Work Instruction Systems 2026 – Retiring the Old Revision Is the Hard Part

When work instructions still move around the plant as paper and Excel files, it is a natural reflex to assume that handing out tablets will settle the matter. But what decides the outcome of a work instruction system is not how quickly a new revision reaches the floor. It is whether the superseded revision reliably disappears from the floor afterwards. This article takes a Japanese-owned parts maker in Chonburi, Thailand as its model and follows a single ledger from what today’s paper operation already costs per year, through the investment required by three implementation configurations, to payback in years, and finally down to the items you should settle on paper before you order anything.

What a work instruction system is – where it ends and process management or electronic forms begin

A work instruction system is the mechanism that authors, revises and approves the master of a work instruction or SOP electronically, delivers it to shop floor terminals and boards, retires the superseded revision, and leaves a record of which revision the work was actually done against. It is not a viewer for SOPs. If a viewer were enough, a shared folder and a PDF reader would already do the job. The moment you decide that is not enough, the thing you are evaluating has quietly become a system for controlling revisions.

This space is crowded with adjacent systems, and internal discussions tend to stall on the question of which one to buy. So it is worth drawing the boundaries first.

The boundary with process management fits in one sentence. A process management system exists to watch progress, and a work instruction system exists to deliver instructions. The former collects output records so you know how far the line has got; the latter hands the floor what to do right now and how. Running both on the same tablet is technically possible, but the investment cases should be built separately. Visualising process progress and designing output collection are covered in the cost and selection of a process management system.

The boundary with 4M change management works the same way. 4M change management is the side that decides the change; the work instruction system is the side that delivers the decided change to the floor. Raising a change request, reviewing it, approving it and logging the change point all belong to 4M change management. Turning what was approved into an instruction, stopping the superseded revision, and confirming that operators have read the new one is where the work instruction system begins. For the design of the request and approval flow and the change point register, see implementing a 4M change management system.

An electronic forms system deals with the paper operators write on – check sheets and inspection records. A work instruction system deals with the paper operators read. In practice these are consecutive actions on the same tablet, so they get conflated, but the required functions differ. On the forms side the centre of gravity is input validation and record integrity; on the instruction side it is revision control and retirement. The cost of digitising forms and how to choose between product types is set out in the guide to implementing an electronic forms system. Identifying defect causes and designing quality data belong to quality data management systems, and the division of roles between ERP and MES belongs to the cost and approach of an MES implementation. This article stays out of all of them and follows one thing only – delivering the instruction and making the superseded revision go away.

With paper work instructions, the real problem is collection, not distribution

Ask what is wrong with paper work instructions and most plants answer that swapping them out never keeps up. At that point the discussion turns towards faster ways to distribute. But what actually produces defects and customer complaints is not slow distribution. It is the superseded revision staying on the floor after the new one has been handed out.

In a paper operation you revise the master, print the new revision and walk the processes swapping copies. What gets swapped is only what the person doing the walk knows about. In the places they do not know about, the superseded revision stays. The residual paths are always the same ones. The spare copy in the workbench drawer. The personal file a team leader keeps. The laminated sheet still stuck on the wall. Somebody’s private photocopy. And the desk of the person who happened to be off on the day of the swap. None of it is malicious; if anything, it is what a floor trying to be conscientious produces.

Collection is hard because it means proving a negative. You can evidence that the new revision was distributed with a distribution log, but you cannot say the superseded revision is nowhere without, in principle, walking the entire site. At a plant running 40 processes, 120 part numbers and 480 work instructions, walking all of it at every revision is impossible. So in a paper operation, residual superseded revisions should be costed not as something that might happen, but as something that will happen at a certain rate.

There is one misunderstanding worth heading off here. It is the idea of running paper and tablets side by side and leaving both in place during a transition period. That is the most dangerous configuration of all, because it institutionalises the state in which a superseded revision is allowed to remain. There is empirical evidence bearing on this point. In the study Letmathe and Rößler published in the Journal of Operations Management in 2022, the experiment was run not in a laboratory but in a university learning factory, and it reported that digitally animated, interactive work instructions made new tasks faster to learn and improved performance compared with paper technical drawings. Just as important, it also found that combining digital with paper produced no advantage over digital alone. The readable implication is that running both is not an insurance policy.

