Thailand Labor Cost Increase 2026 | How to Spot the Processes Automation Can Pay Back

“Wages went up, so we want to look at automation.” Inquiries from Japanese-owned plants in Thailand very often open with exactly that sentence. Yet when you look at what has actually happened since the start of 2026, the hourly rate itself is not what has risen. If you are weighing automation as a countermeasure to Thailand labor cost increases, the first thing to pull apart is not the wage rate but the costs stacking up outside the wage, and whether you will still be able to staff that process next year. This article breaks labor cost into five layers, models the payback across four scenarios, and shows where the line falls between processes that pay back and processes that do not.

Before you answer Thailand labor cost increases with automation, break down what is actually rising

Deciding to automate because labor costs are climbing is not the wrong direction. Studies stall halfway because the payback formula does not match the real shape of the cost. The formula put on the table at the first meeting almost always looks like this.

Headcount removed × hourly rate × annual operating hours, divided by the investment amount.

Numerator and denominator both look tidy at a glance. But the formula misses reality in three places.

First, the hourly rate in the numerator is not rising. As of 2026 the minimum wage in Thailand is on hold, so any model built on an assumed increase in the wage rate collapses at its own premise.

Second, what disappears first when automation goes in is not people but overtime hours. Overtime carries a premium, so the cost per hour is higher than for normal hours. The formula, however, values the saving at the normal hourly rate, and therefore understates the effect. Conversely, if you enter the saving as headcount and write “three people fewer,” you ignore the fact that the shop floor holds on to its people to the very end, and now you overstate it. Either way the number is not accurate.

Third, the formula contains no term for becoming unable to hire. The rate at which recruiting fails is moving faster than the rate at which the unit cost rises. Even with a flat rate, production stops once you can no longer hold the headcount the process needs. The loss from a stopped line cannot be expressed as a difference in hourly wages.

The decision axis is therefore not how many baht the hourly rate will rise. It is whether you can run that process next year with the same number of people. Only the processes where the honest answer is “questionable” are worth an automation study.

The minimum wage is on hold, and what is rising sits outside the wage

Start with the facts. The minimum wage in Bangkok was raised from 372 baht to 400 baht per day on 1 July 2025. The National Wage Committee approved it on 17 June 2025, and roughly 700,000 workers were reportedly covered. In the manufacturing belt, four provinces and one district including Chonburi and Rayong had already moved to 400 baht from January 2025, ahead of the capital.

And there has been no revision since the start of 2026. Nationwide, the level continues to sit between 337 baht and 400 baht per day. In other words, the part of the “labor costs went up” feeling now common on the shop floor that comes from a higher wage rate came to a stop in 2025.

The total still rises. What rises sits outside the wage.

One driver is the ceiling on the social security assessment base. Under a revision published in the Royal Gazette on 12 December 2025 and effective from 1 January 2026, the ceiling on the assessment base was raised from 15,000 baht to 17,500 baht per month. The floor of 1,650 baht is unchanged and the 5% rate is also held. Only the ceiling moves, and it moves in stages: the first stage of 17,500 baht runs from 1 January 2026 to 31 December 2028, then 20,000 baht from 2029 and 23,000 baht from 2032. The amount borne by employer and insured person each rises from 750 baht per month to 875 baht, and thereafter to 1,000 baht and 1,150 baht.

The other driver is the cost that comes with turnover. Job advertising, interviews, onboarding paperwork, and the training needed before a new hire is productive. None of this appears on the payroll ledger under the heading of labor cost, but in a high-turnover process it keeps on occurring. Even with the wage held flat, total labor cost rises if turnover rises. Ignore this and chase the hourly rate alone, and your calculated result will never match the felt increase in cost.

Thailand’s worker shortage bites as an inability to hire, not as a higher rate

Now the structural picture. The NESDC (Office of the National Economic and Social Development Council) projects that Thailand’s labor force will shrink by more than 3 million people every decade from here. The labor force stood at 40.7 million at the end of 2023, while labor demand in 2037 is projected at 44.71 million. Even on a simple subtraction, roughly 4 million people are missing.

The same work suggests that lifting productivity by around 5% through automation could cut labor demand by more than 2 million people. At the same time, the automation adoption rate in Thai manufacturing is understood to sit at only around 5%. You could call that plenty of room to grow, but read from the shop floor the meaning is the opposite. Most plants are still doing the work with people. Which means the pattern of neighboring plants competing for the same limited labor pool continues.

