator’s hands are not moving here, it is rarely recognized as an improvement target. What the video shows is a person waiting, and that does not make it onto the kaizen agenda.

Turning changeover reduction from a one-off event into a state that holds requires measuring at a granularity that matches the structure of the work. Below, changeover is split into four segments, and for each one we set out what is happening and how much comes back for how much invested.

Note that every machine count, changeover frequency, time split, and amount used in this article is a model assumption made by this article, not measured data from a specific plant. When applying it to your own plant, treat measuring your own segments first as the prerequisite. Sourced statistics and model estimates are clearly distinguished throughout the text.

Splitting Changeover Into Four Segments — Where the Time Disappears

Treat changeover as one continuous stretch of time from the last good part out to the return to steady-state speed, and divide it into four segments. There are many defensible ways to cut it. This article draws the boundaries on what has been completed rather than on who is doing what, because cutting on operator activity makes the boundaries ambiguous the moment several people work in parallel.

The total is 48 minutes per changeover, and the ratios round to 8 / 27 / 42 / 23. This split is a model assumption in this article and will naturally differ by industry and process. That said, in processes that require dimensional assurance on the first article, such as metal machining or plastic molding, the structure that makes R3 the largest segment is broadly common.

Two things in this table deserve attention. First, R3 accounts for 42% of the whole. Second, R2, where most shop-floor improvement is concentrated, is only 27%. Even if you halved R2, the total would shrink by just 13.5%. Take R2 to zero and 35 minutes still remain.

R4 is more troublesome still. R4 covers the segment where the first article has passed but the line is not yet at steady-state speed. The machine is running here, so at most plants it falls outside the changeover ledger entirely and gets classified as performance loss, that is, speed loss. It is a segment that structurally disappears from changeover effectiveness measurement. Improve it and nobody credits you; let it degrade and nobody detects it.

Measuring by segment means replacing one number, 48 minutes, with four numbers. When the figure creeps back to 43.5 minutes after an improvement, you know immediately whether R2 stretched to 8.5 minutes or R3 swelled with approval waiting. Once the cause is identified, the remedy has a target. Put the other way around, an improvement that is not measured by segment leaves you with no move when it reverses.

Three Reasons Changeover Pays Off in a Thai Plant in 2026

Changeover reduction is correct in any era, but it is not always the top priority. If equipment runs flat out and orders overflow, the conversation is about capacity investment long before it is about changeover. When utilization is falling, changeover climbs the return-on-investment ranking. Thailand in 2026 is squarely in that situation.

First, falling utilization. Statistics published by Thailand’s Office of Industrial Economics (OIE) and reported on 27 July 2026 put average manufacturing capacity utilization for June 2026 at 57.47%, with the Manufacturing Production Index (MPI) down -3.1% year on year, the steepest decline since last November. A separate series from the Bank of Thailand (BOT) puts utilization at 57.61% for June 2026, well below the long-run average of 67.85%. OIE and BOT compute different series, so the figures must not be mixed even though both are called capacity utilization. What both series share is the fact that equipment is idle.

When equipment sits idle, sales goes after smaller orders. Quantities per lot fall, and the number of changeovers required for the same production volume rises. Even if the time per changeover is unchanged, more changeovers means more total time consumed by setup. Falling utilization automatically raises the changeover burden.

Second, labor cost. According to JETRO, the minimum wage in Bangkok rose to 400 baht per day effective 1 July 2025. Changeover is normally work that several people attach to at once, and operators capable of running a changeover are relatively skilled and relatively well paid. The labor cost poured into each changeover has risen in step with the wage level.

Third, the escape route through inventory is harder to use. The easiest way to cut changeover frequency is to consolidate lots, but that inflates work-in-process and finished-goods inventory. Building inventory when the order book is unpredictable converts directly into the risk of stagnant stock and write-downs. The less stable demand is, the more the only option is to shorten changeover itself rather than absorb it with inventory.

Because all three arrive at once, changeover reduction in a Thai plant in 2026 has the character of a profit defense rather than a routine kaizen activity. As a way to add usable hours without adding machines, it has moved up the return-on-investment ranking.