Vendor material does circulate figures built on this study, of the form “errors fall by so many percent” or “work is so many percent faster”, but those cannot be confirmed from the paper’s abstract. This article does not cite them. Copying another company’s reduction rates straight into a capital request leaves you unable to answer where they came from when the questions start.

The four layers of work instruction – author, translate, deliver, prove

Lay a work instruction mechanism out by feature and it turns into a product comparison, and the decision goes nowhere. Split it into four layers by what happens when each layer breaks, and the question of which layer is weak at your own plant gets settled first.

The point of the author layer is that who holds the authority to fix a revision is unambiguous. In a paper operation it is not unusual for the master Word file to sit in a shared folder that both production engineering and quality assurance can edit. In that state, a revision that has not yet been stamped for approval gets printed and reaches the floor. Digitising changes nothing in this layer if you only move where the file lives without putting approval on a workflow.

The translate layer barely appears in product descriptions written for the Japanese domestic market. In a multilingual plant it is the biggest point of failure. The next section deals with it in detail.

The essence of the deliver layer is not distribution but retirement. Every product can push out a new revision. What separates them is whether the superseded revision can be made impossible to open, and whether you can produce a list of who has acknowledged receipt of the new one. Putting PDFs in a shared folder gives you distribution but not retirement. Overwrite the file and the copies already pulled down to local devices go on living.

The prove layer writes the revision number into the output records and inspection records. Without it, when a defect surfaces later you cannot reconstruct which revision was in front of the operator at that process on that day. Put the other way round, whether you need this layer is decided by customer or standards requirements, not by internal efficiency. In the payback discussion further down, how you treat this layer is what splits the conclusion.

In Thai plants, the revisions that drift are the language versions

This is the centre of the article. In the Japanese domestic discussion, the unit of revision control is a single instruction. Move revision 3 to revision 4 and that instruction is now revision 4. In a Thai plant, the unit of revision control becomes language multiplied by revision.

The model plant has 180 operators, made up of 120 Thai speakers, 45 Burmese speakers and 15 Japanese and English speakers. Instructions are held in 3 languages. So what has to be managed is not 480 documents but 480 documents times 3 languages. When 420 revisions occur in a year, what has to be updated is 420 revisions times 3 languages.

The problem is that the multiplication does not run evenly. When the master is revised, the version updated first in practice is Thai. The majority of operators read Thai, more people in the building can read it, and Thai staff in quality assurance and production engineering can correct it themselves. The Japanese and English versions come next, because the master is usually in Japanese to begin with. The one left until last is Burmese. Few people internally can translate it, outsourcing introduces lead time, and with 45 people affected the priority is relatively lower.

The result is predictable. For the same task in the same process, the Thai version is revision 4 while the Burmese version is still revision 3, and that becomes the normal state. Worse, in a paper operation there is no way to detect it. Opening the Thai file tells you nothing about whether the Burmese version has caught up. Nobody has lied and nobody has cut a corner, and yet two different procedures exist on the floor at the same time.

This is the failure point that does not appear in the Japanese domestic discussion. In a plant in Japan, “the superseded revision is still out there” means physical paper. In a plant in Thailand you get that plus a language version that has not been updated, a form of residue that digitising does not automatically solve. A configuration that merely replaces the shared folder with a tablet leaves this layer completely untouched.

The selection criteria therefore have nothing to do with how pleasant the viewer is or how many terminals you buy. They come down to three questions.

• Can a new revision retire the old one. Not just distribute the new one, but make the old one impossible to open
• Can the system detect whether language versions have followed the master revision. Can it link master and language versions as parent and child and list the ones that have not been updated
• Can the revision that was actually on screen be stamped onto output and inspection records

Of these, the second is the one most often missed at sites in Thailand and across ASEAN. Product demos make a set piece out of switching the display language, but being able to switch languages and being able to detect whether they have kept up are entirely different functions. The question to put in the demo is not “can it switch between Thai and Burmese”, but “revise the master now, and then show me the screen that lists only the language versions that have not been updated”.