This shortage does not show up quickly as a number on the payroll ledger. It shows up like this. You post a vacancy and no one applies. Someone joins and does not stay. You pull relief from another process to cover the gap. Overtime rises in the process that lent the relief. A state in which only one person can do a particular job becomes permanent. The month that person resigns, the production plan falls apart.

What is unfolding on the way to 2037, then, is not a unit cost problem but a continuity of production problem. Build the automation case on the labor cost saving alone and the value of the investment as insurance against that continuity risk drops out of the calculation entirely. Put the other way round, a project whose payback cannot clear the hurdle on savings alone can still be decided once continuity is added. The four scenarios later in this article are the framework for converting that addition into money.

On how to stage which processes you hand to machines and which you keep with people, reading the four-stage roadmap for factory automation in Thailand first makes it easier to place which stage the modelling here belongs to.

Break the real cost of labor into five layers, because wages alone will throw off the payback

Here is the substance. When you employ one operator for one month, what is the company actually paying? On the assumption of a plant near Bangkok and a minimum wage of 400 baht per day, build it up in five layers.

Workmen’s compensation rates differ by industry, so about 0.5% is placed here as an assumption. Substituting your own rate moves the Layer 4 figure, but the skeleton of the discussion that follows does not change.

A note on the Layer 4 assessment base. This model puts the monthly wage including premium pay into the assessment base for social security and workmen’s compensation. That is, the sum of Layers 1 to 3, 10,400 + 1,600 + 3,000, gives a base of 15,000 baht. How much to include in the assessment base is an area where interpretations differ, so the safe course is to confirm your own practice with your advisers and substitute accordingly. The figure lands exactly at the pre-revision assessment ceiling of 15,000 baht. Applying 5% gives the employer’s social security share of 750 baht and 0.5% gives 75 baht of workmen’s compensation, for 825 baht in total.

Note how each layer behaves. Layers 1 and 2 occur in proportion to headcount and vary little month to month. Layer 3 moves every month with production volume and how fully the process is staffed. Layer 4 is set by wage level and rate, and for anyone below the ceiling it tracks Layers 1 to 3. Layer 5 lands in a lump in the month a resignation occurs, so it is invisible monthly and only becomes an amount once averaged over the year.

When you rebuild this with your own numbers, Layers 3 and 5 are the easiest ones to get wrong. The 40 hours of overtime in Layer 3 vary widely by process, so entering a company-wide average pulls you away from the reality of the target process. That number should come from time and attendance records broken out by process. Layer 5 is the product of two assumptions: the turnover rate and the recruiting and training cost per person. Turnover of 20% swings sharply by process, so replace it with the actual figure for the target process. The 8,000 baht of recruiting and training also shifts with the sourcing channel and the ramp-up period. Substitute your own values for these two and Layer 5 can move from 133 baht to several times that.

Take the total to an annual figure and 15,958 × 12 = 191,496, or roughly 191,500 THB per year.

What internal discussions usually use, on the other hand, is Layers 1 and 2 only. 10,400 + 1,600 = 12,000. Annualized, 12,000 × 12 = 144,000 THB per year. Set the two side by side and the gap is plain.

191,496 ÷ 144,000 = 1.3298. The real cost is about 1.33 times base wage plus allowances.

Whether you include that 1.33 coefficient moves the payback substantially. Suppose you recover an investment of 4,000,000 baht from the labor cost of three direct workers. Divide by 144,000 × 3 = 432,000 and you get 9.3 years; divide by 191,496 × 3 = 574,488 and you get 7.0 years. Same equipment, same process, same headcount, and more than two years apart. That is a gap wide enough to decide whether an approval request passes.

What matters is that this 1.33 is not a padded figure. It does not include the future obligation for severance pay. As long as you carry the people, the obligation for compensation tied to length of service keeps accumulating. The amount varies with the individual service profile and the interpretation of the law, so this article does not model it, but it is reasonable to assume the real cost sits above 1.33 times rather than below.

Administrative cost is not included either. Each additional person adds attendance aggregation, shift adjustment, training record keeping, and health and safety handling. These get absorbed as indirect department workload and so do not look like labor cost, but they are not costs that go away.

Who the social security ceiling increase affects, and who it does not

Follow only the tone of the coverage and the assessment ceiling increase from January 2026 reads as “labor costs are going up again.” Check whose cost actually rises inside the plant, however, and the story gets one level finer.

