Most “our costs don’t add up” conversations are not about calculation logic. They are about actual production data that never leaves the shop floor. Before you evaluate any manufacturing cost management system, check how completely your plant records who made what, how much of it, and when. This article breaks down the structural reasons costs stay invisible, the three system types and what each one actually covers, how to design production data collection, indicative budget ranges, and the Thailand-specific tax and BOI issues — at a level of detail you can take straight into an internal meeting.
“We can’t get product costs” is a data problem, not a formula problem
Cost system projects usually start with a sentence from finance. “We can’t see profit by product.” “We have no basis for a price increase negotiation.” “Head office is asking for cost by job order.” Then, almost every time, the discussion drifts into accounting theory: which costing method should we adopt, process costing or job-order costing, absorption or direct costing.
That is the first fork in the road, and it is where a lot of projects go wrong. The choice of costing method matters — but only once the numerator and the denominator both exist. In practice, the dominant problem is not the method. It is that the actual figures you are supposed to allocate do not exist anywhere in a usable form. How many kilograms of material were issued against which work order. How many minutes that operation actually took. How many pieces were good and how many were scrapped. In a factory where those three things are not recorded, no costing module — however sophisticated — will produce anything better than the monthly total divided back out by some ratio.
So the argument running through this article is simple: selecting a manufacturing cost management system is not an exercise in selecting a calculation engine. It is an exercise in designing shop floor data collection. The rest of this section explains why.
Why the total cost is always right but the per-job cost is not
Your accounting system captures everything: material purchases, payroll, electricity, lease payments, depreciation. So the total manufacturing cost for the month always comes out. It has to — the books close and the total ties. The information breaks down one level later, when that total has to be split by product, by job order, and by process.
Break the allocation problem into layers and you get three.
Layer 1 is the quantity link. To assign material cost to a job, the goods-issue record needs a work order number on it. In many plants, the movement from the warehouse to the line is recorded down to date, part number, and quantity — but not “which work order this was issued for.” When you try to produce material cost by job from that state, you fall back on theoretical consumption: the BOM quantity multiplied by units produced. Theoretical figures do not reflect yield loss, partial usage at the line, or how offcuts are handled, so they never match what was physically issued. The gap then gets spread across all products at month end as a difference nobody can explain, and it reappears in the report every month as an unexplained swing.
Layer 2 is the time link. Labour cost and most manufacturing overhead are allocated on a time basis. Your attendance system knows the operator worked eight hours. It does not know how many of those minutes went to machining job A, how many to a changeover on job B, and how many to waiting. So labour gets allocated by output quantity or by standard hours instead. Allocating by standard hours means that however much extra time a job actually consumed, the cost report shows it running exactly to standard. In a plant with no time actuals, the labour efficiency variance — the variance with the largest improvement potential of any of them — is structurally invisible.
Layer 3 is the good/defect link. It is not unusual to find plants that record only good quantity and back-calculate defects as “input minus output.” Under that method, the material and processing cost sunk into defective pieces is quietly absorbed into the cost of good units. Nothing surfaces while the defect rate is stable. Then a month comes where one lot goes bad, and it appears as a cost increase nobody can account for.
If any one of those three layers is missing, cost by job order is not “impossible to calculate.” It is something worse: calculable but untrustworthy. Numbers the shop floor does not believe get rejected; rejected numbers never get used for improvement; unused numbers never improve in accuracy. That loop is why cost management projects keep resurfacing every few years in the same factory.
Four entry points where cost data breaks in Thai plants
From what we see across Japanese-owned plants in Thailand, cost data breaks in roughly four places. There is a common pattern to it: the parent plant in Japan prevented these failures with a mechanism, while the overseas site relies on people following a routine — and the routine erodes.
Entry point 1: unrecorded material issues.
Supply from the warehouse to the line runs on “call out and someone brings it” or “the line comes and takes it.” Daily production still flows, but system inventory and physical inventory diverge, and the difference is cleaned up in one lump at the stock count. That stock-count difference should properly be distributed to material cost by job. In practice it becomes a period-end adjustment smeared thinly across everything. The bigger the difference in a given month, the less reliable product costs are that month.
Entry point 2: rounded changeover times.
Setup time gets written on the daily paper report as “30 minutes” or “1 hour.” Because setup weighs most heavily on small-lot items, rounding distorts products asymmetrically. If a changeover that really takes 45 minutes is logged as 30, small-lot items are systematically undercosted, and management concludes that “small lots are actually quite profitable.” When pricing is built on that number, loss-making orders can run for years before anyone notices.
Entry point 3: no distinction between scrap and rework.
Scrap and rework mean completely different things in cost terms. Scrap means the material and processing cost invested up to that point is lost. Rework means additional processing cost is incurred. Recording both as a single “NG quantity” makes it impossible to tell a material loss from a labour loss, which means there is no way to decide where to act. We also see daily-report formats transplanted directly from the parent plant in Japan that simply do not fit Thai reality — for example, processes where the rework ratio is far higher than the format anticipates.
