Material Misfeed Prevention 2026: What Works Is Removing the Error Conditions, Not Faster Verification

The first thing that comes up in almost every material misfeed prevention discussion is a request to tighten verification. Have the operator scan with a handheld. Have them scan twice. Keep a record of the scan. And yet, walking the floor, you find plenty of plants that are still producing misfeeds after verification went live. The reason is not subtle. Verification is the last move you make inside a world where the error is still possible; it is not the move that eliminates that world. This article works through how to design against errors on the input side, starting from where it differs from outbound inspection design, and then puts numbers to it using a model factory in Thailand with a compounding process, in a form you can re-check on the spot.

One note before the numbers. Every amount and incident count in this article that is labelled as part of the cost model is an original calculation derived from the shared assumptions set out below. None of it is an industry average or a survey result. The effectiveness rates and the incident counts are assumed values placed here as a starting point for discussion, not measurements. Substitute your own figures for the assumptions and the conclusion itself will change. The article is written to be used that way.

Material Misfeed Prevention Is Not an Adaptation of Shipping Error Prevention: Two Decisive Differences

On tomastc.com we have covered verification design on the outbound side in Shipping and Receiving Inspection Systems in 2026: How to Decide Where the Verification Points Go and in Shipping Error Prevention in 2026. The framework set out there does not transfer to the input side as it stands. There are two differences, and both of them reach down into the root of the design.

First, the number of gates is different. Outbound has a single gate called “before it leaves”. Put crudely, if you can make that one check immediately before the truck is loaded absolutely reliable, you can stop the error regardless of what happened upstream. That is why the outbound discussion can concentrate on how reliable to make that single check rather than on where to put the gate at all. On the input side, there are as many gates as there are process steps. One at receiving, one at weighing, one at charging, another one at assembly in the next operation. And each step handles material in a different physical form and a different unit of measure. In a compounding process where liquids, powders, components and secondary materials are all in play, the same verification mechanism will not necessarily work at every step. If you budget with the “one check before it leaves” mental model, the cost of the remaining gates arrives later.

Second, charging material is irreversible. A shipping error can be recovered: if it is noticed at the destination you retrieve the goods and send the correct item. The damage is freight cost and reputational, but the product itself survives. Raw material that has been mixed cannot be separated. A component that has been assembled sometimes has to be destroyed to be removed. In a process involving curing, reaction or heat treatment, the option of going back disappears at the instant of charging. In other words, on the input side a design that says “find it afterwards through verification” is structurally weak. By the time you find it, a batch is already dead.

The conclusion that follows from these two points is clear. Where the optimum on the outbound side is one strong gate at the exit, the optimum on the input side is to create a state in which the error cannot be made in the first place, and to make it impossible to proceed if it is made. The words “verification system” are the same in both cases, but the centre of gravity of the design sits somewhere entirely different.

The Order That Works Is Identification Design, Then Interlock, Then Traceability

Splitting misfeed countermeasures into three layers makes the investment priority visible.

Factories start at Layer 3, but the order that works is 1, then 2, then 3. Records are easy to deploy, they earn credit in audits, and the return on investment is easy to explain. So that is where hands go first. But a record does not remove a single misfeed. Layers 1 and 2 remove them; Layer 3 only affects how far the damage spreads once an event has occurred.

The reason so many plants get stuck at “we verify and we still get misfeeds” is that they have invested in Layers 2 and 3 and left Layer 1 alone. Because Layer 1 costs almost nothing, no owner and no budget line gets attached to it. Reworking the part number scheme looks like a job for the IT department, storage layout looks like a job for the shop floor, and packaging differentiation looks like a job for purchasing. The single most effective move sits in territory that belongs to nobody.

Layer 1 Identification Design Is Simply Making Similar Things Stop Looking Similar

Concretely, it comes down to moves like these. None of them require new equipment.

• Rework the numbering. When part numbers differ only in the last digit, the eye and the hand both get it wrong. The more similar the items, the further toward the front of the code the difference should sit.
• Differentiate appearance. If you have ten kinds of the same white powder, separate them by container colour, container shape and lid colour. Changing the container is more reliable than enlarging the text on the label.
• Fix storage locations and forbid adjacency. Physically separate the combinations that are easy to confuse. Moving them one rack apart is enough to change where the hand reaches.
• Design the packaging unit. If material is pre-portioned into one batch per pack, a quantity error cannot arise. This is the move that removes weighing from the process altogether.

Numbering rules for WIP identification tags and labels are the core of Layer 1, and if they break down, Layers 2 and 3 are unstable from the foundation up. We have covered the operational design in Factory Label Printing Systems in 2026.