The employer share rises from 750 baht to 875 baht per month for people whose monthly wage exceeds the assessment ceiling. The revised ceiling is 17,500 baht. So for the tier earning 17,500 baht a month or more, the share pins at 875 baht, an increase of 125 baht per person per month, or 1,500 baht per year.

What about an operator at entry level? As confirmed in Layer 4, the monthly wage forming the assessment base, overtime included, is around 15,000 baht. That does not reach the revised ceiling of 17,500 baht. So even with the ceiling moving from 15,000 to 17,500, the employer share for this tier stays at 750 baht. The amount owed is 5% of the assessment base, and anyone below the ceiling is unaffected by where the ceiling sits. The only tier whose share rises to 875 baht through the revision is the tier already pinned at the ceiling.

The point to hold on to is that the plant splits into a tier whose cost increases and a tier whose cost does not.

How that split falls feeds directly into the automation investment decision.

Argue that “cutting operators lowers cost” on the grounds of higher social security charges and the argument does not hold, because cost has not increased for that tier. The tier where cost certainly rises is leaders, technicians and staff. That tier is not the target of automation. If anything, adding equipment increases the maintenance and changeover load, and the required headcount in that tier moves upward.

The ceiling increase is therefore not ammunition in favor of automation investment. It is a cost that grows after automation. If it goes into the model at all, it belongs on the additional cost side, not the savings side. Get that the wrong way round when you brief head office and the numbers will fail to reconcile later.

How to tell which processes pay back in Thailand labor saving projects and which do not

With the real cost of labor established, the next question is how to choose the target process. Install the same value of equipment and the payback can differ by more than a factor of two depending on the process. The conditions that separate them come down to four.

Processes meeting three or more of the four conditions come first. Conversely, a process meeting neither condition 1 nor condition 3 will show a longer payback no matter how many people work in it. A large headcount is not in itself grounds for payback.

A note on condition 2. What you are looking at is not difficulty but the amount of variation. A hard job can be handed to a machine if the conditions are settled. An easy job in which a person decides each time that “today’s workpiece is slightly warped, so hold it differently” requires that judgment to be mechanized too. The latter is where the investment jumps. Projects where the quotation comes back higher than expected almost always trace to condition 2. Go out for inquiries without settling the state of the process and the integrator has no choice but to price the uncertainty as risk. How to draw that boundary is covered in choosing a robot system integrator and the boundaries of responsibility in a quotation.

Conditions 1 and 3 can be turned into numbers from records. For condition 1, lining up overtime hours and weekend working days by process for the past 12 months is enough. What you look at there is not the monthly average but the gap between peak and slack months. A process with a large gap needs a specification sized for the peak month, which inflates the investment. A process with steady monthly overtime, by contrast, is easier to size and the effect is more stable. Condition 3 is about whether there is work to run in the hours you want the equipment moving, so it becomes a question for the production planning side. If a process runs day shift only because you cannot crew a night shift, that process’s order backlog is the answer to condition 3.

Condition 4 is the hardest to put in money terms, yet in practice it is the one that bites hardest. Depending on a single person for a particular job is not a labor cost problem but a risk problem. And in an environment with turnover above 20%, that risk materializes with a fixed probability every year. It does not enter the payback calculation, but as a reason for a decision it is quite strong enough.

There are three classic ways to invest in a process that will not pay back. The first is choosing the process with the most people. Headcount is conspicuous, but if that process closes within regular hours the only thing you can remove is low-cost normal time. The second is choosing the process the floor complains about most. The scale of the complaint is proportional to how unpleasant the work is, not to how large the cost is. The third is choosing the process another company has automated. A peer’s configuration was set to fit that company’s model mix and order volume. If your conditions 2 and 3 differ, the same equipment gives a different payback.

In every case the basis for selection sits outside the process, in headcount, complaints, or peers. All four conditions look at the state of the target process itself. Moving the basis for selection back inside the process is the first piece of work.

Deciding the type of equipment first and then hunting for a process is another failure you see often. Enter from the conclusion that you want to automate transport and replacing transport processes by introducing AGVs and AMRs becomes the goal in itself, and the check on how many people spend how many hours on that transport in the first place is skipped. Where people and machines are to work in the same space, placing a collaborative robot alongside the operator is an option too, but either way the premise is that the process satisfies the four conditions.