Entry point 4: flat overhead allocation.
Indirect department costs, equipment depreciation, electricity, and plant administration are allocated at a single uniform rate based on production value or direct labour cost. Flat allocation is simple and easy to defend in an audit, but in a plant that mixes capital-intensive and labour-intensive processes it diverges badly from reality. It matters most where certain processes consume large amounts of electricity. In Thailand, the electricity tariff was raised for the May–August 2026 period to an average of 3.95 baht/kWh, up from 3.88 baht in the January–April period (+1.8%) — the first increase in seven periods. The fuel adjustment charge (Ft) is 0.1623 baht per unit, and the government has allocated roughly 9.472 billion baht in subsidy (source: wisebk.com, 2026). When energy costs are rising, flat allocation blunts your ability to judge which products should absorb that increase.
None of these four has anything to do with a system’s computational power. They are problems of operating procedure and record design, which is exactly why replacing the system does not fix them. The corollary is more useful: redesign these four points and cost resolution improves substantially even on the tooling you already own.
Why cost management came back onto the agenda in 2026
Cost management is a low-priority topic in a good year. When sales are growing, a certain amount of costing vagueness gets absorbed by profit. The reason it has resurfaced in 2026 is a change in the macro environment surrounding manufacturing in Thailand.
The Federation of Thai Industries (FTI) puts its 2026 outlook at 1.6–2.0% GDP growth — slowing from an estimated 2.0% in 2025 — with exports possibly contracting by 1.5% to 0.5%. It also notes that capacity utilisation in several sectors is stuck below 60%, against a normal 70–80%, and that energy costs, raw material costs, wages, and financing costs are all squeezing SMEs (source: nationthailand.com, 2026). The Manufacturing Production Index (MPI) was down 0.36% year on year in April 2026 — hardly a picture of strength (source: english.news.cn, May 2026).
The cost side is moving up at the same time. The minimum wage stands at 337–400 baht per day as of 2026 (a national average of roughly 374 baht), and while there is no clear sign yet of a further increase this year, the ceiling used to calculate social security contributions was raised on 1 January 2026, pushing up the total employment cost borne by employers (sources: thailawonline.com, en.thairath.co.th). Electricity, as noted above, is also on the way up.
Flat revenue, rising costs, falling utilisation. When those three land at once, the question “which products are actually making money?” becomes unavoidable. Whether you are negotiating a price increase, deciding which items to cut back, or switching between in-house production and outsourcing, the evidence you need is cost by product and by process. That, as far as we can tell, is why manufacturing cost management system evaluations have picked up in 2026.
Falling utilisation breaks your fixed-cost absorption
From a costing perspective, a drop in utilisation means more than lost revenue. It breaks the fixed-cost absorption rate.
Take a simplified example. A plant has monthly fixed costs — depreciation, indirect labour, rent, the standing portion of the electricity bill — of JPY 10 million, and normally runs 10,000 machine hours a month. The fixed overhead rate is therefore JPY 1,000 per hour, and the standard cost is built on that rate. Now suppose orders fall and actual running time in a month is only 6,000 hours. The fixed costs incurred are still JPY 10 million, so the real burden is about JPY 1,667 per hour. But products are still absorbing overhead at the standard JPY 1,000 per hour, so only 6,000 hours × JPY 1,000 = JPY 6 million lands on product cost. The remaining JPY 4 million is never absorbed.
That unabsorbed JPY 4 million is the volume variance (also called idle capacity variance). The critical point is that it is not the result of products being made inefficiently. It is the result of not making enough of them. It arises entirely independently of anything the shop floor did.
Two mistakes tend to follow. The first is redistributing the volume variance back onto product cost. Do that and unit costs look higher in exactly the months when volume was low, leading to the false conclusion that “costs went up, so we need a price increase” — when in fact only the volume fell and manufacturing efficiency may be unchanged. The second is lumping the volume variance into an “other” line, at which point you can no longer tell whether the month’s margin deterioration came from shop floor efficiency or from a shortfall in orders.
If utilisation really is below 60% in some sectors, as the FTI indicates, the volume variance will not be a rounding error. In a plant with no mechanism to separate it out, this is precisely the moment when someone frames the problem as “shop floor productivity has fallen,” and improvement resources get spent in the wrong direction. Being able to decompose and read cost variance analysis has real, immediate value in the 2026 environment for exactly this reason.