Separating Internal and External Setup Only Holds Once It Is Measured

SMED (Single Minute Exchange of Die), the classic methodology for changeover improvement, is usually explained in four steps. First, observe and record the current changeover work. Second, classify the work into internal setup, which requires the machine to be stopped, and external setup, which can be done while the machine runs. Third, move whatever can be moved from internal to external. Fourth, shorten the internal setup that remains.

Those four steps are correct and still effective. They have one decisive property, however. All four are designed as work you do once, and none of them contains a mechanism for sustaining the result. Converting to one-touch tooling or putting the next die on a cart beside the machine pays off the day it is done. But nothing on the methodology side guarantees that the state survives into the following month.

The paths by which externalization collapses on the floor are predictable. The forklift that was supposed to bring the next die gets pulled onto another job, so the machine is stopped and only then does somebody go to fetch the die. The operator dedicated to changeover is reassigned to cover an absence on the line, and that day one operator does the changeover alone. Material for the next part number does not arrive because the upstream process is late, so the machine is stopped and then waits. In none of these cases did someone fail to follow the standard. Circumstances arose that made the standard impossible to follow.

In Thai plants this collapse is even more likely. Operator rotation is frequent, so the set of people who can run a changeover does not stay fixed. Turnover removes experienced staff, and what was taught does not stay in the organization. Work instructions exist only in Japanese and English, so the details do not reach Thai-speaking or Myanmar-speaking operators. These are structural conditions that exhortation does not solve.

That is precisely why externalization has to be viewed as how it went today, not as whether it was done at some point. Make it possible to distinguish a day when R1 finished in 2 minutes from a day when it took 6. On the day R1 stretched, you can ask that same day whether the next die was not at the machine or whether people were short. Stare at an average at next month’s QC meeting and nobody remembers the reason any more.

A mechanism for sustaining is, in short, measurement plus daily feedback. Behind the four steps of SMED, add a fifth step of looking at segment-level results every day. That is the core of this article’s argument and the reason segment splitting is necessary.

How to Capture Segment Boundaries — Equipment Signals Alone Cannot Tell You When Changeover Starts and Ends

Once you decide to measure by segment, the first obstacle is acquiring the boundaries. Who decides the boundaries between R1 and R4, and from what data? Install hardware without designing this and you end up collecting data that cannot be cut into segments. The common failure in practice is taking only the stack light signal, which gives a clean running-or-stopped binary but says nothing about what the stop contains.

There are broadly three ways to capture boundaries, and each yields something different.

Trying to close the loop with any one of the three always breaks down. Equipment signals alone do not explain the reason for a stop, operator input alone loses a whole day of data when the presses are missed, and business system events alone do not tell you how long the machine was actually down. The practical answer is a three-layer arrangement. Take the skeleton from equipment signals, add meaning with operator input, and corroborate with business system timestamps.

Missed button presses are designed out, not exhorted away. First, reduce the number of buttons. Separating R1 through R4 requires catching four transitions, but the end of R1, meaning the machine has fully stopped, comes from the equipment stop signal, and the end of R4, meaning steady-state speed has been reached, comes from automatic cycle-time evaluation. In practice the operator only needs to press two or three. Second, build a path for correcting a missed press after the fact. Display the previous day’s gaps at the morning meeting and assign the segments on the spot, and no data is lost.

Equally important is putting the motive to press on the operator’s side. If the result of pressing a segment button only ever surfaces in a manager’s report, the only reason to press is that somebody said so. Show the day’s segment results for that machine and the trend over the last 5 changeovers on the shop-floor terminal, so the operator can see on the spot whether this changeover beat the last one. Once the information pressed comes back to the person pressing, missed presses drop visibly.

The boundary most often overlooked in design is the end of R4, the judgment that steady-state speed has been reached. Leaving this to the operator produces wide variation. In practice it is more stable to evaluate it automatically, at the point where a moving average of measured cycle time stays within a set band of standard cycle time for a set number of consecutive cycles. Fix the logic once and segments stay comparable.

Once segment data is available, the granularity of downtime cause analysis rises automatically. Changeover-related stops separate from minor stoppages, and improvement priority by machine lines up in numbers. That ground is covered in our article on minor stoppages and how to think about OEE, so start there if you want to begin from downtime classification.