Model the automation payback across four scenarios

From here the numbers sit on a concrete project. The target is semi-automation of inter-process transport and case packing. The investment is 4,000,000 baht, broken down as follows.

The equipment itself is 70% of the total, and the remaining 30% is installation, jigs, software integration, spares and training. Drop that 30% out of the quotation when you compare options and the budget will fall short later. Jigs and software integration in particular move with conditions on the process side, so at the rough-estimate stage they need a range around them.

Next, build up the savings side. Take the real labor cost of three direct workers and subtract the additional running cost of keeping the equipment moving.

Here is the make-up of the weekend working figure. Assume attendance twice a month, 10 hours each time, three people. Of those, 8 hours are holiday work at twice the normal rate, 100 baht per hour. The remaining 2 hours are overtime on a holiday at three times, 150 baht per hour. Working it through, (8 × 100 + 2 × 150) × 3 people × 2 occasions = 6,600 baht per month, or 79,200 baht per year. Headcount stays at three, but that 79,200 baht disappears for certain.

For added output from unmanned night running, 4 hours × 20 days of extra operation is entered as 40,000 baht per month of marginal profit, or 480,000 baht per year. This item requires an assumption that the output sells, so a plant with no evidence on the demand side cannot book it.

Combine all of the above and you get four scenarios.

Check the arithmetic. 4,000,000 ÷ 514,488 = 7.775, so 7.8 years. 4,000,000 ÷ 593,688 = 6.737, so 6.7 years. 4,000,000 ÷ 1,073,688 = 3.726, so 3.7 years. In scenario D, where the effective investment falls, 2,000,000 ÷ 1,073,688 = 1.86 gives 1.9 years and 3,000,000 ÷ 1,073,688 = 2.79 gives 2.8 years.

Replacing labor cost alone will not clear the five-year hurdle

How should you read the 7.8 years in scenario A? This article sets the payback hurdle at five years. Internal standards differ by company, but the premise is that beyond this a head office approval request becomes hard to pass.

On that premise, 7.8 years does not pass. And note that scenario A is not a half-hearted model. Labor cost is entered at the real figure of 191,496 baht rather than the base wage, and 60,000 baht of additional running cost is properly deducted. In other words, stacking up the idea of labor cost replacement as far as it will go still only reaches this level.

Why it falls short becomes clear from the structure of the numerator. The 514,488 baht per year is only 12.9% of a 4,000,000 baht investment. In a country where the real cost of one operator is 191,496 baht per year, replacing three people with one machine caps the effect in the 500,000 baht range per year. That cap does not rise while wages are on hold. Which means waiting will not shorten the payback.

Two conclusions follow. One is that in Thailand, an automation approval request justified purely on the labor cost saving basically does not pass. The other is that where such projects are passing, effects other than labor cost are in the calculation.

It is tempting to stack the reduction to four or five people to shorten the years, but that breaks down in execution. The shop floor will not release those people for some time after the equipment arrives. During ramp-up people need to stand and watch, and when there is trouble the work reverts to manual. Headcount reduction comes last. Write the headcount in early to get the approval through and the following year’s actuals will disagree with it, without exception. How far savings were actually realized on projects of this kind is covered in labor saving cases and cost effectiveness tracked at real figures.

What clears it is extending operating hours

What scenarios B and C show is that there are two directions in which to grow the numerator.

Scenario B adds 79,200 baht by eliminating weekend working. The effect moves 7.8 years to 6.7 years, an improvement of 1.1 years. It still does not clear the hurdle, but this item has the advantage of high certainty. Weekend working is in the attendance records, so the saving can be verified after the fact. Unlike headcount reduction, resistance from the floor is small. The size of the effect is moderate, but its power to explain itself in an approval request is strong.

Scenario C adds 480,000 baht through unmanned night running. Here the payback becomes 3.7 years and clears the hurdle. The effect is an order of magnitude larger because this is not a cost reduction but an addition of time. Run the equipment in hours you cannot staff and the effect accumulates independently of the ceiling on savings. Same equipment, same headcount, and only the operating hours grow.

Understand that structure and the way you choose the target process changes. Instead of the process with the most people, you go looking for the process where you want the equipment running but cannot put anyone there. A process running day shift only because a night crew cannot be secured, a process you want to run at holidays but cannot staff. These become the first candidates.

Scenario C does come with a premise, though: the extra output has to sell. The 40,000 baht per month of marginal profit only holds if there is evidence on the demand side that added volume will be bought. Produce at night while orders are flat and inventory simply grows; the effect is not merely zero but negative. A plant where that premise does not hold must not put scenario C into the model.

