Inventory Management System Comparison: ASEAN Plants 2026

Most inventory management system comparison exercises start the wrong way round. Someone opens a spreadsheet, writes vendor names down the left-hand column, features across the top, and starts collecting ticks. Three months and six demos later, the decision is still made on gut feel, because nobody agreed on what the plant was actually trying to fix. If you run a factory in Thailand or Vietnam, or you carry regional IT responsibility across several ASEAN sites, what you need first is not a product list. You need a measuring stick of your own.

This article gives you three of them. The first sorts the market into four system types so you stop comparing things that are not comparable. The second breaks cost into five layers, because a licence fee is rarely more than a fifth of what you will actually spend. The third splits inventory discrepancy into six root causes and states plainly which of them software can fix and which it cannot. Around those three axes we cover the issues that only bite at ASEAN plants: multilingual master data, multi-currency receipts, duty-free material reconciliation, statutory labour cost movements, the shift to 2D barcodes, and a 90-day rollout sequence that survives contact with a real warehouse.

One position up front, because it shapes everything that follows. Installing an inventory management system will not make your inventory discrepancy disappear. Roughly half the causes can be structurally suppressed by software. The other half only respond to operating rules and clear ownership. Skip that distinction and you land in the worst possible outcome: an expensive system whose only visible effect is more data entry for the shop floor. The second half of this article shows how to draw the line before you sign anything.

What to Decide Before You Build an Inventory Management System Comparison Table

Three things need internal agreement before you talk to a single vendor. Without them, you will receive quotations built on three different sets of assumptions, and the comparison itself becomes meaningless.

Decision 1: which inventory, at which granularity

A factory does not have “inventory”. It has raw materials, purchased components, work in process, finished goods, maintenance spares, indirect consumables, and tooling and moulds. Those seven things behave completely differently, and squeezing them into one word is where most scope creep begins.

Granularity is the second half of the same decision:

• Reconcile at item level, or track down to lot and serial level?
• Reconcile at warehouse level, or down to bin location?
• Is it enough to be correct at month-end close, or does the balance need to be correct in real time?

Choosing lot level plus bin level plus real time multiplies the number of shop-floor scan events several times over. That is a legitimate choice, but it is only legitimate if you also fund the handheld terminals and the training hours that make it possible. Conversely, if “item level, correct at month-end” genuinely covers your reporting obligations, your investment drops sharply and your rollout gets much shorter.

Decision 2: measure your current inventory accuracy

You cannot argue effect without a baseline. Inventory accuracy is normally measured as the percentage of counted items whose physical quantity matches the system quantity.

Industry benchmarks put 97% inventory accuracy as the minimum acceptable level, with best-in-class operations at 99% or above. At the same time, surveys of plants already running barcode-scanner WMS operations report real-world performance in the 95% to 98% band (source: Factory AI, guide to cycle counting).

Those percentages become far more useful when you convert them into item counts. Take a plant managing 5,000 active SKUs:

• 95% inventory accuracy: 250 items do not match
• 97% inventory accuracy: 150 items do not match
• 99% inventory accuracy: 50 items do not match

Moving from 97% to 99% therefore means taking the mismatch list from 150 items down to 50, a reduction of 100 items. That is a target a warehouse supervisor can act on, and it is far easier to report to group headquarters than “improve inventory management”.

If you have no baseline at all, pull 100 to 200 items and count them before you go any further. You do not need to redo a full physical inventory to get a usable starting number.

Decision 3: draw the line between system fixes and process fixes

This is the decision that determines whether the project is judged a success. Some causes of inventory discrepancy are structurally preventable by software. Others are only preventable by rules and accountability. Expect the system to handle both and you will be disappointed.

The macro backdrop in Thailand makes that honesty commercially necessary. The Manufacturing Production Index published by Thailand’s Office of Industrial Economics fell 3.10% year on year in June 2026, the steepest drop since November of the previous year. The second quarter of 2026 was down 1.79%, with average capacity utilisation for the quarter at 57.47%. Reduced output in automotive and petroleum-related sectors was cited as the main driver (source: Xinhua, 27 July 2026). Note that the single-month June decline was 1.31 percentage points worse than the quarterly average.