Cutting R3 (Adjustment, First-Article Inspection, Approval Wait) — Least Investment on the Largest Segment

R3, the largest of the four segments, is not actually a single activity. Four kinds of time with different characters are mixed into that 20 minutes. Telling people to speed up first-article inspection without breaking this apart moves nothing.

When plants actually measure, it is not unusual for approval wait to be the largest of the four. Trial shots get onto the improvement agenda because the operator’s hands are moving, but the 7 minutes spent waiting for the approver to come back are not recorded as anyone’s work time. As long as changeover is measured as a lump, those 7 minutes dissolve invisibly into the 48.

Measures that cut approval wait need almost no capital equipment. Move first-article inspection records from paper to terminal entry and let approval happen in the system. Decide in advance, by part-number criticality, who can approve in the approver’s absence. Push a notification to the approver the moment an approval request is raised. Those three make most of the approval wait disappear.

Shortening the dialing-in hinges on recipe management for process conditions. Store the temperature, pressure, speed, and time used the last time that part number ran, keyed on the part-number-and-die combination, and recall them as initial values next time. What matters here is storing the conditions under which good parts came out, not building a standard conditions table. Most plants already have a standard conditions table, and it usually diverges from the conditions that actually produced good parts, which is exactly why the dialing-in is redone every time.

This article’s model assumes R3 falls from 20 minutes to 11, a reduction of 9 minutes. The breakdown assumes roughly 5 minutes from electronic approval routing, 3 minutes from recipe recall, and 1 minute from moving measurement to the machine. This is an assumption made by this article, not measured data. What matters for the investment decision is that most of the 9 minutes comes from changing a business process. What R3 reduction requires is not expensive equipment but changes to approval rules and to how data is held.

R4 (Ramp-Up and Condition Stabilization) Disappears Inside Speed Loss

R4 runs from the moment the first article passes to the moment steady-state speed is reached. This article’s model puts it at 11 minutes, 23% of the whole. It is shorter than R2’s 13 minutes and a little over half of R3’s 20, but it carries a difficulty of its own. In measurement terms it easily falls off the changeover ledger.

During R4 the machine is running. Parts are counted. So a collection system that watches only equipment signals classifies the segment as running. The actual speed, though, is below steady state, and defects are more likely. The result is that this time stays inside operating time, reappears as performance loss or quality loss, and drops out of changeover effectiveness measurement entirely.

Consider what that does in practice. Cut R1 through R3 hard and, if R4 has stretched, total output does not rise as much as expected, yet the changeover time record shows improvement. A report goes up saying changeover got shorter but output did not rise, and confidence in the improvement activity itself is lost. The cause is not that the improvement failed, but that the scope of effectiveness measurement does not match reality.

Even within the ISO 22400 framework, the treatment of ramp-up time remains an open point. Whether it is handled as equipment downtime or as reduced performance changes the composition of the OEE that comes out. The practical recommendation is to decide the treatment once internally, fix it, and keep the series comparable over time. Change the definition every fiscal year and you can no longer tell improvement from redefinition.

Measures that shorten R4 are continuous with R3’s recipe management. Store and recall the conditions that produced good parts, and the time to condition stability shortens by itself. In addition, make the handling rules for ramp-up parts explicit. From which piece does the count of legitimate good parts begin, and is product made during ramp-up shippable or not? Where this is vague, someone has to be called for every judgment, and that time lands straight in R4.

In plants where the method of counting production is itself vague, the reality of R4 is harder still to see. Automating the count and measuring changeover are often cheaper when designed as a set, and that ground is covered in our article on automating production counts.

Where Changeover Sits in OEE — ISO 22400 and the Six Big Losses

Once you decide to measure changeover, you have to decide where it lands in the KPIs you already run. Most plants already operate OEE (Overall Equipment Effectiveness), so this becomes a design question about where changeover is reflected in OEE.

OEE is defined in ISO 22400-2:2014 as the product of three factors, Availability, Performance, and Quality. As a formula, OEE = Availability × Performance × Quality. Changeover time lands on availability. That is, time spent on changeover is treated as time the equipment could not run, and it lowers availability.

A practical fork appears here. A non-trivial number of plants exclude changeover time from the denominator as planned downtime. Treating it as planned downtime raises the OEE number, but it also means removing changeover from the scope of improvement. Changeover is a genuine availability loss and must not be subtracted from the denominator as planned time, a point that general guidance on running OEE makes repeatedly.