Three types of manufacturing cost management system and what each covers
“Cost management system” covers products with three quite different origins. This is not a ranking. They cover different ground, and which one you should choose depends on where your problem actually sits.
| Dimension | Type A: costing module in accounting/ERP | Type B: costing built into a production management system | Type C: dedicated costing package + BI |
|---|---|---|---|
| Origin | Financial and management accounting | Production planning and shop floor execution | Cost accounting and analysis |
| Strong at | Total-level consistency, tying to the closed books, departmental P&L | Actual cost by job order and process, feeding variances back to the floor | Multi-axis analysis, simulation, visualisation |
| Weak at | Process-level granularity, timeliness | Reconciling to accounting, rigorous overhead allocation | Holds no actual data of its own |
| Source of actual data | Depends on an interface from production systems | Collects it directly (manual entry + machine integration) | Imports from both |
| Realistic update frequency | Monthly | Daily to real time | Depends on the source |
| Primary users | Finance, administration, head office | Production control, plant manager, manufacturing sections | Corporate planning, cost engineering, management accounting |
| Implementation burden | Medium (entails redesigning accounting) | High (entails changing shop floor practice) | Medium (data integration design is the crux) |
| Best when | You need cost to agree with the closed books | You need the real picture by job order | You already have data but cannot analyse it |
Type A: the costing module in accounting or ERP
This is the cost accounting module bundled with an ERP suite or accounting system. Because it is wired directly to the general ledger, the cost it produces always agrees with the closed books. For audit response and head office reporting, it is the safest choice available.
Its limits are granularity and freshness. Since it works from journal entries, cost is only final after the monthly close. It is also usually not designed to handle detailed actuals at the process or machine level, so producing cost by job order requires importing actual data from the production side. In other words, choosing Type A does not solve the data collection problem. In a large share of the “we implemented ERP and still can’t get costs” conversations we have, the project went ahead without anyone recognising this structure.
Type B: costing built into a production management system
This is the costing capability inside a production management system, including MRP and MES-class products. Because work orders and actuals sit in the same database, its greatest strength is producing actual cost by job order and by process on a daily basis. When a variance appears, you can trace it to the specific order and the specific operation where it arose. Of the three types, this is the one that most readily produces information at a granularity you can hand straight back to the shop floor.
Its weakness is reconciliation with accounting. Costing on the production side exists to manage operations, so overhead treatment is often simplified and period attribution may not match the accounting view. You end up with two sets of numbers — “the plant’s management cost” and “the accounting cost” — and you need a routine that explains the difference between them. Whether you can build a team capable of explaining that difference every month is what decides whether Type B succeeds. If you are looking at the wider picture and not just the costing function, our 2026 production management system comparison works through how to evaluate these platforms as a whole.
Type C: a dedicated costing package plus BI
This approach uses a package specialised in cost accounting, or builds a costing analysis layer on a data warehouse plus BI tools. Multi-axis comparison of standard cost vs actual cost, price revision simulation, profitability impact analysis for product mix changes — analytical freedom here is in a different league.
The catch is that this type collects no actuals of its own. It works extremely well where a production system or a data collection layer already exists, data has been accumulating, and nobody can make use of it. Conversely, if a plant without reliable actual data picks this first, the result is the worst possible outcome: a beautiful dashboard displaying estimates.
How to position this against an existing ERP
The most common real-world situation is “head office mandated an ERP and we can’t touch it.” In that case you have two options: go deeper into the ERP’s costing module (strengthen Type A), or hold actual cost on the production side and interface it into the ERP monthly (add Type B).
The decision rule is simple. If the output being asked for is product profitability for the closed books and head office reporting, strengthen Type A. If it is evidence for shop floor improvement and pricing, add Type B. The first is valued for accuracy and consistency; the second for freshness and granularity. Trying to satisfy both fully in a single system is how requirements balloon and projects stall. The realistic discipline is to pick one primary purpose and cover the other through a monthly reconciliation.
Standard cost, actual cost, and estimated cost: using each for its purpose
Cost comes in several varieties, each with a different job. They get conflated constantly, so let us define them first — this matters especially for readers who did not grow up inside a Japanese management accounting department.
Standard cost is a predetermined cost set before production begins, based on scientific and statistical investigation. It defines what cost *ought* to be: standard material price × standard consumption, standard wage rate × standard operation time, and so on (sources: smartf-nexta.com, nec-solutioninnovators.co.jp). The role of standard cost is to be compared against actuals so that variances surface and improvement targets can be identified.
Actual cost is calculated from the material, labour, and expense actually incurred. This is what the financial statements use. Looked at alone it offers no benchmark for whether a number is high or low, which makes it a poor management indicator in isolation.
Estimated cost is what you calculate at the quotation or new-product development stage, when no actuals exist yet. It is built from actuals for similar past products or bottom-up from design information. The accuracy of your estimated cost is, directly, the accuracy of your decision on whether an order is worth taking.
Translated into an operating routine, the three work like this. Estimated cost decides whether to accept an order and at what price. Standard cost provides the day-to-day management benchmark. Actual cost confirms the result. Then “estimated vs actual” tests whether your order acceptance judgement was sound, and “standard cost vs actual cost” tests whether the manufacturing process is behaving. When both of those verification loops are turning, cost management is functioning.