Capacity utilisation of 57.47% is another way of saying that 42.53% of installed capacity is idle. When utilisation falls, work in process and slow-moving component stock accumulate, which is precisely when inventory visibility pays for itself. It is also precisely when capital requests get the hardest scrutiny. If you promise that a system will eliminate discrepancies and it does not, the next request will not be approved.

Inventory Management System Comparison: The Four Types

“Inventory management system” covers four categories with genuinely different characters. Put coverage on the horizontal axis, from warehouse-only at one end to procurement-through-shipping at the other, and customisation depth on the vertical axis. The four types separate cleanly.

Type A: cloud inventory management (SaaS)

A subscription service focused on receipts, issues and stock balances. Most run in a browser and on a smartphone, and you can be entering transactions within days of signing.

The strengths are speed to value and a light entry cost. The weakness is manufacturing depth: bill-of-materials driven requirement calculation, work in process by operation, and lot or serial traceability are either out of scope or handled superficially. Assume you cannot customise it, which means your process bends to the product rather than the reverse. For a plant whose only real problem is that nobody trusts the stock balance, that trade is often worth making.

Type B: WMS (warehouse management system)

A WMS exists to optimise work inside the four walls of the warehouse: location management, receiving inspection, put-away and picking instructions, enforced FIFO management by lot, and task confirmation through handheld terminals.

This is the strongest answer to warehouse-origin problems, the classic pair being “the system says we have it but nobody can find it” and “old lots keep getting stranded at the back of the rack”. What a WMS does not do is connect to production orders and operation results, so work in process and cost accounting stay outside its reach. We look at the wider warehouse and distribution picture in our article on logistics DX and WMS practice in Southeast Asia.

The WMS market itself is growing quickly. Mordor Intelligence estimates the market at USD 4.04 billion in 2025 rising to USD 4.77 billion in 2026, with a compound annual growth rate of 17.98% for 2026 to 2031. The implied 2025-to-2026 increase is about 18.1%, essentially in line with that CAGR, so the near-term estimate and the long-term rate tell a consistent story. Extending 17.98% mechanically across five years gives a multiplier of roughly 2.29 times, which would put the 2031 market at about USD 10.9 billion. Cloud deployment is the stated growth engine.

Type C: the inventory module of a production management system or ERP

Here inventory lives inside a wider system. Bills of materials, requirement calculation, purchase orders, operation results, cost accounting and stock all sit on one database, so every stock movement carries a business reason with it.

The decisive advantage is root-cause tracing: you can follow a discrepancy back to the transaction and the process that created it. The disadvantage is weight. Master data scope expands, more departments have to agree, and decisions take longer. We approach the same territory from the production side in production management system rollouts at Thai factories, which is worth reading in parallel if cost and inventory have to reconcile.

Type D: custom build

Built to your specification. The upside is that it can mirror your existing paper forms and unusual units of measure exactly, which sometimes produces the least shop-floor resistance of any option.

The risk sits after go-live: maintenance, and handover when the people involved rotate out. At an overseas site, the practical question is whether your relationship with the development vendor will outlast the project. If source code ownership and documentation obligations are not written into the contract, you may be looking at an untouchable black box in three years, at exactly the moment a new regional standard arrives.

Four-type comparison table

The go-live durations reflect what we have seen on projects in Thailand and Vietnam, and they move substantially with SKU count, number of sites, and the number of interfaces to existing systems. Treat them as a planning starting point, not a commitment.

One warning about how to read that table: wider coverage is not better. If your dominant problem is that material cannot be found in the rack, an ERP inventory module with weak location handling will not solve it. If regional HQ is questioning the gap between inventory value in the accounts and physical stock, no amount of WMS polish will answer the question. The real work of an inventory management system comparison is matching where your problem lives to where each type is strong.

Inventory Management Software Cost: Breaking It Into Five Layers

Answer “what does an inventory management system cost” with a licence figure and you will have an uncomfortable conversation later. Real spend splits into five layers, and the licence is one of them.