The other premise is that the equipment actually runs. Extra operation of 4 hours × 20 days assumes the equipment does not stop in those hours. Stop when no one is watching and everything up to recovery the next morning becomes loss. The 60,000 baht per year of additional running cost covers power, maintenance and consumables, and stoppage loss is not inside it. So if scenario C goes into the model, how to secure continuous unattended running time becomes a design item. Stocker capacity on the feed side, receiving capacity on the discharge side, automatic stop on fault with the state held until morning. Book the effect of night running on equipment where these three are not in the design and the actuals will not reach the model.

Where they cannot be secured, the alternative is to use unmanned night running for leveling rather than added output. Push load concentrated in the day into the night and daytime overtime and weekend working fall. The effect then sits on the extension of scenario B, smaller in money terms but free of any demand premise. Levelling without holding inventory often presupposes automating storage and retrieval, in which case the price bands for automated warehouses and how to judge the investment is worth reading alongside this.

Cut overtime before headcount, and the order in which to take savings

Comparing the scenarios shows that savings have an order to them. The principle is to take the items with the highest unit cost and the easiest execution first.

There are reasons for the order.

Overtime ranks first because of unit cost. As set in Layer 3, the overtime rate is 1.5 times the normal hourly rate of 50 baht, at 75 baht. If you are removing one hour either way, removing an hour of overtime is 1.5 times as effective as removing an hour of normal time. Overtime is also in the attendance records, so the before and after comparison comes out as a number. Being able to see the effect the month after the equipment goes in counts for a great deal in building internal agreement.

Weekend working, in second place, carries an even higher unit cost. Twice normal, or three times for overtime on a holiday. The model here reaches 79,200 baht per year at twice a month; a process working every weekend would be more than four times that. But weekend working usually happens because “we miss the delivery date if we do not come in that day,” so it cannot be eliminated unless equipment capacity is sufficient. Elimination of weekend working needs to be designed into the capacity from the start.

Resilience to absence, third, is the hardest area to reduce to money. Relief when someone is out, additional agency labor, training hours for multi-skilling. These are visible only in the month they occur and rarely appear in the annual budget. Even so, ignore the item and part of the value of automation vanishes from the calculation. To pin it down with numbers, the practical approach is to pick up the days on which absence occurred over the past year and the extra overtime hours incurred at those times.

Headcount is fourth. It sits last because the unit cost of the saving is the lowest and execution is the hardest. The unit cost of one person is base wage and allowances with statutory benefits on top. Against overtime and weekend premiums, that is low-cost time. On top of that, headcount reduction involves reassignment and retraining and takes time to realize. It does not happen at the same moment the equipment starts up.

Why a model built in the reverse order is dangerous is that the payback then rests on a premise that will not be realized. A model that puts headcount reduction first produces the shortest payback on paper. But if headcount does not fall in execution, that number is never achieved. A model built up from overtime and weekend working gives a longer figure, but it is achieved. Keeping the number that wins approval and the number you can actually achieve aligned pays off more over the long run.

There is one more practical caution. Cut overtime and the operator’s take-home pay falls. The 3,000 baht in Layer 3 is a cost to the company but income to the individual. It is not unusual for resignations to rise in the month overtime disappears. If the 20% turnover set in Layer 5 deteriorates, the cost you thought you had saved comes back as recruiting and training. Alongside cutting overtime, you need to consider reviewing base wages and allowances, or redistributing through something like a multi-skill allowance. Labor saving that does not build this in sees labor cost return to where it started within a year.

The BOI productivity improvement measure, and the traps on the scheme side

Scenario D lowers the effective investment to between 2,000,000 and 3,000,000 baht because it assumes the BOI productivity improvement measure. The outline of the scheme is understood to be as follows.

The minimum investment is 1,000,000 baht or more (500,000 baht or more for small and medium enterprises). Investment in replacing manual work with automation is reportedly granted a three-year corporate income tax exemption, with the exemption capped at 50% of the automation investment. Where conditions such as a purchase ratio of 30% or more from domestic manufacturers are met, the cap is understood to rise to 100%.

That said, the description of this measure is kept to the outline of the scheme. BOI announcements are revised, so confirm with the BOI and your advisers before applying. The figures in this article must not be used as your own application conditions. The scope of the incentive, the application deadline, the relationship with existing operations, and the required documents are areas where the judgment differs project by project, and the places to confirm are the BOI and your accounting and legal advisers.