Where changeover sits within a loss taxonomy has a classic answer in the TPM literature. In the six big losses defined by Seiichi Nakajima in 1988, changeover sits in the second category, Setup and adjustments. What is worth noting is that this classification bundles setup and adjustment together from the outset. It corresponds to this article’s R2 and R3 combined, meaning adjustment time was recognized as part of changeover nearly 40 years ago. Shop-floor measurement has nonetheless gone on skewing toward the clamping work.

As a reference point for judging your own level, there are industry OEE benchmarks. Benchmarks published in 2026 put discrete manufacturing OEE at a median of 65 to 75% and an upper quartile of 78 to 85%. Plants with high-mix low-volume production and frequent changeovers usually sit at the bottom of that median range. Turn that around and cutting changeover works by pushing OEE up as a whole through availability.

How to present segment-level changeover results alongside OEE becomes a KPI design question in its own right. Next to OEE by machine, put the segment averages per changeover, the recent trend, and the gap to target on the same screen. KPI structure and how to present it on the floor are organized in our article on factory KPI management, so start there if you are working from overall indicator design.

Options for Cutting Changeover Frequency, and What Grows Behind Them

Total time consumed by changeover is the product of time per changeover and frequency. So there are two directions for reduction. Shorten each one, or do fewer. This article has dealt with the former up to here, and it needs to state explicitly why it does not pursue the latter.

There are three main ways to cut frequency. Consolidate production of the same part number into larger lots. Sequence part numbers that run on the same die back to back to cut the number of die changes. Concentrate part numbers on the machines that need changeover and dedicate the others.

What to read in this table is that every entry in the right-hand column is a cost that appears somewhere else. Lot consolidation merely converts changeover time into inventory. Halve the frequency and half of the 4,800 hours a year lost to changeover is indeed freed, but cash is tied up in stock instead. As noted above, in Thailand in 2026 demand is hard to forecast and there is little room to take risk in inventory.

Part-number sequencing is superior in that it cuts frequency without adding inventory. Making it work, though, presupposes a structure that can rebuild the production plan daily and organized information on dies and conditions for each part number. In practice, the data groundwork needed to satisfy that prerequisite overlaps almost exactly with the data groundwork needed for segment measurement. In other words, the sequence of measuring first also pays off in the direction of cutting frequency.

Machine dedication requires capital and goes obsolete when the part-number mix shifts. With utilization in the 57% range in Thailand today, it is hard to justify dedicating machines and pushing utilization lower still.

For these reasons this article takes the position of concentrating on shortening the time per changeover. Cutting frequency trades off against other KPIs, namely inventory and delivery, and belongs in a separate discussion about production management design as a whole. The two are not mutually exclusive, however. Getting one changeover down to 28 minutes and then applying sequencing only to part numbers with stable demand is a perfectly realistic combination.

Five-Layer Cost Breakdown and Payback Estimate

From here the discussion turns to money. The model plant assumed is as follows. These are assumptions for an estimate, not figures from a real plant.

• A Japanese-owned metal machining and plastic molding plant in Ayutthaya, Thailand
• 12 machines that require changeover, 2 changeovers per machine per day, 250 operating days a year
• Annual changeovers = 12 × 2 × 250 = 6,000
• At 48 minutes each today, annual changeover time is 6,000 × 48 = 288,000 minutes = 4,800 hours
• Lost contribution margin per machine-hour is set at 1,200 THB per hour
• Annual loss attributable to changeover = 4,800 × 1,200 = 5,760,000 THB

That 1,200 THB per hour unit rate assumes an order book that is full, so that any hour freed on a machine can be filled with additional orders. In a plant whose order book is not full, the benefit of changeover reduction appears only on the labor cost side, not in contribution margin. This should be confirmed first, as a precondition of the investment decision. When utilization is falling in particular, you need a cool look at whether the time you cut actually converts into revenue.

Two improvement scenarios follow. Case A is the approach most plants actually take, which is to attack R2 only. R2 goes from 13 minutes to 7 and the 48 becomes 42. A 12.5% reduction. So far it looks like a success. But because segment results are not measured, nobody notices when R2 creeps back to 8.5 minutes six months later. Total time becomes 43.5 minutes and most of the gain is gone before the year is out. This is what the reversal within six months described at the top actually consists of.