Layer 2 Poka-Yoke Interlock Means Moving the Authority to Stop From the Person to the Machine

The essence of Layer 2 is wiring the verification result into the operating conditions of the equipment. If a verification result only turns a screen red, that is a warning, and the decision about whether to stop still sits with the operator. When people are in a hurry, warnings get ignored. A poka-yoke interlock is a mechanism that takes that decision away from the person. On a verification NG, the charging valve does not open, the machine does not start, the next work instruction does not appear on screen. Only at that point does the line reliably stop short of the point of no return.

And as the cost model below shows, Layer 2 is the layer where quotations diverge most. The most common failure in practice is that the discussion ends as a software comparison and the contract is signed while the cost of equipment-side electrical work and PLC I/O is still invisible.

The Four Patterns of Material Misfeed: One Verification Step Does Not Stop All of Them

Treating every mixed-material error as a single category leads to the wrong countermeasure, because the mechanism that stops them differs by pattern.

Pattern 1, wrong item, is the pattern where barcode verification manufacturing controls work most directly. But they only work when the scan actually happens. If even one route remains by which material can be charged without being scanned, errors will leak out through it. The proper approach is to combine this with identification design so that similar items are not sitting within arm’s reach of each other in the first place.

Pattern 2, wrong lot, will never be stopped by verification against the item code alone. What is scanned has to include the lot number and expiry date, and unless the lot is embedded from the point where the WIP tag is issued, no amount of scanning downstream gives the system anything to judge against. In areas such as chemicals and food where lot traceability is required, verification at the point of charging and lot control in inventory need to be looking at the same master data. We cover this design in Chemical Lot Management in 2026.

Pattern 3, wrong quantity or ratio, is not stopped by the verification framework at all. If the correct raw material, from the correct lot, is charged in the wrong amount, every code verification passes. To stop it, the system has to read the value from the scale directly and compare it against the upper and lower limits in the recipe. A substantial share of the plants that say “we installed barcode verification and misfeeds did not fall” are in fact dealing with Pattern 3 as the main cause. That is what happens when the investment target is chosen without separating the patterns first.

Pattern 4, wrong sequence, is harder still. The item, the lot and the quantity are all correct, and only the order of charging is different. In processes where the charging order determines reaction or dispersion, as with catalysts and additives, this translates directly into a quality defect. Worse, the record shows that all required raw materials were charged, so nothing looks abnormal even when you go back through the traceability data. Stopping it requires a sequence interlock that unlocks the next charging instruction only on completion of the previous one.

The choice of reading device changes with the pattern. If you want each item scanned individually and reliably, a handheld unit suits, and we have summarised the cost picture in Barcode Picking With Handheld Terminals in 2026. Where you instead want to confirm several materials at once, or to remove the act of scanning from the process entirely, bulk reading via RFID becomes a candidate, and that investment decision is covered in The Cost of RFID Inventory Management in 2026.

As for productised configurations aimed at compounding processes, packages do exist that manage raw materials with handheld terminals and barcodes and unify recipe verification and work procedure instructions with the production plan; Toshiba Mitsubishi-Electric Industrial Systems offers such a system for misfeed prevention in compounding, covering the receiving, weighing and charging stages and handling both liquids and powders. There are also products that treat picking error prevention and wrong material prevention within the same master verification framework, as Mars Tohken Solution does. Precisely because close-fitting products exist, failing to first establish which pattern your own misfeeds cluster into means signing a contract with Patterns 3 and 4 still untreated.

The Cost of a Material Misfeed Prevention System Splits Into Five Layers

Put several quotations side by side and the amounts diverge sharply. They diverge because vendors differ in how many of the five layers they have included.

Layer 4 diverges the most. How much spare I/O the existing equipment has, whether the control panel can be modified, what the equipment maker’s warranty conditions say: none of this is settled without going to the floor. If you are comparing on the basis of a single software package price, an additional quotation for electrical work turns up after the contract is signed and the original budget envelope is breached.

Layer 5 is the least visible cost. Verification only holds together on the premise that the master data and the recipes are correct. If one recipe is left at an old revision, the system will “correctly” verify against that old recipe and pass an incorrect charge as the right answer. This state is more dangerous than not verifying at all, because operators reason that if the system let it through, it must be correct. Unless the introduction of a recipe verification system comes packaged with an operating routine that protects master data freshness, it becomes a machine for making errors permanent.

Cost Model: Comparing a Verification-Only Plan Against a Plan That Goes as Far as Interlocks at Model Factory M

What follows is an original cost model built for this article. Everything is derived from the shared assumptions, so you can check it with a calculator as you read. The incident counts and the effectiveness rates are not measurements; they are assumed values placed here as a starting point for discussion.