In most factories, standard cost was set some years ago and has never been revised. Material prices and wage rates have moved, but the standards have not, so variances run persistently in one direction and the whole exercise decays into “the same variance every month, so there’s no point looking.” Decide the rule up front: revise standard cost once a year, and mid-year as well if material markets or wages move significantly. In an environment like Thailand’s — a minimum wage spread of 337–400 baht per day and a revised ceiling for social security contribution calculations — it is worth putting the wage rate review on the calendar as a fixed event.
Classifying cost variances and how to read them
The point of variance analysis is not “how much did we miss by in total” but “why did we miss,” broken out by cause. The basic split is between price factors and quantity factors.
| Variance | How it is calculated | Typical causes | Where responsibility usually sits | Direction of action |
|---|---|---|---|---|
| Material price variance | (actual unit price − standard unit price) × actual quantity consumed | Purchase price movement, FX, supplier change, freight | Purchasing / procurement | Review suppliers, consolidate orders, revise standard price |
| Material quantity variance | (actual quantity consumed − standard quantity) × standard unit price | Yield deterioration, defects, offcuts, weighing accuracy, loss or theft | Manufacturing / engineering | Yield improvement, BOM review, tighter issue control |
| Labour rate variance | (actual wage rate − standard wage rate) × actual hours worked | Overtime ratio, change in workforce mix, wage increases, allowances | HR / production management | Shift design, multi-skilling, revise standard rate |
| Labour efficiency variance | (actual hours worked − standard hours) × standard wage rate | More changeovers, equipment trouble, rework, insufficient training | Manufacturing / production engineering | Setup reduction, training, equipment maintenance |
| Volume variance | (actual activity level − normal activity level) × fixed overhead rate | Order shortfall, low utilisation, unplanned stoppages | Sales / management | Secure orders, revise production plan, rethink fixed cost structure |
Note: the formulas above give the magnitude of each difference. Whether a given variance is favourable or unfavourable depends on the variance type. Before you design your internal reporting format, agree a single convention for which sign is treated as unfavourable.
The column worth dwelling on is the second from the right: where responsibility sits. The greatest practical benefit of decomposing variances is that it identifies which department owns the improvement. “Material cost went up” moves nobody. “That increase is a price variance, so it belongs to purchasing” or “that is a quantity variance, so it belongs to manufacturing” produces a decision for next month.
Three practical tips on reading these. First, price variances are heavily driven by external factors, so start with quantity and time variances — that is the territory you actually control. Second, where the labour efficiency variance is chronically large on a given process, suspect that the standard time is out of date. What needs fixing may be the standard, not the shop floor. Third, as noted above, the volume variance is not the shop floor’s responsibility, so separate it out of any report shown to the floor. If you do not, operators feel they are being evaluated on numbers that are not their doing, and trust in the whole cost data set evaporates.
Production data collection design decides 90% of your cost accuracy
Here is the core of it. In evaluating a manufacturing cost management system, the area that deserves the most comparison time is not the calculation functionality. It is how you will collect the actuals.
The five data points to collect, and no more at first
Many plants set out to “capture everything,” watch requirements balloon, and then collapse because the shop floor cannot absorb the entry workload. The opposite is also true: with the following five items in place, cost by job order and by process can be calculated to a genuinely usable accuracy. Start by narrowing to exactly these.
| # | Data point | What it is used for in costing | Difficulty of capture | Alternative approach |
|---|---|---|---|---|
| 1 | Start time / finish time | Labour efficiency variance, machine running time, basis for overhead allocation | Medium | Automatic capture from machine signals |
| 2 | Good quantity | Denominator of unit cost, confirming output | Low | Inspection process records, back-calculation from shipments |
| 3 | Defect quantity (by process and defect type) | Material loss, rework hours, yield variance | Medium | Digitising inspection records |
| 4 | Material issue quantity (linked to work order) | Material quantity variance, actual material cost | High | Weighing scale integration, barcode goods issue |
| 5 | Changeover time | Cost of small-lot items, measuring setup improvement | Medium | Inferred from non-processing time on the machine |
A note on why these five. Items 1 and 5 form the basis for allocating labour and overhead, and they are also where the largest improvement headroom sits. Items 2 and 3 are essential to understanding yield and defects — without them the material quantity variance cannot be explained. Item 4 is the hardest to capture, but in industries where material accounts for a high share of cost (assembly, plastic moulding, metalworking), material is more than half of total cost, so there is no way around it.
There are equally things you do not need to force into scope at the start: actuals by individual operator, tool and jig usage counts, consumable usage by job order, electricity consumption by machine. These deepen analysis, but including them from day one doubles the number of entry fields and undermines adoption on the floor. The right sequence is to get the skeleton of costing moving first, then add depth on the specific processes that need it.