The five cost layers

Those shares are set to total 100% for a mid-case project, so they are internally consistent by construction. On that basis, total cost of ownership is about five times the licence layer (100 divided by 20 equals 5).

Across the range of projects we have been involved in locally, each layer tends to fall within these bands when the licence layer is indexed to 1:

• Layer 2 hardware: 0.3 to 1.0
• Layer 3 master data and physical count: 0.5 to 1.5
• Layer 4 integration: 0.5 to 2.0
• Layer 5 first-year operation and training: 0.3 to 0.8

Adding all the low ends gives 1 + 0.3 + 0.5 + 0.5 + 0.3 = 2.6. Adding all the high ends gives 1 + 1.0 + 1.5 + 2.0 + 0.8 = 6.3. So total spend runs between 2.6 and 6.3 times the licence layer, and budgeting 3 to 6 times is the practical rule. The mid-case multiple of 5 sits inside that band, which is the arithmetic check you want before you put a number in front of a finance committee.

When you compare quotations, establish which of the five layers each one includes. A vendor quoting licence only and a vendor quoting licence plus master data plus integration are not describing the same project, and putting their totals side by side tells you nothing.

How to use published price ranges without misleading anyone

Published inventory management software cost ranges are almost always drawn from a single national market, and the Japanese market is one of the few with widely circulated figures. Use it as a reference point for structure, not as a local price list.

As a Japanese market reference point, cloud products are described as costing from zero up to a few tens of thousands of Japanese yen to set up, plus roughly JPY 20,000 to JPY 50,000 per month, while on-premise deployments are described as JPY 1 million to JPY 10 million up front plus roughly JPY 100,000 to JPY 400,000 per year (source: System Kanji). All figures in this paragraph are Japanese yen. Converting those ranges into three-year and five-year totals, still in Japanese yen, changes how the comparison looks:

Cloud. Annual subscription is JPY 240,000 at the low end (JPY 20,000 x 12) and JPY 600,000 at the high end (JPY 50,000 x 12). Three-year subscription total is therefore JPY 720,000 to JPY 1.8 million, and five-year is JPY 1.2 million to JPY 3 million, plus a small setup fee.

On-premise. Three-year total runs from JPY 1.3 million (JPY 1 million up front plus JPY 100,000 x 3) to JPY 11.2 million (JPY 10 million up front plus JPY 400,000 x 3). Over five years it runs from JPY 1.5 million (JPY 1 million plus JPY 100,000 x 5) to JPY 12 million (JPY 10 million plus JPY 400,000 x 5).

At the five-year mark, the cloud high end of JPY 3 million compares with an on-premise high end of JPY 12 million, a factor of four. But compare the low ends and you get JPY 1.2 million against JPY 1.5 million, which is close to indistinguishable. The correct reading is not “cloud is cheap and on-premise is expensive”. It is that on-premise has a far wider upside tail, and your job is to work out which end of that tail your requirements sit in.

Do not present those Japanese yen ranges to your board as the going rate in Thailand or Vietnam, and do not invent an exchange rate to make them look local. Local reality depends on the size of the entity, whether you need support in a specific language, whether the vendor has an on-the-ground team in Bangkok, Ho Chi Minh City or Hanoi, and how much of layers 3 and 4 you outsource. The transferable part of the reference point is the shape of the cost curve, not the amounts. Build your own numbers in THB or VND from local quotations, express them in your reporting currency with an explicit note that any conversion is approximate, and apply the 3-to-6-times licence multiple as your sanity check.

Layers 3 and 5 are the two that inflate most reliably at ASEAN sites. Item masters are maintained twice, once at group HQ and once locally, and the two have drifted. Units of measure are mixed across the plant, with the same part ordered by carton, issued by piece and counted by weight. Work instructions have to exist in the group language, in English, and in Thai or Vietnamese, which triples the documentation effort before anyone has scanned a single label. For the hardware and counting side specifically, our article on the cost of RFID-based factory stocktaking works through device costs in more detail.