Shared Assumptions (Model Factory M)

Current Misfeeds Run at 51 a Year

The counts for the four patterns are mutually exclusive. A single misfeed is classified into exactly one pattern, and the total does not exceed 51. Every reduction figure below is calculated only by subtracting from this breakdown of 51.

At 51 events against 22,500 batches a year, that is roughly one in 441 batches, a rate of 0.23%. Of those, the 6 that escape as far as the customer work out at one in 3,750 batches. Escapes account for only 11.8% of the count, yet as we will see they account for more than half of the loss in money terms.

Cost per Incident

• One in-process detection = 22,000 baht (18,000 in raw material cost for the scrapped batch plus 4,000 in labour and equipment for recompounding)
• One escape = 220,000 baht (120,000 for rework and sorting in the customer’s process, 60,000 for emergency production and freight of replacement goods, and 40,000 in man-hours for investigation and the corrective action report)

An escape is 10 times the cost of an in-process detection. That factor of 10 governs every investment decision that follows.

Current Annual Loss Is 2,422,500 Baht

• In process, 45 events at 22,000 = 990,000 baht
• Escapes, 6 events at 220,000 = 1,320,000 baht
• Labour cost of the double check = 112,500 baht
• Total 2,422,500 baht per year

The breakdown of the double check labour cost is as follows. Four minutes of mutual confirmation per batch across 22,500 batches gives 4 times 22,500 = 90,000 minutes, or 1,500 hours. The hourly labour rate is set at 75 baht per hour. That is an assumed value: it takes the upper end of the minimum wage, 400 baht (this is a daily amount, not an hourly rate), divides it by 8 hours to get 50 baht per hour, and multiplies by 1.5 to allow for social security, benefits and similar on-costs. So 1,500 times 75 = 112,500 baht.

Note that the minimum wage in Thailand is not uniform; the range published as of 2026 is 337 to 400 baht per day. That is a daily amount, not an hourly rate. Misreading it as hourly inflates every subsequent labour figure by a factor of 8 and throws the cost model out by an order of magnitude. What the source supports is the range itself; which point in the range applies to a given province, district or sector is a separate question. The upper end of 400 baht per day has not been applied uniformly across whole provinces, and there is a history of it being applied only to specific provinces, districts and sectors. Please confirm the amount that applies to your own establishment with the labour office that has jurisdiction over it. For the purposes of this cost model, this article uses the upper end of the range, 400 baht per day.

Dividing the annual loss of 2,422,500 baht by 22,500 batches means that roughly 108 baht per batch is being consumed by misfeeds.

Ranking by Loss Value Reshuffles the Ranking by Count

(This table excludes the 112,500 baht of double check labour cost. It is the breakdown of losses arising from incident counts only. 990,000 plus 1,320,000 gives 2,310,000 baht, which matches.)

Pattern 2, wrong lot, is half the count of Pattern 1, yet it comes close to Pattern 1 in loss value, because it has the highest number of escapes at 3. Ordering countermeasures by descending incident count pushes Pattern 2 down the priority list. And Patterns 3 and 4 together, 15 events, account for 528,000 baht of loss, 22.9% of the total. As we will see next, verification alone does not reduce that 22.9% by a single baht.

Scenario 1, Verification Only

Initial investment

Annual running cost = software maintenance (15% of 850,000) 127,500 plus label consumables 60,000 = 187,500 baht per year

Effectiveness assumptions (assumed values used in this article, not measurements)

Effective on Patterns 1 and 2 at 90%, ineffective on Patterns 3 and 4, because a code verification passes for both a quantity error and a sequence error. And since the double check remains in place, the labour cost saving is taken as zero. In practice, a routine in which a person confirms that verification passed cannot be abolished immediately after go-live.

• Patterns 1 and 2, in process, 22 plus 9 = 31 events, reduced by 27.9 events (3.1 remaining)
• Patterns 1 and 2, escapes, 2 plus 3 = 5 events, reduced by 4.5 events
• Escapes stand at 6 minus 4.5 = 1.5 remaining per year
• Saving = 27.9 times 22,000 plus 4.5 times 220,000 = 613,800 plus 990,000 = 1,603,800 baht per year
• Annual net benefit = 1,603,800 minus 187,500 = 1,416,300 baht per year
• Simple payback = 1,530,000 divided by 1,416,300 = 1.08 years

It is worth confirming the loss that remains. The remainder of Patterns 1 and 2 is 3.1 times 22,000 plus 0.5 times 220,000 = 178,200 baht; Patterns 3 and 4 are untouched at 14 times 22,000 plus 1 times 220,000 = 528,000 baht; and the double check labour cost is 112,500 baht. That totals 818,700 baht, which matches the current 2,422,500 baht less the 1,603,800 baht saving.