Three traps on the scheme side are worth naming.

The first is that a corporate income tax exemption has no value without taxable income. Even if the exemption cap is 50% of the automation investment, you cannot use up the allowance without commensurate taxable income over three years. For a company running losses, or one already inside an exemption period under an existing incentive, part of the allowance is lost. The range of 2,000,000 to 3,000,000 baht on the effective investment in scenario D expresses exactly this uncertainty about the utilization rate. Build an approval request on the lower bound of the range and the actuals will not reach it.

The second is the condition on where you buy. If a purchase ratio from domestic manufacturers is what lifts the cap to 100%, then the choice of supplier feeds directly into the value of the incentive. A configuration importing Japanese-built equipment from Japan and a configuration built inside Thailand differ not only in the investment itself but in the effective investment. This viewpoint needs to be shared with procurement before the equipment specification is frozen. The comparison between local build and Japanese build is also touched on in the questions later in this article.

The third is the order of application and ordering. It happens that a model is built on the incentive and then, in practice, the purchase order goes out before the application. What has to be filed by when depends on the detail of the scheme, so here again the places to confirm are the BOI and your advisers. If you want the whole picture of the scheme first, see the article setting out how factory automation in Thailand relates to BOI incentives.

In terms of how to write the approval request, it looks different depending on whether you put the incentive into the denominator as a reduction in the effective investment or show it separately as a lighter tax burden. Scenario D here puts it into the denominator, but that form makes the payback period stretch sharply if the incentive turns out not to be usable as assumed. Fix the denominator at 4,000,000 baht and write the incentive as a separate incidental effect, and the main scenario stays at 3.7 years while the incentive is treated as upside. If you are deciding while carrying scheme uncertainty, the latter is less likely to fall apart in the telling.

A model resting on a scheme collapses when the conditions change. So the safe course is to put the main scenario in the approval request at C, 3.7 years, and note D alongside as the upside if the conditions come together. Put the 1.9-year figure at the center of your case and the whole project’s rationale collapses the moment the incentive turns out not to be usable as expected.

A 90-day path to a go or no-go decision

Now turn all of the above into a procedure that reaches a decision. Take 90 days as the guide. A longer study does not add information, and the labor situation moves on ahead of you.

Phase 1, roughly 30 days, is for assembling the numbers. There are three tasks. First, for candidate processes, extract overtime hours and weekend working days for the past 12 months, broken out by process. Broken out by process, not as a company total, is the important part, and a plant that cannot extract that will start with the aggregation itself. Second, rebuild the real labor cost in five layers under your own conditions. Layer 4 workmen’s compensation rates and Layer 2 allowances differ by company, so do not use the 15,958 baht in this article as it stands; build it with your own values. Third, pick up the absences over the past 12 months and the extra overtime they caused. With those three in hand, screening processes against the four conditions becomes a numerical exercise.

Phase 2, roughly 30 days, is for settling the conditions on the process. Narrow to one or two target processes and write out the workpiece shape, feed orientation, changeover frequency and method, and the required takt time. Go out for inquiries without settling this and the quotation will carry a price for uncertainty. At the same time, confirm with sales whether unmanned night running goes into the effect. If sales cannot answer whether the extra output will sell, scenario C is unusable and you switch to designing for leveling. Defer that decision and you will be redoing the capacity design.

Phase 3, roughly 30 days, is for setting money against conditions. Take rough quotations and run your own numbers through the four scenarios to produce payback figures. In parallel, confirm with your advisers whether the BOI incentive is available, and build both a payback assuming it is not and one assuming it is. Then decide the main scenario for the approval request.

If those three phases do not produce a decision, the cause is almost always that the Phase 1 numbers are not in place. In a plant with no overtime actuals by process, there is no basis for deciding which process is the candidate, so the study drifts into an argument about equipment types. Consult outside help in that state and the process selection gets done together with you, which extends the timeline. Whether to take the procedure for choosing target processes with automation consulting first can be decided simply on whether you can produce the Phase 1 numbers in house.