Building the benefit side of the case

Cost alone never gets approved. You need a benefit estimate, and inventing a number is the fastest way to lose credibility six months later. The approach we recommend is one simple model with its assumptions stated openly.

The following figures are illustrative, not from a real project. Take a plant holding THB 50 million of inventory and turning it 6 times a year. Annual issue value is 50 million x 6 = THB 300 million. If turnover improved to 8 times a year, the inventory required to support the same throughput would be 300 million divided by 8 = THB 37.5 million, which is THB 12.5 million less than today.

The point of that calculation is not to promise a THB 12.5 million release of working capital. It is to establish what one turn of inventory is worth in your own currency and at your own scale, so that the investment and the benefit can be discussed in the same order of magnitude. Substitute your own inventory value and turnover and the conversation with finance changes character.

A second, smaller calculation makes the operational benefit concrete. If 20 people spend 2 days on a physical inventory, that is 40 person-days. Valued at the 400 baht per day minimum wage that applies in Bangkok and the eastern industrial provinces, purely for illustration, one count costs 16,000 baht in direct labour, and four counts a year cost 64,000 baht. Actual wages are higher than the statutory minimum, and this arithmetic excludes overtime and holiday premiums, the time of supervisors and finance staff who attend, and the opportunity cost of suspending shipments during the count. Present it as a floor, clearly labelled as such, and it will hold up under questioning.

Why Inventory Discrepancy Never Fully Disappears: Six Root Causes

This is the heart of the article. Six causes, each with an honest statement of how much software can do about it.

The three that software genuinely suppresses

Unit of measure conversion errors (cause 3) are where systems shine. Define a base unit and conversion factors once in the item master, allow ordering by carton and issuing by piece, and let the system normalise internally. Mental arithmetic errors disappear. Note the condition attached: this only works if the master data is correct, which lands squarely in cost layer 3.

No location definition (cause 6) is the same story. Assign bin locations, record the bin at put-away, direct the picker to the bin, and “it should be here somewhere” shrinks dramatically. This is the single area where a handheld-driven WMS earns its keep fastest.

Issue postings done later (cause 2) sits on the boundary between system and process. Put a terminal at the issue point so that moving the material and recording the movement are one action, and accuracy improves sharply. Replace the software but keep the practice of batch-entering from an office PC in the afternoon and nothing changes. Terminal placement and terminal count effectively determine how much this cause improves. Saving money here undermines the entire investment, and it is the most common false economy we see.

The three that only operating rules fix

Informal emergency withdrawals (cause 4) cannot be blocked by software. When a line is about to stop, someone will take the part from the rack and write the paperwork later, or not at all. This behaviour comes from good intentions, so punishing it simply drives it underground. What works is accepting that emergency withdrawals will happen and providing a fast, low-friction way to record them: a dedicated document series, a rule that it must be converted to a formal transaction within 24 hours, and a named owner who checks the open list daily.

Scrap and rejects not booked (cause 5) is an organisational design question disguised as a data problem. Asking the operator who produced the defect to raise the scrap document is psychologically difficult, so material is physically discarded while the book balance survives. The workable pattern is to move the booking to quality assurance on a fixed cycle, so the person raising the document is not the person being measured by it.

Receiving quantity not verified (cause 1) can be partly supported by software, through matching against the purchase order and alerting on differences, but no system can detect a quantity nobody counted. Full inspection or sampling, and at what percentage, is a policy your own team has to set and publish.

What cycle counting is actually for

Cycle counting means counting a rotating subset of inventory continuously, alongside or instead of the annual or quarterly wall-to-wall count, without stopping operations.

The purpose is widely misunderstood. Cycle counting exists not to correct numbers but to identify the cause of discrepancies (source: Factory AI). If correcting numbers were the goal, one annual full count would be enough, but a full count tells you nothing about why. The value comes from what you do in the hour after a variance is found: trace the recent receipts and issues for that item and classify the variance against the six causes above. Repeat weekly and structural improvement becomes possible, because you finally know which cause dominates in your plant.

Put differently, cycle counting is a measurement instrument, not an accuracy programme. Running it for two or three weeks before you shortlist vendors is the cheapest way to make your comparison criteria concrete.