Scenario 2, Verification Plus Scale Integration Plus Interlock

Initial investment

Annual running cost = software maintenance (15% of 1,250,000) 187,500 plus label consumables 60,000 plus periodic verification of the interlocks 120,000 = 367,500 baht per year

The reason 120,000 baht a year is carried for periodic verification is to meet the IATF 16949 requirement discussed later. A poka-yoke device is not something you install and forget; it is subject to periodic confirmation that it is still functioning, and the frequency of that confirmation belongs in the control plan. Vendors who leave this line item out of their quotation are not unusual, but leaving it out invites a finding at audit.

Effectiveness assumptions (again, assumed values)

98% on Patterns 1 and 2, 90% on Pattern 3, 95% on Pattern 4.

• Patterns 1 and 2, in process, 31 events, reduced by 30.38 events; Patterns 1 and 2, escapes, 5 events, reduced by 4.9 events
• Pattern 3, in process, 9 events, reduced by 8.1 events; Pattern 3, escapes, 1 event, reduced by 0.9 events
• Pattern 4, in process, 5 events, reduced by 4.75 events; Pattern 4 escapes were 0 to begin with
• In-process reduction = 30.38 plus 8.1 plus 4.75 = 43.23 events, times 22,000 = 951,060 baht
• Escape reduction = 4.9 plus 0.9 = 5.8 events, times 220,000 = 1,276,000 baht
• Escapes stand at 6 minus 5.8 = 0.2 remaining per year
• Double check reduction = 112,500 times 70% = 78,750 baht (witnessing is retained, so it is not abolished entirely)
• Total saving = 951,060 plus 1,276,000 plus 78,750 = 2,305,810 baht per year
• Annual net benefit = 2,305,810 minus 367,500 = 1,938,310 baht per year
• Simple payback = 3,140,000 divided by 1,938,310 = 1.62 years

One instance of double counting is being avoided here. The 78,750 baht saving in double check labour is credited to Scenario 2 only. In Scenario 1 the premise is that human confirmation remains even after verification is deployed, so the saving is taken as zero. Loading the same benefit onto both scenarios would overstate Scenario 1’s net benefit and destroy the basis for comparison.

Here too it is worth confirming the residual loss. The remainder in process is 45 minus 43.23 = 1.77 events, or 38,940 baht; escapes at 0.2 events are 44,000 baht; and the remaining 30% of the double check is 33,750 baht. That totals 116,690 baht, which matches 2,422,500 minus 2,305,810.

The Crux: Choose on Payback and You Get Scenario 1, Choose on the Five-Year Total and You Get Scenario 2

Setting the two plans side by side.

• The investment is 3,140,000 divided by 1,530,000 = 2.05 times larger
• The annual net benefit is 1,938,310 divided by 1,416,300 = 1.37 times larger

Because the investment grows 2.05 times while the benefit grows only 1.37 times, choosing on payback period alone makes Scenario 1 the faster option (1.08 years against 1.62 years). If the capital request form used by your parent company or regional HQ has only a payback period field, Scenario 1 is selected at this point.

But one decisive difference remains. Scenario 1 leaves 1.5 escapes a year and Scenario 2 leaves 0.2. That difference does not appear anywhere in the payback field. And an escape means incorrect material arriving in a customer’s process.

Pricing the Residual Risk Widens the Five-Year Gap to Roughly Double

What follows is a separate calculation with one additional assumption. Please read it without mixing it into the figures above.

We assume a 3% probability that an escape develops into a product recall. This is an assumption made in this article, not a statistic. We also provisionally set the cost of one recall at 5,000,000 baht. Marsh Japan notes that the average damages from a product recall exceed USD 1.5 million per incident, which at roughly 32 baht to the dollar corresponds to something on the order of 48 million baht. The 5,000,000 baht used here is around one tenth of that, deliberately set low. This too is an assumption made in this article.

In Japan, voluntary recalls of food and similar products have been subject to mandatory notification to the authorities since June 2021, and the Consumer Affairs Agency recall information site carries more than 700 filings a year. The assumption placed here is that recalls are not exceptional events but something that occurs in reality at a certain frequency.

• Scenario 1: 1.5 remaining events times 3% = 0.045 events per year, times 5,000,000 = 225,000 baht per year of expected loss
• Scenario 2: 0.2 remaining events times 3% = 0.006 events per year, times 5,000,000 = 30,000 baht per year

Deducting these from the net benefit.