Case B measures across four segments first and then attacks segment by segment.

Seen annually, Case B delivers 6,000 × 20 minutes = 120,000 minutes = 2,000 hours saved. Converted to money that is 2,000 × 1,200 = 2,400,000 THB per year. Set against Case A, where attacking R2 alone yields 6 min × 6,000 = 36,000 minutes and then reverses within six months, the difference is obvious.

Next, build up the cost side in five layers. Layering it is what makes it possible to decide to stop partway.

Amounts are in THB. Layers 1 to 3 are the measurement platform, layer 4 is the business process change, and layer 5 is embedding. Note in particular that layer 5 is shown explicitly as a cost. A rollout that does not budget for training, standard revision, and hands-on support tends to land in a state where the hardware runs but nobody looks at it.

The payback calculation is set up as follows. Because benefits only materialize in stages in year one, the annual benefit is taken at 60%. 2,400,000 × 0.6 = 1,440,000 THB. Subtracting the 384,000 annual operating cost gives a year-one cash flow of 1,056,000 THB. Deducting the 1,934,000 initial investment leaves a cumulative position at the end of year one of -878,000 THB.

From year two the benefit is assumed to land in full, giving 2,400,000 – 384,000 = 2,016,000 THB per year, or 168,000 THB per month. Dividing the 878,000 still unrecovered at the end of year one by 168,000 per month gives 5.2 months. Cumulative payback therefore works out to 17.2 months, roughly one year and five months.

Sensitivity When the Benefit Does Not Land — How Many Months If Approval Routing Never Changes

An investment decision always has to model the case where things do not go as planned. The most realistic failure scenario here is that R3’s approval routing does not change. The hardware can go in, but moving first-article approval off paper and stamps requires agreement from quality assurance and production control, and there are plenty of plants where that does not move.

The thing to watch here is how the sensitivity is applied. Do not take a blanket coefficient, of the “only 70% of the benefit lands” kind, and multiply it uniformly across every effect. Externalizing R1 and running SMED on R2 work regardless of approval routing. Only R3 changes, and swapping out only that part is the correct treatment.

Assume the R3 reduction stops at 3 minutes instead of 9. Total reduction goes from 20 minutes to 14 minutes. Annually that is 6,000 × 14 = 84,000 minutes = 1,400 hours, worth 1,680,000 THB per year.

The cost side does not change. The layer 4 initial cost and the 384,000 THB annual operating cost are both still incurred, because the hardware and platform are in and only the approval process is stuck. Year one is 1,680,000 × 0.6 = 1,008,000, less 384,000, giving 624,000 THB. The cumulative position at the end of year one is -1,934,000 + 624,000 = -1,310,000 THB.

From year two, 1,680,000 – 384,000 = 1,296,000 THB per year, or 108,000 THB per month. Dividing the 1,310,000 unrecovered by 108,000 gives 12.1 months, so cumulative payback is 24.1 months, roughly two years.

The conclusion is that even with approval routing unchanged, the investment pays back in a little over two years. The practical implication, however, is not that you should invest anyway. It is that buying layer 4 first is the wrong sequence. If there is no prospect of changing R3’s approval routing, start with layers 1 to 3, the measurement platform, alone. The initial cost drops to 1,064,000, and improving R1, R2, and R4 alone still delivers benefit. And once segment measurement produces a number for how many minutes approval wait actually runs, that number is itself the material that moves the quality assurance department. Keep the sequence and layer 4 stops being something you buy and becomes something you buy once the need has been proven.

Points Specific to Plants in Thailand and ASEAN

Everything designed above assumes the shop-floor conditions of a Japanese-owned plant in Thailand. Several points do not transfer directly to a plant in Japan, so they are stated explicitly.

First, multiple languages. Changeover work instructions and shop-floor terminal displays need to be designed with Thai as the first language. On processes with many operators from Myanmar or Cambodia, more languages are added. What works in practice here is less about translating sentences than about showing procedures with images and step numbers and designing terminal operation so that it does not depend on text. Simply assigning colors and process icons to the segment buttons instead of words cuts mispresses substantially.