Signs Your Excel Inventory Management Has Hit Its Limits

Excel inventory management is not a mistake. With a modest SKU count, a stable owner and a single site, it works. The difficulty is that the limits are crossed quietly. If three or more of the following are true at your plant, it is time to look at alternatives.

1. Multiple versions of the same file exist. Filenames like “latest”, “latest_2” and “latest_revised” sit in the same folder and nobody can say which numbers are authoritative.
2. Month-end close takes three days or more. By the time the figures are agreed, they describe a warehouse that no longer exists.
3. Broken formulas go unnoticed. A row insertion shifts a reference range and the totals stop adding up, which surfaces at month-end if at all.
4. Reorder points live in someone’s memory. “We order this part when it drops below about 200” is knowledge held by one person, and stockouts follow their annual leave.
5. Inventory value does not agree with the accounting balance. Operations and finance hold different numbers and nobody can explain the gap.
6. Only one person can touch the file. Macros and nested formulas have made it unmaintainable by anyone but the author.
7. The file has become slow. Opening takes noticeable time and filtering freezes, which typically starts a few thousand rows in.
8. Multi-site stock cannot be consolidated. Each site keeps its own file and any group-level inventory figure requires manual aggregation.

An important caveat: leaving Excel does not automatically remove these problems. Item 4 in particular survives the migration untouched unless somebody actively enters reorder points and safety stock values into the new system. No product decides your optimal inventory for you. A named person has to derive the numbers from lead time and demand variability, and revisit them on a schedule. Going live without an owner for that task is one of the most common ways a good system quietly stops reflecting reality.

Five Issues Specific to Factory Inventory Management in Thailand and Vietnam

These are the points that domestic comparison articles never raise and that regional operations cannot avoid.

Issue 1: a multilingual UI and multilingual master data are different problems

Switching the screen language is a product feature. Having item names appear in Thai or Vietnamese is a master data project. Decide whether the item master will carry separate fields for the group HQ language, English and the local language, and if so, who does the translating and who approves it. Translating several thousand item names is real effort and it lands in cost layer 3, usually unbudgeted.

There is a further wrinkle in Thailand, where a large share of warehouse operators may speak Burmese or Khmer as a first language rather than Thai. Adding a fourth and fifth screen language is rarely the efficient answer. Reducing the amount of text on screen is: lean on codes, colour and photographs so that identification does not depend on reading. Attaching a single photograph to each item master record has cut mis-picks visibly on projects we have supported, at a fraction of the cost of another translation pass.

Issue 2: multi-currency receipts and inventory valuation

Purchasing in a mix of Thai baht, US dollars and the group currency is normal for a manufacturing site in the region, and Vietnamese entities frequently buy in USD while reporting in VND. Decide which currency and which rate the receipt unit cost is held at, and how monthly revaluation is handled. This is as much an accounting policy question as a system question. If the inventory system can only hold a single currency, a structural gap with the general ledger is guaranteed, and it will be discovered during an audit rather than during the project. Confirm during the comparison stage that the product can hold receipt costs in multiple currencies and that rate application rules are configurable.

Issue 3: duty-free imported material under investment incentives

If you import raw materials duty-free under Thailand Board of Investment privileges, the quantity balance you report has to reconcile with the material physically present. Many plants end up maintaining a second quantity view for BOI reporting alongside the internal stock ledger, and how the system handles that dual view is a question to settle early rather than after go-live. When a physical count disagrees with a reported balance, it stops being an internal matter. The same principle applies to Vietnamese export-processing enterprises operating under bonded material regimes: the customs-facing material balance and the operational stock ledger have to be reconcilable on demand, ideally from one data source rather than two spreadsheets maintained by different departments.

Investment appetite in Thailand remains strong despite the soft production numbers. Applications to the Thai BOI in the first half of 2026 reached 1,299 projects worth THB 1.473 trillion, up 37% year on year. Digital industries accounted for THB 1.115 trillion across 90 projects, roughly 76% of total application value, with electronics and electrical products at THB 120.23 billion across 179 projects, about 8%. Foreign direct investment came to 877 projects worth THB 1.368 trillion, up 80%, equivalent to about 93% of total application value (source: The Nation).