There is one more issue specific to multilingual plants – translation granularity. If your operating rule is to retranslate the whole document, then even a small numeric change in the master triggers a full retranslation, lead time stretches, and the Burmese version falls further behind. Translate only the changed paragraphs as a diff, and translation consistency degrades. What works in practice is to hold the instruction split into process steps and send only the changed steps for translation. Whether that structure is available depends on the product’s internals, so it needs to be confirmed during selection.

Assumptions – the model plant’s production conditions and the size of its instruction set

Before getting to cost, fix the assumptions. Every value below, unless otherwise noted, is an assumption. These are not figures backed by statistics or public sources; they are a starting point so that you can substitute your own values and walk the same steps. Apart from the sensitivity section, where they are deliberately broken, this article does not change any value in this table.

Some readers will find 420 revisions implausible. At a site with 40 processes and 120 part numbers, 420 per year works out to just under 1.7 revisions on every one of 250 operating days. That looks like a lot, but the breakdown is procedure corrections following 4M changes, working conditions added as defect countermeasures, and new documents created for new part introductions. At a contract-heavy site with a wide part mix, this level is unremarkable. If your own number is lower, the sensitivity section puts a figure on what that does.

Operator turnover of 54 people is likewise nothing unusual for manufacturing in Thailand. Turnover bites in two places – the ramp-up time for new operators, and the risk of superseded revisions surviving. A large part of how instructions are operated is maintained by word of mouth, and every time people change, the tacit understanding that you do not use the paper in the drawer wears a little thinner.

Counting what today’s paper operation already costs, in four items

The starting point of an investment decision is not the cost of the new mechanism but what you are already paying for the current one. A paper operation never issues an invoice, so it is mistaken for free. Count it in 4 items.

Figures are in baht per year. The total is 341,250 + 392,000 + 205,200 + 32,000 = 970,450 baht.

Look at the breakdown of the revision effort. Of the 195 minutes per revision, 90 minutes – nearly half – is translation into 2 languages. Revising the master itself takes 60 minutes, and approval, printing and swapping take 45 minutes. One thing follows from this. At a multilingual site, the cost of revising an instruction is set not by the time spent thinking about the procedure but by the time spent propagating it sideways. Revisions are heavier than at a comparable plant in Japan not because the people are less capable but because the structure is different.

The 14 cases a year at 28,000 baht each for rework from escaped superseded revisions is an assumption that bundles repair labour, sorting and writing the report for the customer. At 392,000 baht this single item is the largest of the four. It is also, in most companies, not aggregated anywhere as a cost. Repair shows up as production overtime, sorting as routine quality assurance work, and the customer report as something sales handled, each dissolved into a different department’s daily life. The main reason capital requests for work instruction systems fail is that the largest item is on nobody’s books.

New operator ramp-up at 54 people × 40 hours × 95 baht = 205,200 baht is calculated at the direct labour rate. The 40 hours is an assumption standing in for the ramp-up until someone can work independently while following the instruction. Audit response at 4 times per year × 32 hours × 250 baht = 32,000 baht is the smallest of the four. The amount is small, but as the next sections show, this item can be the reason you choose configuration C. Size in money and weight in a decision are not the same thing.

Implementation splits into three configurations – delivery only, revision control, records integration

Digitising work instructions is far less varied than the product catalogues suggest. Seen through the investment figure, it splits into three configurations.

Figures are in baht. The first thing to look at in this table is the share taken by the software itself. It is 16.4% in A, 24.2% in B and 25.9% in C. In every configuration, more than seven tenths of what you pay goes somewhere other than the software. Comparing licence unit prices across competing quotations will not tell you how the totals rank. Terminals and mounts, the wireless environment, migrating the existing instructions, and the multilingual linkage design – that is where the money moves.

Note in particular that migrating the existing 480 instructions splits into 150,000 baht for A and 480,000 baht for B. Migration in A means importing the existing Word and PDF files as they are and lining them up. Migration in B means linking master and language versions as parent and child, deciding the revision numbering scheme, and registering the content split into process steps. The same word, very different work. If a quotation says “data migration, one lot”, always ask which of the two it means.

The next table maps the three configurations onto the four layers. When this mapping slips, internal discussions stop meeting each other.