Decide up front who to involve as well. Phase 1 needs HR and administration for the five layers of labor cost and the attendance actuals, and manufacturing for overtime and absence by process. Phase 2 is led by production engineering and maintenance, plus sales to confirm whether added output is saleable. Phase 3 brings in finance and the advisers handling BOI incentives. Involve them out of that order and Phase 3 produces objections along the lines of “sales cannot guarantee that effect” or “that incentive is not available in our situation,” and the model has to be rebuilt. What breaks the 90 days is more often a delay in involving people than a delay in the technical study.

Reaching a decision of “no” in 90 days is also a legitimate outcome. Investing in a process that meets none of the four conditions only lengthens the payback. But do not stop at “no” without answering the question of whether you can run that process next year with the same number of people. If the answer is “questionable,” you need a plan to address that with something other than automation.

Frequently asked questions

How much do you need to start automation as a Thailand labor cost countermeasure?

The semi-automation of inter-process transport and case packing modelled here is a 4,000,000 baht project, but that figure is not a floor. The fact that the BOI productivity improvement measure sets a minimum investment of 1,000,000 baht or more (500,000 baht or more for small and medium enterprises) shows that projects on a 1,000,000 baht scale are contemplated by the scheme as well.

Rather than hunting for a floor on the amount, checking the floor on the effect is more practical. If you put an annual 514,488 baht from labor cost replacement alone into the numerator, recovering in five years means the investment has to fit inside roughly 2,500,000 baht. Conversely, if you are spending 4,000,000 baht, the payback will not arrive unless extended operating hours are added to the effect. The reasonableness of the investment, in other words, is decided by the structure of the numerator, not by the absolute size of the amount.

There are two ways to think about starting small. One is to narrow the scope, limiting the project to transport only or case packing only. The other is to defer the integration with existing equipment, but cutting software integration means you cannot pull the actual data and so cannot verify the effect, which makes it a poor thing to cut first. If you are cutting, cut the scope of the target process rather than spares and training.

Will automation really solve Thailand’s worker shortage?

At the level of a single plant, reducing the headcount you need makes you less exposed to the shortage. NESDC work also suggests that lifting productivity by around 5% could cut labor demand by more than 2 million people, so the direction is sound. On the other hand, given that the automation adoption rate in Thai manufacturing is understood to sit at only around 5%, it is hard to assume the shortage itself will be resolved by automation.

The practical answer is this. Automation does not solve the shortage; it is a way to shrink the surface exposed to it. You remove the state in which only one person can do a particular job, and you cut the path by which a failed recruitment leads straight to stopped production. On a projection where labor demand grows to 44.71 million by 2037, shrinking that surface is the only realistic way to protect continuity of production.

The priority target is therefore not the process with the most people. It is the process that stops when someone is absent. Handle the processes matching condition 4 of the four conditions first and, even where the labor cost saving is small, resilience to the shortage rises sharply.

Can you think about labor saving at an overseas plant the same way as in Japan?

The framework is the same but the coefficients differ. The differences come down to three.

First, the ratio of wage to real labor cost differs. In the model here the real cost was about 1.33 times base wage plus allowances. That coefficient is set by the level of statutory benefits and turnover, so it is not a number you can carry over to a plant in Japan. Before producing a payback figure you have to rebuild the five layers under the conditions of your own country and site.

Second, the way the wage rate moves differs. In Thailand the minimum wage is on hold as of 2026, and the level continues to sit between 337 baht and 400 baht per day. Build a growing future saving on assumed wage increases into the model and the premise fails. The “it will keep rising” assumption used in a study in Japan cannot be carried across as it is.

Third, the maintenance setup differs. Whether you can fix the equipment yourself after it goes in governs the effective utilization. If unmanned night running goes into the effect, the premise is a setup that can restore operation by morning. A design that redirects part of the headcount you removed into maintenance needs to be built in at the investment stage. Cut that and the 480,000 baht in scenario C stays a number on paper.

What automation payback period is reasonable?

This article set five years as the hurdle. That is a premise, though, and what is reasonable as an internal standard depends on how many years you will use the equipment and how many years the target process will continue.

In practice you should look at the structure of the payback rather than its absolute value. Set 7.8 years and 3.7 years side by side and, of the 4.1-year gap, 1.1 years comes from eliminating weekend working and the remaining 3.0 years from extending operating hours. Most of the improvement, in other words, is not labor cost reduction. If you can explain that breakdown, a decision can be made even at 7.8 years. Conversely, even a short 3.7 years will not be achieved if the breakdown depends on headcount reduction that is unlikely to happen.