Choosing your input method
Even with the same five items, how you collect them changes both accuracy and shop floor burden substantially. Here are the three main options compared.
| Method | Accuracy | Burden on the floor | Initial cost | Suited to | Watch out for |
|---|---|---|---|---|---|
| Handheld terminal (barcode / QR) | High — scanning produces few entry errors | Low to medium | Medium | Material issues, receiving and shipping, inter-process moves, lot traceability | Label operations must be designed first; an unlabelled physical item cannot be scanned |
| Tablet-based electronic forms | Medium — human judgement is involved | Medium | Low to medium | Inspection records, defect classification, changeover records, daily reports | Without a design that narrows the choices, free text proliferates and analysis becomes impossible |
| Automatic collection from PLC / machines | Highest — no human involvement | Minimal | High | Running time, processed quantity, micro-stoppages, electricity by machine | Requires investigating the signal specification; older machines need retrofitted sensors |
The deciding question is whether that data point requires human judgement. Running time and processed quantity are things the machine already knows — there is no rational case for having a person type them in. If you can take them from the equipment, take them. Defect cause classification and the decision on whether a piece needs rework, on the other hand, do require judgement, so a tablet form with predefined options is the right fit. Material issues need physical goods and data to agree, which makes barcode scanning the most reliable option.
Two practical notes before you commit to a method. If you go the handheld route, design the label operation first — which physical units carry a label, who prints it, and when — because a scanner cannot read an item that was never labelled, and that single gap is what breaks material issue records in practice. If you are replacing paper daily reports and inspection sheets with tablets, the work is less about the hardware than about converting free-text fields into a fixed, short list of selectable options; free text is what makes the resulting data unanalysable.
A realistic answer to “more data entry will stop production”
Raise cost management on the shop floor and someone will say, without fail, “add any more data entry and production stops.” That concern is legitimate. We have watched several projects add too many input fields and see the routine collapse within a week.
There are three realistic answers.
First, fill in the automatically capturable items first. Running time, processed quantity, and downtime can be taken from the equipment. Automate that and items 1 and part of 5 of the five are covered with zero human effort. Even where machines are too old for a PLC connection, you can retrofit a current sensor or photoelectric sensor and capture running versus not-running. Accuracy is lower, but it is unquestionably more accurate than rounded numbers on a paper report.
Second, ride on work people already do. Do not create a new task; layer data capture onto a movement that already happens. Fetching material is already happening, so add exactly one barcode scan to it. Inspection is already happening, so change where the record goes — paper to tablet — and nothing else. “One more task” and “the record goes somewhere different” land completely differently with operators.
Third, build the return path at the same time. The single biggest reason data entry stops is that what gets entered is only ever used by head office or finance, and nothing comes back. Simply displaying actual time by process and defect rate on a floor monitor daily changes what the entry means to the people doing it. You do not need to show them cost. Translate it into something the floor can act on — “today’s changeover time was shorter than last week’s average” — and send that back.
Where exactly to draw the line on process-level granularity is covered in our guide to process management system cost and selection for 2026. If your plan extends to equipment integration, read it alongside our MES implementation cost comparison for 2026.
Breaking the cost into five layers
At budgeting time, the most common failure is to get approval based on licence cost alone, then come back for a supplementary request when data collection infrastructure and integration development pile up on top. Manufacturing cost management system cost should be estimated across the following five layers.
The figures below are indicative ranges based on general market price bands. Actual pricing varies considerably by product, vendor, and requirements. The assumption behind them is a mid-sized Japanese-owned plant at a single site in Thailand: 1,000–5,000 part numbers, 10–30 main processes, 10–30 data entry terminals, and roughly 20–50 users. More sites or more part numbers push the figure up proportionally; simpler processes and fewer items pull it down.
| Layer | What it covers | Indicative range (year 1) | What moves it | Easy to overlook |
|---|---|---|---|---|
| ① Licence / subscription | Software usage fee, perpetual or annual | JPY 1.0M–6.0M equivalent per year | Users, modules, sites | Annual version upgrade fees are sometimes billed separately |
| ② Initial build and master data | Item master, BOM, routings, rate setting, initial parameters | JPY 2.0M–8.0M equivalent | Number of items, routing complexity, condition of existing data | The most under-estimated layer. Poor existing master quality inflates it |
| ③ Data collection infrastructure | Handheld terminals, tablets, wireless LAN, label printers, PLC connection hardware | JPY 1.5M–7.0M equivalent | Number of terminals, plant floor area, number of machines in scope | In-plant wireless coverage, dust/water-resistant specs, spare units |
| ④ Integration with other systems | Interfaces to accounting/ERP, production management, sales | JPY 1.0M–5.0M equivalent | Number of endpoints, availability of APIs, data conversion complexity | If head office ERP needs modification, that budget and approval sit at head office |
| ⑤ Operation, maintenance, improvement | Annual maintenance, help desk, master maintenance, standard cost revision, enhancements | Roughly 15–25% of licence annually, plus internal effort | How much you can do in-house, vendor support model | Internal staff time is frequently left out entirely |
Layers ① to ④ together come to roughly JPY 5.5M–26M equivalent for year one. Layer ⑤ is an annual figure from year two onward and is not included in that year-one total. If you are considering a perpetual licence, split it into the one-off initial fee and annual maintenance, then compare it against subscription pricing on an annualised basis.