One caveat about how this deduction is framed. The table above subtracts the residual risk each scenario carries from its benefit as an absolute value, while carrying no recall risk on the current state that serves as the basis for comparison. Apply the same assumption to the current state and it too carries an expected loss of 6 escapes times 3% times 5,000,000 = 900,000 baht per year. In other words, this table is a conservative framing that deliberately understates the benefit of both plans. Level the treatment by placing 900,000 baht on the current state as well, and the annual net benefit after deduction becomes 2,091,300 baht for Scenario 1 and 2,808,310 baht for Scenario 2, with paybacks of 0.73 years and 1.12 years respectively, both shorter than before the deduction. Since the whole point of the investment is to reduce risk, moving in that direction is arguably the more natural result.

What matters is that the conclusion drawn from the five-year totals below, a gap of 1,975,050 baht, is the same under either framing. (With 900,000 baht placed on the current state as well, the five-year totals become 8,926,500 baht for Scenario 1 and 10,901,550 baht for Scenario 2, and the gap is the same 1,975,050 baht.) The rest of this section proceeds on the conservative side, that is, on the figures in the table above.

• Scenario 1 = 1,191,300 times 5 minus 1,530,000 = 4,426,500 baht
• Scenario 2 = 1,908,310 times 5 minus 3,140,000 = 6,401,550 baht
• Gap = 1,975,050 baht (Scenario 2 is about 1.98 million baht better over five years)

The easy misreading here is to think the ranking flipped because the expected loss was deducted. Scenario 2 was already ahead on the five-year total before any deduction. Calculated before deduction, Scenario 1 gives 1,416,300 times 5 minus 1,530,000 = 5,551,500 baht and Scenario 2 gives 1,938,310 times 5 minus 3,140,000 = 6,551,550 baht, a gap of 1,000,050 baht. What the deduction changes is not the ranking but the size of the gap, which widens from 1,000,050 baht to 1,975,050 baht, roughly doubling Scenario 2’s advantage.

So the conclusion is this. Choose on payback period and you get Scenario 1; choose on the five-year total and you get Scenario 2. That contrast holds whether or not the deduction is applied. Which one is right depends on the time horizon over which the plant evaluates capital investment. A form that looks only at payback period mechanically favours the plan with the smaller initial outlay. We would not say that is wrong, but “pays back faster” and “leaves you better off over five years” are different questions, and that is worth sharing before the request goes up for approval.

The Practical Answer Is Phased Deployment

It is also possible to dismantle the premise that you must pick one of the two plans. The alternative is to stop Patterns 1 and 2 first with Scenario 1, then add interlocks only on the processes where Patterns 3 and 4 actually remain.

Looking at the increment makes the structure clear. The incremental investment in moving from Scenario 1 to Scenario 2 is 3,140,000 minus 1,530,000 = 1,610,000 baht, and the incremental annual net benefit is 1,938,310 minus 1,416,300 = 522,010 baht. Payback on the increment alone is 1,610,000 divided by 522,010 = about 3.08 years. On the assumption where expected loss is deducted, the incremental net benefit becomes 1,908,310 minus 1,191,300 = 717,010 baht and the incremental payback shortens to 1,610,000 divided by 717,010 = about 2.25 years.

In other words, the second stage on its own is a project with a payback of around three years. If the group investment threshold is three years it passes; if it is two years it does not. The advantage of phased deployment is that this judgement can be made a year later on real data. Run Scenario 1 for a year and the records will show how many Pattern 3 and Pattern 4 events actually remain. The assumed 9 and 5 might turn out to be 3, or they might turn out to be 20. Deciding the second stage on those real counts is more reliable than settling everything up front.

Issues Specific to Designing Material Misfeed Prevention in a Thai Factory

On a Multilingual Floor, Identification by Text Can Fail to Work

On manufacturing floors in Thailand it is not unusual for operators whose first language is Burmese or Khmer to be working the same line as Thai speakers. This bears directly on misfeed prevention design, because identification by written text depends on the language of the reader.

You can post “Caution, first in first out” in large Thai characters, and to an operator who cannot read that script it is indistinguishable from a pattern. Raw material names written in Japanese, abbreviations in English, cautions in Thai: with three languages mixed on the floor, every posting is only partially readable to everyone. And postings that cannot be read get skipped.

Layer 1 identification design should therefore give priority to means that do not depend on language.

• Colour. Align container, label and storage location colour by material family. Without reading a word, taking a red container from a blue rack feels wrong.
• Shape. Vary container geometry or lid form. It registers the moment a hand touches it.
• Codes. Have a machine judge the result of reading a barcode or 2D code. That takes the operator’s language ability out of the judgement.
• Photographs. Replace text in work instructions with photographs. One image shows “from this container, into this charging port”.
• Physical shape mismatch. Build the geometry so the wrong container will not fit the charging port. This is the strongest move, but it involves equipment modification.