Another view is to compare against the payback of not investing. If the answer to whether you can run that process next year with the same number of people is “questionable,” then the comparison is not against “the status quo with a payback of zero.” The comparison is a state in which you have lost output in the month someone was absent, incurred overtime in other processes to provide relief, and paid additional recruiting and training. Make that comparison explicit and 6.7 years looks different.

For Thailand automation equipment, is a local build or a Japanese build better value?

Compared on investment alone, a local build often comes out ahead, but there are three things to weigh.

The first is the effective investment. Under the BOI productivity improvement measure, the exemption cap is understood to change with conditions such as a purchase ratio of 30% or more from domestic manufacturers. The make-up of your sourcing therefore affects the value of the incentive. Because the conditions are revised, however, confirm with the BOI and your advisers before applying. Complete that check before the equipment specification is frozen, or you will not be able to change suppliers afterwards.

The second is response speed after start-up. Is it a configuration that can be attended to locally when there is trouble, or one where parts have to be shipped from Japan? On a project that books unmanned night running as an effect, that difference shows up directly as money through the utilization rate. The model here sets additional running cost at 60,000 baht per year, but on a slow-response configuration this cost and the stoppage loss both grow.

The third is how firm the specification is. A project where the workpiece conditions are settled and can be conveyed on a drawing gives predictable quality even on a local build. Send a project whose conditions are not settled to the cheaper option and rework multiplies until the total flips the other way. The order of judgment is not “where do we build it” but “is the specification settled” first. Settle it and then compare, and neither choice will be far wrong.

Summary

To pull the argument together.

Thailand’s labor cost, in the sense of the wage rate, is on hold as of 2026. Bangkok moved from 372 baht to 400 baht on 1 July 2025, but there has been no revision since, and nationwide the level continues between 337 baht and 400 baht. What is rising sits outside the wage: the social security assessment ceiling and the recruiting and training cost that comes with turnover.

The real cost of labor is made up of five layers and comes to 15,958 baht per month for one operator, or roughly 191,500 baht per year. That is about 1.33 times the 144,000 baht you see looking at base wage and allowances alone. Produce a payback without that coefficient and you will be off by more than two years.

The effect of the social security assessment ceiling rising from 15,000 baht to 17,500 baht, with the employer share going from 750 baht to 875 baht, does not reach the operator at entry level. It reaches the leader, technician and staff tier, and that tier is the side that grows with automation rather than the side that shrinks. The ceiling increase therefore belongs in the model as an additional cost, not as grounds for a saving.

On a 4,000,000 baht project, labor cost replacement alone gives a payback of 7.8 years and does not clear the hurdle. Add elimination of weekend working for 6.7 years, and added output from unmanned night running for 3.7 years. With the BOI productivity improvement measure the figure narrows to between 1.9 and 2.8 years, but announcements are revised, so confirm with the BOI and your advisers before applying, and the allowance cannot be used up without sufficient taxable income across three years.

The order of savings is overtime, weekend working, resilience to absence, then headcount. A model that puts headcount first produces the shortest figure on paper but is not achieved in execution.

And the decision axis is not how many baht the hourly rate will rise. It is whether you can run that process next year with the same number of people. In an environment where labor demand grows to 44.71 million by 2037 while the labor force shrinks by more than 3 million every decade, having an answer to that question is itself the investment decision.

It is perfectly reasonable to start at the stage of working out which of the four conditions apply to your own processes, or what the five layers of labor cost look like rebuilt under your own conditions. TOMAS TECH is based in Bangkok and works with Japanese manufacturers in Thailand on FA system construction, control panel design and fabrication, and production management system implementation, and we take inquiries from the concept stage before any decision to invest has been made. Even if you only want to check how the numbers are set, get in touch through the contact page and we will help.

References

• https://www.jetro.go.jp/biznews/2025/07/b21007a1ac8f7fca.html
• https://www.jetro.go.jp/biznews/2025/12/98e85145415653f4.html
• https://bangkokshuho.com/thaieconomy-87/
• https://bangkokshuho.com/thainews-948/
• https://www.jetro.go.jp/world/asia/th/invest_03.html
• https://www.boi.go.th/upload/content/BOI%20Thailand%20Investment%20Promotion%20Policy%20for%20Automation%20Robotics%20Industries%20Japanese_619b411183526.pdf
• https://www.jetro.go.jp/biz/areareports/special/2024/0303/f5b4d6344434b2a9.html
• https://kuno-cpa.co.jp/thailand_blog/