Second, turnover and retraining. Changeover is work where experience counts and training takes time. In a high-turnover environment, what was taught does not accumulate in the organization and walks out with the individual. The realistic hedge is to move as much of the skilled part of changeover as possible onto the data side. Turning process conditions into recipes is also a mechanism for keeping the feel for dialing in inside the organization.

Third, operator rotation. Japanese plants can often assign dedicated changeover staff, but in Thailand absence and sudden output increases move the rotation, and the person doing the changeover differs from day to day. Under those conditions, improvement premised on an experienced operator being fast does not hold. Investing in kitted preparation and fixed procedures, so that the procedure is the same whoever performs it, gives more stable results. Making segment measurements viewable by operator lets you identify which segment and which person need training.

Fourth, the handling of BOI-privileged equipment. Equipment imported under BOI investment promotion may carry restrictions on modification and relocation. Whether taking a signal out of the PLC or adding external wiring for changeover measurement conflicts with the conditions of the privilege needs to be confirmed in advance. In practice, choosing a non-contact acquisition method that does not touch the inside of the machine, such as reading the stack light with an external sensor, largely sidesteps the issue. Confirm it when designing layer 1.

Fifth, the physical whereabouts of the approver. In Thai plants the quality assurance manager is often a Japanese expatriate who spends long stretches away from the production floor in meetings or with visitors. That is a textbook condition for structurally long approval waits in R3, and electronic approval with delegated-approval rules tends to pay off more than it would in a plant in Japan.

The 90-Day Roadmap

Introducing segment measurement does not end when the hardware is in. How the first 90 days are designed determines whether the habit of looking at the numbers survives afterward. The approach this article recommends is set out in three phases.

The most important thing in this roadmap is to make no improvements during days 0 to 30. Act during the measurement period and you lose the baseline. Without it you cannot prove the effect afterward, and the next investment does not get approved. For 30 days you do nothing but measure the current state. During this period the floor will ask what the point of measuring is, and whether you can show what happens from day 31 onward determines the project’s credibility.

The reason for taking on R3 in days 31 to 60 is not only that it is the largest segment. Changing approval routing requires coordination with other departments, and building agreement takes longer than implementation. Start late and it will not fit inside 90 days. R1 and R2 measures, by contrast, run from start to effect quickly, so the back half is soon enough.

Days 61 to 90 go to designing the embedding. Cut corners here and you follow the same path as Case A. Do not settle for handing out a revised standard on paper. Finish the job of putting segment results back in front of the floor every day. The 4 months of hands-on support carried as a layer 5 cost exist because one more month after these 90 days is needed to confirm that the habit of looking at the numbers runs on its own.

Ten-Point Decision Checklist

Here are the items worth confirming before an investment decision. If more than half come back as no, there are things to sort out before bringing in hardware.

• Is changeover time recorded for each individual changeover, rather than existing only as a monthly average
• Is the record broken into segments, rather than managed as a single figure of 48 minutes
• Are the definitions of changeover start and end documented and shared across the company
• Is it settled who approves the first article and who can stand in when that person is away
• Are process conditions stored on the part-number-and-die combination and recallable next time
• Is changeover time kept inside the OEE denominator, rather than excluded as planned downtime
• Is the treatment of ramp-up time, the R4 equivalent, fixed by internal definition as either changeover or performance loss
• When a machine frees up, is there an order or an insourcing candidate to fill the time
• For BOI-privileged equipment, is there a contact point that can confirm whether modification for signal acquisition is permitted
• Does the display language on the shop-floor terminal match the language of the operators who actually do the changeover

The most important of these is whether there is an order or an insourcing candidate to fill the time freed up. If there is nowhere to use the hours you carve out of changeover, the financial benefit is limited to the labor cost saved, and this article’s 1,200 THB per hour assumption does not hold. Measurement still has value in that case, but keeping the investment scoped to layers 1 through 3 is the sensible call.

Frequently Asked Questions

Where should a changeover and setup time reduction program start?

Start by measuring the current state in segments for one month, before changing anything about the setup time itself. Act first and you lose the baseline needed to prove the effect. Once measured, most plants find that R3, covering adjustment, first-article inspection, and approval wait, is the largest, and the improvement priority shifts away from what was originally assumed. The urge is to start with work you can put your hands on, but the correct sequence puts measurement first.