This point is largely absent from articles on misfeed prevention written for the domestic Japanese market, because on a Japanese floor the ability to read Japanese signage is an unspoken given. Apply the same assumption in Thailand and you get a phenomenon where the more signage you add, the more diluted the effect becomes. Before adding signage, it is worth examining how many postings could be replaced by colour, shape and codes. The cost is close to nothing, and it works even before Layer 2 interlocks are in place.

One further point: recording training is also harder on a multilingual floor. You need a way to confirm that proof of “having taught” corresponds to the trainee’s actual understanding. Here a hands-on practical check with real material is more reliable than a written test.

For Food, Thai MOPH Notification No. 420 (GMP) Is the Baseline

In the food sector, a misfeed is not merely a yield problem; it is a regulatory one. In Thailand, Ministry of Public Health Notification No. 420 sets standards for food production methods, production equipment and food storage, and when exporting food from Japan, proof that the manufacturing facility meets or exceeds that standard is required in the form of a GMP certificate. Certificates such as ISO 22000, FSSC 22000 and JFS-B/C are accepted as supporting documentation (MAFF, JETRO).

What matters from a misfeed prevention standpoint is that these standards require control at each stage of raw material identification, receiving, storage and use. That means Layer 1 identification design and Layer 3 traceability are not optional quality improvement measures but mandatory requirements for maintaining certification. There are areas, in other words, where the decision criterion of “we will invest if it produces a return” does not apply in the first place.

Which certification to obtain and for what scope, and whether an existing certification satisfies the requirements of the destination market, differ by product and by destination. We recommend confirming these individually with the competent authority or a specialist.

For Automotive Parts, IATF 16949 Clause 10.2.4 Governs How Poka-Yoke Devices Are Operated

In the automotive parts sector, clause 10.2.4 of IATF 16949 deals with poka-yoke, or error proofing. What it requires is not “install poka-yoke” but having a documented process for determining poka-yoke methods. Methods are to be selected on the basis of risk analysis such as PFMEA, and the frequency of their verification is to be included in the control plan. Beyond that, the poka-yoke devices themselves become subject to preventive maintenance.

There are three practical implications.

1. What method was applied to which pattern has to be documented, together with the rationale. The four-pattern split used in this article can be used directly as the unit of analysis in a PFMEA.
2. The devices have to be periodically confirmed to be functioning. You set a verification frequency and check whether an interlock has degraded into a state where an NG still passes. This is why 120,000 baht a year of periodic verification is carried in the Scenario 2 running cost.
3. A response to poka-yoke device failure has to be defined in advance. How is control maintained by other means while the device is down? Leave this undecided and the floor acquires an incentive to keep running without reporting the failure.

The cost of guaranteeing that a device keeps functioning is overlooked more often than the cost of installing it. When comparing quotations, check that both the Layer 4 amount and the Layer 5 running cost are present.

Read the Conditions on BOI Incentives Precisely

The Thailand Board of Investment (BOI) has a Smart and Sustainable Industry measure covering equipment upgrades and automation. Under this measure, corporate income tax exemption is available against qualifying equipment upgrade expenditure, but the exemption is capped at 50% of qualifying expenditure as the basic rule.

The cap rising to 100% is not unconditional. It applies only where automation or robotics are introduced into the production line and at least 30% of the value of the upgraded machinery is sourced from Thailand’s domestic automation industry. Misreading this as “automation investment gets 100% exemption” throws out the entire premise of the investment plan. In the terms of this article’s cost model, it changes how the incentive bears on Scenario 2’s initial investment of 3,140,000 baht, which moves the payback discussion itself.

For reference, the same measure was reported to have attracted 132 applications worth roughly 17.2 billion baht in the first half of 2026 (Asia News Network). The scheme is operating and is being used in practice.

That said, please confirm eligibility individually in every case. Application categories, the definition of qualifying equipment and the test for what counts as “sourced from Thailand’s domestic automation industry” vary case by case. How much of a misfeed prevention system investment falls within qualifying equipment also depends on the configuration. BOI incentives and tax treatment presuppose confirmation with the competent authority or a specialist.

Confirm Every Time That the Labour Cost Assumption Is a Daily Rate

As noted above, the published minimum wage range in Thailand is 337 to 400 baht per day. It is a daily amount. This article takes the upper end of that range, 400 baht per day, as the basis for the cost model, divides by 8 hours to get 50 baht per hour, and applies a factor of 1.5 to arrive at the 75 baht per hour used as the hourly labour rate.