What this table says is unambiguous. A covers only part of one of the four layers, the deliver layer. Distribution gets faster, but approval in the author layer stays a matter of operating rules, the translate layer stays fully manual, and the prove layer does not exist. That column is where the original instinct – hand out tablets and the problem is solved – actually takes you.

The difference between B and C is one line, the prove layer. Author, translate and deliver are identical, and C adds the ability to write the revision number into the output records system and the inspection records. For that one line, the investment rises from 2,560,000 baht to 3,400,000 baht, an increase of 840,000 baht. Whether that increment is worth it is the subject of the next section.

Payback – delivery alone does not pay back

Against the investment figures above, set the amounts that can be reduced. The reduction rates are assumptions, but they are tied to which of the four layers each configuration covers. Costs attached to layers a configuration does not cover do not fall.

Set that alongside the investment.

Three conclusions follow, in order.

First, A, which is handing out PDFs on tablets, does not pay back. A payback of 16.58 years is far beyond the useful life of the terminals. The reason is visible if you read the reduction table across. In A, rework from escaped superseded revisions falls by only 15%. Distribution gets faster, but there is no mechanism that makes the superseded revision go away. The largest cost item, 392,000 baht, survives almost untouched. The claim at the top of this article – that the point is not being able to distribute but making the superseded revision disappear – shows up as a number on that one line.

Second, B is the shortest at 9.68 years, and what creates the difference is the reduction in escaped superseded revisions, 313,600, not the reduction in revision effort. This is where the emphasis is easiest to get wrong. When you propose a work instruction system internally, the easiest benefit to explain is that revising documents gets easier. But in B the revision effort saving is 204,750 baht while the escaped-revision saving is 313,600 baht, and the latter is larger. The main axis of the investment is not that work gets easier, it is that incidents get rarer. Write the capital request with that order reversed and you leave most of the benefit unexplained.

Third, C cannot be justified on cost-benefit grounds. Its payback of 17.13 years is longer than B and roughly level with A. The subtotal of benefits does rise from 624,430 baht in B to 668,430 baht in C, an increase of 44,000 baht, but the annual running cost rises from 360,000 baht to 470,000 baht, an increase of 110,000 baht, so the net annual benefit actually falls from 264,430 baht to 198,430 baht. The investment is 840,000 baht higher, so payback naturally recedes.

Stating this asymmetry rather than hiding it is the position of this article. C is not a higher grade of B. It is a configuration with a different purpose. The reason to choose C is not payback; it is that customer audits, IATF, medical devices and similar contexts require proof of which revision a specific person was looking at on a specific day. Where that proof is a requirement, C’s cost is not a subject for payback arithmetic – it is a condition of continuing to trade. Conversely, choosing C at a site with no such requirement means buying an investment that does not pay back, at a higher price than one that does. Sales proposals tend to steer towards the higher configuration, but the only thing to decide is whether the prove layer is required of you.

Finally, be explicit about what is not counted as a benefit. A work instruction system does not reduce headcount. What was counted here is indirect revision effort, rework from escaped superseded revisions, new operator ramp-up time, and audit response effort. All of them reduce hours and incidents, not the number of people standing on the line. The 180 operators are still 180 operators afterwards. Put labour reduction into the benefits and you can write any payback figure you like, but it will not happen, and it comes back as a shortfall after the capital request has been approved. The reason payback comes out as long as 9.68 years is precisely that labour reduction is excluded. Presenting that openly tends, in the end, to get internal agreement faster.

What happens when the assumptions break – three sensitivity cases

Every figure so far sits on the assumptions fixed earlier. If your values differ, so does the conclusion. Vary a limited set of items and see how B’s payback moves. The important discipline here is to move only the items that are actually affected. Apply a uniform factor to every benefit and you also cut benefits that do not depend on that factor, and the conclusion flips for no reason.

The first case has the largest effect, pushing payback from 9.68 years to 40.74 years. At a site where superseded revisions escape only 5 times a year, investing in B cannot be explained on cost-benefit grounds. But a caution applies. Whether the true figure is 14 or 5 is something almost no plant actually counts, because repair dissolves into production overtime and sorting into quality assurance’s daily routine. Running this model on a feeling that it is probably about 5 a year can only err in the direction of understating reality. The first task in any evaluation is to go back over the last year of repairs, sorting and customer reports and count how many were caused by use of a superseded revision.