Two layers deserve particular attention: ② and ⑤.
Layer ② gets under-estimated because in almost every case someone says “we already have master data.” But master data capable of supporting cost accounting means something specific: item codes uniquely rationalised, BOMs matching the actual input configuration, standard times set on the routings, and rates (labour rate, machine hour rate) defined per process. Master data built for production control very often has those last two blank. Setting standard times and rates is not a data entry job — it involves studying the processes. Fail to budget the effort for it and go-live can slip by months.
The internal effort in layer ⑤ is the same story. Standard cost needs revising annually, new items require master registration, and process changes require routing updates. Implement without deciding who does that upkeep and within a year the master has drifted from reality and cost credibility is gone. Explicitly reserve a few person-days a month in the production control section or with the cost accounting owner.
How to build the business case
Approval processes ask for a benefit figure. Inventing a number you cannot support will hurt you later, so the honest structure is roughly this.
First, finding loss-making orders. In a plant with no visibility of actual cost by product, a certain proportion of items is running at a real loss. What proportion varies by factory, so the credible approach is to trial-calculate a handful of items from existing data before implementation and present an estimate extrapolated from that sample.
Second, yield and changeover improvement. Once actual time by process and defect rate become visible, the priority order for improvement targets becomes clear. You cannot promise a benefit figure in advance here, but you can frame it accurately: today there is no basis for choosing what to improve, and that is the state you are exiting.
Third, reducing the effort of costing itself. In many plants the monthly cost roll-up consumes several person-days of manual Excel work. That is measurable, which makes it the most defensible benefit line you have.
Fourth, lower tax and audit compliance cost. The effort of producing transfer pricing documentation and BOI incentive segmentation material — both discussed below — falls reliably as cost data granularity improves.
Implementation roadmap: the first 90 days and the first six months
Cost management projects have a wide scope, so without phase boundaries they always overrun. As a working target, design the first 90 days to reach “we understand our real material cost” and the six-month mark to reach “actual cost by job order, calculated daily.”
Days 1–30: inventory the current state and fix the objective.
Start by mapping what data exists where: the chart of accounts in the accounting system, work orders and actuals in the production system, paper daily reports on the floor, Excel roll-ups. Lay all of it out on one page and make visible how much of the five required data points is actually covered. In parallel, agree the output with management. “Actual cost by job order, monthly” and “actual work time by process, daily” require different designs. Going into product comparisons while this is still vague is the single biggest failure pattern.
Days 31–60: master data and process design.
Deduplicate the item master, verify BOMs against reality, review routings, and set rates. In parallel, decide the operating rules for material issues: who issues, when, in what unit, and how it is recorded. You will barely discuss systems during this period. The success of everything downstream is nonetheless decided here.
Days 61–90: run material cost through first, on its own.
Do not switch everything on at once. Start with material issue recording. Material is the largest cost component in most plants, and it is also the area where the effect of linking to a work order shows up fastest. Pilot on one or two lines, and use it to surface missing records and impractical procedures. Even at this stage — with nothing more than material cost by job order — the signs of loss-making items begin to appear.
Months 4–6: add time and defect actuals, move cost to daily.
Once material handling is stable, add start/finish times and defect quantities. Prioritise processes where the data can be captured automatically from equipment, and keep manual entry to a minimum. With this in place you can calculate actual cost by job order and by process on a daily cycle. At the same time, finalise the format of the variance report against standard cost, and build a monthly variance review meeting into the business process.
Month 7 onward: reconcile to accounting, then roll out.
Reconcile the cost calculated on the production side against the accounting total each month, and get to a state where the difference can be explained. Only after that reconciliation is established should you roll out to other lines or other sites. Roll out without an agreed reconciliation rule and you end up with a different cost basis at each site and no ability to compare them.
Thailand-specific considerations
Designing cost management for a plant in Thailand raises three issues that do not arise domestically in Japan. All three share a property: addressing them after the fact requires rebuilding your cost data. That makes them worth designing in from the start.
Transfer pricing documentation and consistency of cost data
Transfer pricing provisions were added to the Thai Revenue Code in November 2018, and companies above a certain size are required to prepare a Local File. Documentation may be requested up to five years from the filing date of the disclosure form; where requested, it must generally be submitted within 60 days (180 days for a first-time notice only) (sources: KPMG Thailand GJP Newsletter, March 2026; PwC). Into 2026, a growing number of Japanese-owned companies are reported to be receiving summons for transfer pricing audits from the Thai Revenue Department (TRD).