There are two things to watch. The first is scope of application. The range itself is published, but the upper figure does not necessarily apply as-is to your establishment. The applicable amount differs by province, district and sector, and the upper end has historically been applied to specific regions and sectors only. Before putting it into a cost model, please confirm individually with the labour office having jurisdiction, or with a specialist.

The second is the unit. Get this wrong and the model breaks. Read 400 baht as an hourly rate and the hourly labour cost becomes 600 baht per hour, inflating the double check labour cost by a factor of 8 and producing a completely different composition of the current loss. When reviewing a cost model relating to labour in Thailand, the first thing to check is whether the figure is a daily rate or an hourly rate. This is the most common error in models brought in from Japan.

Master Data and Recipe Freshness Directly Determine What Verification Is Worth

Finally, back to Layer 5 and operations.

A verification system never doubts that its master data is correct. If a recipe has not been updated, the system will judge a charge made to the old recipe as normal. And at that moment, the operator does not doubt it either. The system showed green, so it must be right, which is the natural inference. Under a paper routine there was room for an operator to notice that something was different from last time and speak up; introduce verification and that room disappears.

So the introduction of verification has to come packaged with a change management mechanism.

• When a design change or recipe change occurs, does production stop until the master data update is complete, or does it run on the old recipe? Decide this rule in advance.
• Who confirms that the change has been reflected? Separate the person who updates from the person who confirms.
• Record the switchover timing at lot level. If “from when is the new recipe in effect” is vague, you cannot narrow the scope when a problem surfaces.
• Include raw material changes in scope. Where the supplier changes under the same part number, verification passes the material as identical.

This is 4M (man, machine, material, method) change management itself. When designing a system for wrong material prevention, place the change management mechanism inside the same project. We cover the design and operation in 4M Change Management Systems in 2026. Master data preparation is carried at 220,000 baht in Scenario 1 and 350,000 baht in Scenario 2 precisely so that this work is not treated as something the floor does after go-live. Pushed onto the floor, it is invariably deferred.

Frequently Asked Questions

What is a material misfeed prevention system, and is barcode verification enough on its own?

It is the general term for a mechanism in which, at the moment raw material or components are charged, a machine judges whether what is about to be charged matches the instruction, and makes it impossible to proceed if it does not. Barcode verification is one part of it, and verification alone stops only the wrong item and wrong lot patterns. A wrong quantity or ratio cannot be judged without integration between the scale and the system, and a wrong sequence is not stopped without a sequence interlock. At the model factory in this article, wrong quantity and wrong sequence account for 15 events a year and 528,000 baht of loss. The first step is to count, from your own historical records, which of the four patterns your misfeeds cluster into.

How much does material misfeed prevention cost?

Under the assumptions used in this article, a verification-only configuration is set at 1,530,000 baht initial and 187,500 baht annual running cost, and a configuration adding scale integration and equipment-side interlocks to verification is set at 3,140,000 baht initial and 367,500 baht annual running cost. These are not survey-based market rates; they are assumed values that make the model work. In real quotations, the layer that diverges most is the equipment-side interlock layer (PLC I/O and electrical work), which varies greatly with the condition of existing equipment. Comparing on a single software package price makes a quotation that omits this layer look cheap. If you want a net figure that incorporates BOI incentives or tax treatment, note that application categories and the definition of qualifying equipment differ case by case, so individual confirmation with the competent authority or a specialist is a prerequisite. The same applies to GMP requirements in the food sector.

Can a shipping error prevention setup be used as-is in charging processes?

It cannot. There are two reasons. Outbound has a single gate called “before it leaves”, whereas on the input side there are as many gates as there are process steps. And charging is irreversible: once material has been mixed or components assembled there is no going back, so a design that says “find it afterwards through verification” is structurally weak. Outbound design is covered in Shipping and Receiving Inspection Systems in 2026, but on the input side you additionally need cost for as many gates as there are process steps, and a mechanism that reliably stops the line short of the point of no return.

What is the difference between poka-yoke and an interlock?

Poka-yoke is the broad concept covering any mechanism that prevents an error or detects it immediately, including making misassembly physically impossible through jig geometry and detecting an error with a sensor and raising a warning. An interlock is the subset in which the judgement result is wired directly into the operating conditions of the equipment, so that the equipment does not move unless the condition is satisfied. Where the difference bites in practice is that with a warning-only mechanism, the decision about whether to stop still sits with the operator. When people are in a hurry, warnings get ignored. For automotive parts, IATF 16949 requires documentation of the process for determining poka-yoke methods and inclusion of verification frequency in the control plan.

In a factory with multilingual operators, where should we start?