The second case, revision frequency, stretches payback to 19.28 years. At a site where 420 a year falls to 150, the investment in B is too heavy. Lines with a fixed part mix that have settled into long-run volume production sit in this territory. The more the part mix churns, the more the investment is justified.

The third case is the most important sensitivity in this article. It assumes the master itself is authored in Thai. If there is only one language and the master is in Thai as well, translation effort disappears entirely and the 350,000 baht for multilingual linkage design is not needed, so payback shortens to 13.01 years. The implication is worth stating plainly. A single language pays back faster. But the risk of superseded revisions surviving is greater in a multilingual plant. In other words, the plants that show the shorter payback are the ones that need the system least.

That inversion is a warning about the decision framework itself. Choose a configuration on payback alone and the highest-risk sites end up concluding that it does not pay, so they defer. When a multilingual site evaluates B, the payback figure needs to sit next to actual counts – how many superseded-revision escapes occurred in the year, and how many of those were caused by a language version that had not been updated.

Ten things to settle on paper before choosing a work instruction system

Settle the following on paper before you watch a product demo. Go into a comparison with any of these unsettled and the axis of comparison will be set by the vendor’s proposal.

Of these, items 4 and 5 are the questions that genuinely separate configuration A from configuration B. In the demo, do not ask what happens to the superseded revision. Ask this instead – approve a revision now, then try to open the old one in front of me and show me that screen. Some products display a warning but still let you read it; others stop the viewing outright, and that difference maps directly onto the reduction rate for escaped superseded revisions.

The prove layer requirement in item 7 should be confirmed in documents, not by feel. The customer quality agreement, the audit checklist, the requirements of the applicable standard. Somewhere there either is or is not a requirement equivalent to being able to identify the revision of the instruction in force at the time of work. Choose C without one and you pay an extra 840,000 baht for proof nobody asked for. Stop at B when there is one and you leave an item you cannot answer in the audit room.

Item 10 is a matter of quotation format. Line up quotations that show initial cost only and the differences in annual running cost stay invisible. In this article’s model the annual running cost is 120,000 baht for A, 360,000 baht for B and 470,000 baht for C, a wide spread that heavily influences payback. Specify the format from the buying side at the point of requesting quotations – initial cost and annual running cost separated, three years shown.

How to roll it out – a 90-day roadmap for one process

A plan to migrate every process at once will almost certainly stall part way. Trying to migrate 480 instructions in one go makes the migration itself crowd out the day job, the floor keeps using paper until it is finished, and the period during which paper and electronic coexist stretches out. As noted earlier, there is no empirical support for the idea that running paper alongside digital adds anything.

What works is to pick one process, run a full cycle in 90 days, and roll the revision numbering scheme and translation operating rules you learn there out to the rest.

The most important judgement in this plan is the one on days 46-60 – anything not complete comes out of scope. Digitising an instruction provisionally when one of its 3 languages is missing simply reproduces revision drift from that point on. Leave those on paper and record them explicitly as out of scope. A half-digitised instruction is the most dangerous object of all.

Pushing one revision through and measuring the elapsed time is inside the 90 days because that measurement becomes the basis for estimating the rollout. This article’s ledger puts a revision at 195 minutes, but that is an assumption. Once you have a measured figure from your own site, substitute it into your ledger and redo the payback calculation.

There is a criterion for choosing the target process – pick one with frequent revisions. Choose a stable process with few revisions and there is a real chance that not a single revision occurs during the 90 days, and the period ends without your ever confirming the one cycle you most wanted to test, from revision to retirement.

Issues specific to Thailand and ASEAN

When a plant in Thailand evaluates digitising its work instructions, four conditions apply that never come up in the Japanese domestic discussion.