Cost data comes under scrutiny in a transfer pricing context mainly where the cost plus method or the transactional net margin method (TNMM) is applied. Where the Thai entity is positioned as a contract manufacturer and sets its selling price to the parent as “cost plus a defined margin,” it has to be able to explain both the scope of that “cost” and how it was derived.
The question that bites is whether your method of calculating cost by product can be explained at all. “We allocate overhead in proportion to sales” invites the follow-up: why is that allocation base reasonable? Allocation based on actual time by process is far easier to defend. And within a 60-day window, reconstructing several years of product-level cost and its allocation basis is realistically very difficult without daily actual data. Viewed this way, implementing a cost management system carries a tax risk mitigation dimension.
Three practical design points: (1) document your allocation bases and do not change them mid-period; (2) where a change is necessary, record the reason and the effective date; (3) preserve the product costing process in a reproducible form. If the calculation runs automatically in a system, all three are satisfied as a by-product.
BOI incentives and cost segmentation
Companies holding incentives from the Board of Investment (BOI) must compute income separately for promoted and non-promoted activities. Where both kinds of product are made in the same plant, how to apportion shared costs becomes the issue.
Looking at investment trends, BOI applications in the first half of 2026 totalled 1.473 trillion baht across 1,299 projects, of which digital industries accounted for 1.115 trillion baht across 90 projects. Approvals came to 1.306 trillion baht across 1,300 projects, with reported job creation of more than 82,000. Notably, upgrade measures under “Smart and Sustainable Industry” reached 132 projects worth roughly 17.2 billion baht (source: nationthailand.com, 2026) — evidence that incentives for equipment renewal, automation, and energy efficiency investment at existing plants are actively being used.
Two implications for cost management. First, investment in data collection infrastructure — machine integration, sensors, systems — may qualify under those upgrade measures. It is worth checking with your BOI contact or accounting firm early in the evaluation. Second, taking on a new incentive creates a promoted/non-promoted split, which means your cost data has to be structured to withstand that split.
Concretely, “structured to withstand it” means: the item master can carry an incentive classification attribute; actual time on shared processes is held at work order level and can be aggregated to promoted and non-promoted buckets; and the allocation basis for shared costs can be explained on an actuals basis. Retrofitting the classification item by item after the fact is extremely difficult unless past actuals were linked to work orders.
Language, turnover, and making shop floor entry stick
More than any technical consideration, this third point is what decides whether a project succeeds.
Multilingual support should be treated as mandatory. The standard configuration is Thai on the shop floor entry screens and Japanese or English on the management and analysis screens. Compromise here and run everything in English only, and entry errors rise and data quality falls. Cost accuracy is almost entirely governed by entry accuracy, which makes screen language a direct investment in accuracy. On lines with a high proportion of Myanmar or Cambodian operators, consider icon-led screen designs that do not depend on reading, or picture-based options.
Turnover is addressed by changing the structure of training cost, not by fighting the turnover itself. A system that takes half a day to learn generates training effort every time a person is replaced. Design it so that the operator scans a barcode or presses a button on screen, and training takes minutes. This is precisely why “few things to remember” should rank above “rich in features” in your evaluation criteria. Manuals in Thai, picture-led, one screen per page, make adoption noticeably faster.
Making entry stick comes down to bringing the supervisory layer along. The dividing line is whether line leaders and team leaders see the data as usable for their own improvement work. Before implementation, show that group concrete examples of what the current data cannot tell them about a given process, and explain what will become visible. That earns cooperation. The opposite approach — finance instructing the floor to “enter this so we can produce costs” — reads to operators as nothing but tighter control.
Five common failure patterns
Here are the failure patterns we see most often in cost management projects. None of them is technical.
Failure 1: starting from the choice of costing method.
Months disappear into accounting theory — process costing or job-order costing, whether to adopt direct costing. Method selection matters, but without actual data no method will produce accuracy. Start from “which data are we actually capturing today.”
Failure 2: deferring data collection.
The system is chosen first and data collection is handled “through operating procedure.” After go-live the floor cannot sustain entry, data gaps appear within a few months, and you are back to no usable cost. Be prepared to spend half the project’s total effort on designing data collection and building agreement on the floor.
Failure 3: mis-estimating master data effort.
The plan assumes the item master and BOM “already exist,” and duplicates, gaps, and divergence from reality only emerge after work starts. Routings and standard times take particularly long. Before the implementation decision, check the actual state of the master for about ten representative items — that gives you a usable sense of the effort required.
Failure 4: never updating standard cost.
The standard set in year one is carried forward for years and variances run persistently in one direction. The moment the variance becomes a number you can predict without looking, variance analysis is dead. Put standard cost revision on the calendar as an annual task.
Failure 5: data never returns to the shop floor.
Only finance and head office see the cost report; the people entering the data get nothing back. When the only motivation for entry is “because we were told to,” accuracy will fall. Design the daily return of indicators the floor can act on — actual time by process, defect rate, changeover time — at the same time as you design the cost report itself.