Start with identification that does not depend on text. In Thai factories it is not unusual for Thai, Burmese and Khmer to be mixed on one floor, and because signage in text depends on the language of the reader, it can fail to work. Colour coding, differentiation of container geometry, procedure display by photograph, and code verification in which a machine judges the reading result: these four cost almost nothing and work even before any system investment. Countermeasures that go in the direction of adding more signage tend to plateau on a multilingual floor.

How long does deployment take?

It depends on the configuration, but what determines the schedule is not the software rollout; it is getting master data and recipes in order. Even at the scale of the model factory here, 180 raw material SKUs and 60 recipes, taking stock of current recipes, reworking the part number scheme and fixing the label numbering rules all take a meaningful amount of time. Where equipment-side interlocks are included, you also need a survey of the existing control panels and a plan for the electrical work, plus confirmation of the equipment maker’s warranty conditions. With phased deployment, a realistic approach is to run verification alone for a year, confirm from the records how many wrong quantity and wrong sequence events actually remain, and set the scope of the interlocks from those real counts.

Summary

What works in material misfeed prevention is not the speed of verification. It is removing the conditions that allow the error in the first place. The order is Layer 1 identification design, then Layer 2 interlock, then Layer 3 traceability. Because most plants invest in Layer 3 and in verification speed while leaving Layer 1 alone, they end up stuck at “we verify and we still get misfeeds”.

Misfeeds also split into four patterns, and the mechanism that stops each one is different. At the model factory in this article, of the 51 events a year, the 36 in Patterns 1 and 2 are stopped by verification alone and the 15 in Patterns 3 and 4 are not. In money terms, Patterns 3 and 4 are 528,000 baht, 22.9% of the 2,310,000 baht of loss arising from incident counts (the figure excluding the 112,500 baht of double check labour). Before deploying verification, count which pattern your own misfeeds cluster into. Invest without doing that and Patterns 3 and 4 remain in full.

The cost model has two conclusions. Choosing on payback period favours the verification-only Scenario 1 (1.08 years); choosing on the five-year total favours Scenario 2 with interlocks (6,401,550 baht against 4,426,500 baht, a gap of 1,975,050 baht). The investment grows 2.05 times while the annual net benefit grows only 1.37 times, which is why Scenario 2 will never pass under a capital request form that shows only the payback field. Yet Scenario 1 leaves 1.5 escapes a year and Scenario 2 leaves 0.2. That difference does not appear in the payback field.

In practice this points to phased deployment. Stop Patterns 1 and 2 with Scenario 1, run it for a year, confirm from the records how many Pattern 3 and Pattern 4 events actually remain, and add interlocks only on the processes where they do. Viewed as an increment alone it is a project with a payback of about 3.08 years (about 2.25 years on the assumption where expected loss is deducted), so there is no penalty in waiting for real data before deciding.

Do not lose sight of the Thailand-specific issues either. On a multilingual floor, identification by text can fail to work, and replacement by colour, shape and codes works first. The minimum wage of 337 to 400 baht that underpins the labour assumption is a daily rate, and where in that range your own site falls needs confirmation with the authority having jurisdiction. Under the BOI Smart and Sustainable Industry measure, corporate income tax exemption is capped at 50% of qualifying expenditure as the basic rule, and 100% applies only where automation or robotics are introduced into the production line and at least 30% of the value of the upgraded machinery is sourced from Thailand’s domestic automation industry.

There is one thing to do tomorrow. Open the misfeed records for the past year and sort them into four columns: wrong item, wrong lot, wrong quantity or ratio, and wrong sequence. If many records cannot be sorted, that in itself is the first problem to solve.

Separating the patterns is work that can proceed well before any system study begins. At TOMAS TECH we take enquiries about production control and WIP identification from manufacturers with sites in Thailand from the earliest stage of consideration. It is entirely fine to come to us at the point where you have not yet decided which processes need interlocks, or where you would simply like help sorting your own records into the four patterns. We are happy to start by working through what your existing records can tell you, so please feel free to get in touch through our contact form.

References

• Toshiba Mitsubishi-Electric Industrial Systems, misfeed prevention system for compounding processes
• Mars Tohken Solution, master verification by barcode
• Consumer Affairs Agency of Japan, recall information site
• MAFF, hygiene standards required of facilities producing food for export to Thailand
• JETRO, food import regulations in Thailand, August 2025
• Commentary on IATF 16949 clause 10.2.4, poka-yoke
• Pretesh Biswas, IATF 16949:2016 Clause 10.2.4 Error Proofing
• Thai Law Online, Minimum wage in Thailand, 337 to 400 baht per day
• Marsh Japan, product recall risk
• Alvarez & Marsal, Thailand’s Renewed BOI Incentives 2026-2027
• Asia News Network, Thailand 1H 2026 investment