Is SMED alone not enough?

SMED is effective, and separating internal from external setup remains central to improvement. What is missing is a mechanism for sustaining it. All four SMED steps are designed to be complete once performed, and they have no function for detecting that things went back six months later. Only by adding a fifth step, looking at segment-level results every day, does the improvement settle into a state that holds.

Does measuring changeover time require modifying equipment?

Not necessarily. Reading the stack light state with an external sensor captures stop and start timing without touching the inside of the machine. Choose this non-contact method where BOI-privileged equipment restricts modification or where the machine builder’s warranty terms are strict. Equipment signals alone cannot cut the segment boundaries, though, so the design combines them with operator input on a shop-floor terminal.

What data does changeover time reduction analysis require?

At minimum, the start and end times for R1 through R4 for each individual changeover, the machine number, the part numbers before and after the switch, and the operator who performed it. Add the first-article inspection request and approval times and the process condition settings, and the components of R3 can be broken out as well. Put the other way around, a monthly total of changeover time and a count of changeovers do not constitute analysis. Per-changeover granularity, split into segments, is the precondition for analysis.

How much does it cost?

In this article’s model of 12 machines, the assumption is 1,934,000 THB initial and 384,000 THB annual operating across five layers. This is an estimate and varies widely with equipment type, existing wiring, and whether business systems are in place. Excluding layer 4, which covers electronic approval routing, and scoping to layers 1 through 3, the measurement platform, the initial cost is 1,064,000 THB. Starting from measurement alone can begin within that range.

Can high-mix low-volume production recover the investment?

High-mix low-volume tends to pay back faster. Because changeovers are frequent, the reduction per changeover is multiplied by a larger count. This article’s model assumes a modest 2 changeovers per machine per day, and at 3 or 4 the benefit scales proportionally. What does not change is the premise that there are orders to fill the time freed up. Where the order book is not full, the benefit appears only on the labor cost side, so scoping the investment to the measurement platform is the sensible call.

Summary

Changeover reduction reversing within six months is not a problem with SMED as a method. It is that changeover is only ever measured as a single lump, so nobody can see which part went back. 48 minutes becomes 42, and six months later it is 43.5. As long as you measure the lump, no move comes out of that 43.5.

Split changeover into four segments and the structure becomes visible. Under this article’s model assumptions, R1 is 4 minutes, R2 is 13, R3 is 20, and R4 is 11. R2, where most floors concentrate their improvement, is only 27% of the whole, and the largest is R3 at 42%. R3 is also weighted toward waiting time, and because film shows no visible room for improvement it sits largely unmeasured. R4 goes further still and disappears from the changeover ledger on the grounds that the machine is running.

Measuring by segment requires combining equipment signals, operator input, and business system events. No single one of them cuts the boundaries. And building the measurement platform in layers preserves the option to stop partway. In this article’s model the estimate came to 1,934,000 THB initial and 384,000 THB annual operating, with payback in 17.2 months. Even with approval routing unchanged it is 24.1 months. In that case, though, layer 4 should not be bought first. Starting with layers 1 to 3, the measurement platform, is the correct sequence.

In 2026, with Thai manufacturing utilization in the 57% range and the Bangkok minimum wage at 400 baht per day, changeover time reduction is a way to win back usable hours without adding machines. Whether there is somewhere to use the hours you cut, though, needs to be confirmed before the investment decision.

You do not need to know how many minutes your changeovers actually run, or which segment the time is concentrated in, before talking through how to proceed. Even sorting out which segment to measure first, and how far signals can be taken from your existing equipment, raises the precision of the investment decision. For help designing a measurement approach that fits your shop-floor conditions, reach us through our contact page at any time.

References

• Thailand Office of Industrial Economics (OIE) Industrial Indices
• Bank of Thailand Economic and Financial Conditions in June and the Second Quarter of 2026
• Xinhua Thailand Industrial Output in June 2026
• Thailand Capacity Utilization Series Data
• ISO 22400-2 2014 Key Performance Indicators for Manufacturing Operations Management
• How to Calculate OEE and the Six Big Losses
• OEE Benchmarks by Industry 2026
• JETRO Bangkok Minimum Wage of 400 Baht per Day