You have to design on the assumption that people turn over. Manufacturing in Thailand depends to a considerable degree on workers from Myanmar, Laos, Cambodia and Vietnam, and the framework for their residence and employment is renewed periodically by administrative measures. On 14 July 2026 the Thai cabinet approved an extension of work permits for migrant workers of Lao, Myanmar and Vietnamese nationality, covering approximately 770,000 people as at 30 June 2026. The previous measure was due to expire on 11 December 2026, and securing labour for manufacturing and services was given as the reason. In an environment where numbers of that scale move in step with regulatory deadlines, an operation that assumes procedural knowledge accumulates in individual experience does not hold. Instructions have to be written to a standard where reading them is enough, provided in the reader’s own language, and kept from falling out of date.

Wage levels change the basis of the cost calculation. The statutory minimum wage in force from 1 July 2025 is 400 baht per day in Bangkok, Phuket, Chachoengsao, Chonburi, Rayong and Koh Samui district in Surat Thani, 380 baht in the municipal areas of Chiang Mai and Songkhla, 372 baht in Nonthaburi, Pathum Thani, Samut Prakan, Samut Sakhon and Nakhon Pathom, and 359 to 337 baht in the remaining provinces. This article’s model sits in Chonburi, but the direct labour rate of 95 baht per hour used in the calculation is not derived directly from the minimum wage. It is set independently, as an assumption on an employer-cost basis including allowances and statutory benefits. Calculate the return on a work instruction system using a Japanese parent company’s standards and the order of magnitude of the labour rate is wrong, so the savings come out overstated. Always substitute local labour rates. At the same time, a low labour rate does not lead to the conclusion that you should just do it by hand. As the preceding sections show, the main axis of the benefit is not labour reduction but the reduction of escaped superseded revisions.

A revision of ISO 9001 is coming. The revision of ISO 9001 is at the FDIS, or final draft international standard, stage, with publication expected in September 2026 and a transition deadline of September 2029, allowing a 3 year transition period. It is not a published standard yet, so designing your mechanism today around the requirements of the revised edition would be premature. That said, the framework of controlling documented information carries over from the current edition, and the questions of identifying revisions, approving them, ensuring the current version is available, and preventing unintended use of superseded revisions will be asked regardless of what the revision contains. Whether to align the timing of transition work with the timing of digitising your instructions is a judgement to make against your site’s audit cycle.

The wireless environment is a real slice of the investment. The investment table puts the wireless LAN upgrade, 8 access points and design, at 260,000 baht in all three configurations because it is needed whichever configuration you pick. Plants in Thailand are often split across buildings, metal racking and parts bins block signals, and the existing wireless was frequently designed around office requirements. In a configuration that shows instructions on terminals, you have to survey where on the floor the terminals will actually be used before deciding access point placement. Build the quotation on the assumption that the existing infrastructure will do, and additional cost appears at start-up while the floor reverts to pulling out paper whenever the connection drops. That is why the wireless survey sits in days 61-75 of the 90 day roadmap.

Skills transfer and workforce retention are treated as the same class of problem in Japan too. The White Paper on Manufacturing Industries 2026 (Ministry of Economy, Trade and Industry, Ministry of Health, Labour and Welfare, and Ministry of Education, Culture, Sports, Science and Technology, published in May 2026) devotes Chapter 2 to employment trends and securing and developing human resources, and Section 3 of that chapter deals with securing and retaining people and transferring skills at manufacturing companies. In that context, putting instructions in order is the work of moving skill that lives in individuals into documents. This article’s payback calculation, however, includes no monetary value for skills transfer, because there is no defensible basis for converting it into an amount.

Frequently asked questions

What is a work instruction system?

It is the mechanism that authors, revises and approves the master of a work instruction or SOP electronically, delivers it to shop floor terminals and boards, retires the superseded revision, and records which revision the work was done against. It is distinct from a viewer that merely shows instructions on a screen. The decision comes down to 3 points – whether a new revision can retire the old one, whether the system can detect that language versions have followed the master, and whether the revision number can be stamped onto output and inspection records.

Where should we start when digitising work instructions?

Pick one target process and run a full cycle in 90 days, ending with one revision pushed all the way through. What to do in the first 15 days is not to compare products, but to go back over the last year of repairs, sorting and customer reports and count, as a real number, how many were caused by use of a superseded revision. That count is the denominator of the investment decision. In this article’s model, changing escaped superseded revisions from 14 a year to 5 a year alone moves configuration B’s payback from 9.68 years to 40.74 years.