FAQ
What is a manufacturing cost management system?
It is a system for capturing the cost incurred to make a product at the level of individual products, job orders, and processes, and for comparing that against standards or estimates to analyse variances. Where an accounting system handles cost in total, a cost management system handles the granularity of “how much was spent on which process for which product.” In practice it consists of two things: a mechanism to allocate material, labour, and manufacturing overhead based on actual data, and a mechanism to organise the result into variances you can read.
How is a cost management system different from a production management system?
They have different purposes. A production management system exists to make things according to plan — orders, requirements calculation, work orders, progress tracking, inventory. A cost management system exists to establish what it cost. The two overlap heavily in the data they use: the manufacturing actuals held by the production system (work time, good quantity, defect quantity, material issues) are precisely the inputs to cost accounting. That is why many production management products include built-in costing. When evaluating, the key check is whether the actuals on the production side can be captured at the granularity cost accounting requires. Our 2026 production management system comparison sets out the wider landscape.
How much does it cost?
As set out in the five-layer breakdown above, the estimate has to cover more than licence fees: master data preparation, data collection infrastructure, system integration, and ongoing operation and maintenance. For a mid-sized Japanese-owned plant in Thailand (one site, 1,000–5,000 items, 10–30 terminals), the indicative range is roughly JPY 5.5M–26M equivalent for layers ① to ④ in year one, with annual running cost thereafter at around 15–25% of licence plus internal staff time. These are indicative figures based on general market price bands only, and they vary considerably with the configuration of your existing systems, the number of integration endpoints, and the age of your equipment. Layers ② and ③ in particular depend heavily on the state of the plant, which is why nobody can responsibly quote a fixed number without seeing the site.
Where should we start?
We suggest three steps in order. First, take inventory of which actual data you currently capture. For each of the five items — start/finish time, good quantity, defect quantity, material issues, changeover time — confirm whether a record exists and how accurate it is. Second, fix the purpose of the output. Whether it is for the closed books and head office reporting, or for shop floor improvement and pricing, changes which system type you should choose. Third, run a pilot that links material cost alone to work orders. One line is enough. Material accounts for a high share of cost and the effect shows quickly, which also makes it useful for building internal agreement.
When does Excel-based cost management hit its limit?
There are four warning signs: (1) the monthly roll-up takes three days or more; (2) only one person can update the file; (3) comparisons against historical data do not tie and you cannot trace why; (4) you now need granularity by job order and process that manual work cannot keep up with. Sign (4) is the practical turning point. Excel is an excellent analysis tool, but as a database for continuously accumulating daily actuals it hits limits on change history, concurrent editing, and master data integrity. The realistic migration path is a division of labour: accumulate actuals in a system, and do analysis and manipulation in Excel or BI. You do not have to abandon Excel.
Summary
The most important thing in selecting a manufacturing cost management system is not comparing calculation features. It is designing shop floor data collection. Total manufacturing cost will always come out of the accounting system, but cost by job order and by process only becomes calculable once five data points — material issues, work time, good quantity, defect quantity, and changeover time — are linked to work orders. Establishing which of those five you are missing is the starting point of the whole evaluation.
Systems cover three different areas: the costing module in accounting/ERP (Type A), costing built into a production management system (Type B), and a dedicated costing package plus BI (Type C). If the goal is consistency with the closed books, A; if it is shop floor improvement and pricing, B; if you already have data and analysis is the bottleneck, C. Trying to satisfy all of it with one system is how requirements balloon.
Estimate cost across five layers — licence, initial build and master data, data collection infrastructure, system integration, and operation and maintenance — and take particular care not to under-estimate master data preparation and internal operating effort. Phase the implementation with the first 90 days targeting a real understanding of material cost and the six-month mark targeting daily actual cost by job order.
And in Thailand there are two issues that touch the structure of the cost data itself: preparing for transfer pricing documentation (generally within 60 days of a request) and supporting BOI incentive segmentation. 2026 is a year in which the FTI is forecasting GDP growth of 1.6–2.0% with capacity utilisation below 60% in several sectors, while electricity tariffs and social security burdens both rise. When utilisation falls, fixed cost absorption breaks, and a plant that cannot separate out the volume variance loses the ability to identify why margins deteriorated. The practical value of higher cost resolution is greatest in exactly this kind of environment.
Whether you want to change how cost becomes visible, or you are still at the stage of deciding where to start, either is a fine place to begin a conversation. TOMAS TECH is based in Bangkok and supports Japanese manufacturers across Thailand and ASEAN with production management, shop floor data collection, and cost visibility. We are happy to start with something as simple as reviewing how far your current actual data can be trusted, and which layer would give you the fastest result. This is not a product pitch — the first step is taking stock of where you are. Get in touch through our contact form whenever it is useful.