What is the difference between a work instruction system and an electronic forms system?

An electronic forms system digitises the paper operators write on, meaning check sheets and inspection records. A work instruction system deals with the paper operators read. The required functions differ as well – on the forms side, input validation and record integrity; on the instruction side, revision control and retirement. They are consecutive actions on the same tablet, which is why they get conflated, but the investment cases should be built separately. The cost of the forms side and how to choose between product types are covered in the guide to implementing an electronic forms system.

How much does a work instruction system cost?

For this article’s model site, with 40 processes, 120 part numbers, 480 instructions, 180 operators and 3 languages, delivery-only configuration A is 1,100,000 baht, revision-control configuration B is 2,560,000 baht, and records-integration configuration C is 3,400,000 baht. All are estimates built on assumptions, and all include 24 tablets and a wireless LAN upgrade. Watch the share taken by the software itself, which is only 16.4% in A, 24.2% in B and 25.9% in C. The money moves on the terminal, wireless, existing instruction migration and multilingual linkage side. Annual running cost is expected to be 120,000 baht for A, 360,000 baht for B and 470,000 baht for C.

Is showing PDFs on a tablet not enough?

No, because distribution gets faster but the superseded revision does not go away. In this article’s model, configuration A reduces rework from escaped superseded revisions by only 15%, giving a net annual benefit of 66,350 baht and a payback of 16.58 years. That exceeds the useful life of the terminals, so it does not stand up as an investment. With PDFs in a shared folder, overwriting the file leaves the copies already pulled down to local devices alive, and retirement is impossible.

How do you manage multilingual work instructions?

Link master and language versions as parent and child, so that when the master is revised you can extract a list of the language versions that have not been updated. At a multilingual site the unit of revision control becomes language multiplied by revision, and in practice the Thai version is updated first while the Burmese version stays on the superseded revision. Being able to switch the display language and being able to detect that versions have kept up are different functions. In selection, do not ask whether it can switch languages; ask to be shown a list of only the language versions that have not been updated. Translation granularity matters too, so check whether the structure lets you split an instruction into process steps and send only the changed steps for translation.

Summary

What you should be buying when you digitise work instructions is not delivery speed. It is a mechanism that makes the superseded revision disappear from the floor. At this article’s model site, the largest of the 970,450 baht that the paper operation costs each year is the 392,000 baht of rework from escaped superseded revisions, and in most companies that amount appears on nobody’s books. Delivery-only configuration A cuts that item by just 15%, giving a payback of 16.58 years, which does not stand up as an investment. Revision-control configuration B is the shortest at 9.68 years, but what creates the gap is not the reduction in revision effort – it is the 313,600 baht reduction in escaped superseded revisions. Records-integration configuration C comes in at 17.13 years, longer than B, and cannot be justified on cost-benefit grounds. The reason to choose C is confined to cases where a customer audit or a standard requires proof of which revision a specific person saw on a specific day. And a work instruction system does not reduce headcount. What falls is indirect effort and rework, which is exactly why payback comes out long. At a multilingual site in Thailand, the unit of revision control becomes language multiplied by revision, and revising the master leaves the Burmese version on the old revision. Single-language sites show faster payback, but the risk of superseded revisions surviving is greater at multilingual ones. Before you decide, take on board that the plants showing the shorter payback are the ones that need the system least.

How many instructions do you actually have, how many languages are they running in, and how many pieces of rework last year were caused by a superseded revision. TOMAS TECH is happy to be consulted on nothing more than counting those three real numbers. Which of configurations A, B and C stands up under your own conditions is largely settled once those three are on the table. Please get in touch through the contact page for details.

References

• Should firms use digital work instructions? Individual learning in an agile manufacturing setting – Journal of Operations Management
• White Paper on Manufacturing Industries 2026 – Ministry of Economy, Trade and Industry, Japan
• Starting July 1 – Ministry of Labour Revises Minimum Wage – Ministry of Labour, Thailand
• Thailand extends work permits for migrant workers from Laos, Myanmar, Vietnam – The Star
• Overview and schedule – Revision of ISO 9001 coming in 2026 – TUV Rheinland
• Thailand Labour Migration Profile – IOM Thailand