The measure most relevant to factory IT spending is the Smart and Sustainable Industry package, which recorded 132 applications worth THB 17.158 billion. That is only about 1.2% of total application value, but it works out to roughly THB 130 million per project (THB 17.158 billion divided by 132 applications). For context, the average across all applications is about THB 1.13 billion per project, and digital industries alone average roughly THB 12.4 billion per project. Large data centre projects are clearly inflating those averages, so do not use the headline average as a benchmark for your own investment size. The per-project figure under Smart and Sustainable Industry, which covers machinery replacement, energy efficiency, digital technology adoption, and automation and robotics, is the more relevant order of magnitude.

Issue 4: can you actually use the depa 200% deduction

In Thailand, Royal Decree No. 802 (B.E. 2569) was gazetted on 6 February 2026, granting a 200% deduction for the purchase, commissioning or subscription of software, hardware, smart devices and digital services registered in depa’s Thailand Digital Catalog. The cap is THB 300,000 and the spending deadline is 31 December 2027 (source: Mahanakorn Partners).

The eligibility conditions matter more than the headline. The measure targets SMEs with paid-up capital of THB 5 million or less and revenue of THB 30 million or less, and general-purpose laptops and desktop PCs are excluded. Manufacturing subsidiaries of foreign groups in Thailand usually exceed both thresholds comfortably, so the honest answer for most readers of this article is that the deduction is not available to them.

Even where it is available, a THB 300,000 cap is small relative to an inventory system project. Outsourced work in cost layer 3 alone will typically consume the whole allowance. Do not build the investment case around this incentive. If you think you might qualify, confirm it with a tax adviser on your specific numbers rather than on a summary like this one.

Issue 5: GS1 Sunrise 2027 and the move to 2D codes

This one affects label design, and label design is expensive to redo. GS1’s Sunrise 2027 initiative aims for retail point-of-sale systems to be able to read both traditional one-dimensional barcodes and GS1-compliant two-dimensional codes, meaning GS1 DataMatrix and GS1 Digital Link QR codes, by the end of 2027. During the transition, labels carry both 1D and 2D symbols, and from 2028 onward operating with 2D alone becomes possible. Companies representing 48 countries and approximately 88% of world GDP are reported to be preparing (source: GS1 US).

Labels used purely inside your own plant can stay in whatever format you like. Labels that travel to customers or into retail channels cannot. If you are selecting an inventory management system now, confirm that its label printing function supports 2D symbologies and can produce a dual 1D-plus-2D layout, and you avoid a rebuild in two or three years. There is an operational bonus too: a 2D code carries far more data in the same area, so lot number and expiry date can live in one symbol instead of three separate fields. For the wider move away from paper on the shop floor, see our article on electronic forms and the paperless factory.

Reading Vietnam differently from Thailand

If your responsibility spans both countries, resist the temptation to copy one specification to the other, because the two are in different phases of the cycle.

Vietnam’s manufacturing PMI was 51.8 in June 2026, down 1.0 point from 52.8 in May but still 1.8 points above the 50.0 no-change mark. Output expanded for the fourteenth consecutive month, and June’s pace of output growth was the fastest since February. Manufacturing value added for the first half of 2026 grew 9.97%, close to 10% (source: VietnamPlus).

Inventory management fails differently in growth than in contraction. In a contracting plant, the pain is slow-moving stock, obsolescence and write-downs, so the priority is visibility of ageing inventory and a documented rule for classifying long-held items. In an expanding plant, the pain is stockouts that halt a line and the freight premiums caused by emergency reordering, so the priority is reorder point management, safety stock accuracy and supplier lead-time discipline. Reading the Thai MPI figures alongside the Vietnamese PMI figures above, a Thai site and a Vietnamese site inside the same group can rationally rank the same feature list in a different order. Say so explicitly in your specification document, or someone at group headquarters will ask why the two proposals differ.

Parts, Spares and WIP: Requirements Change by Inventory Type

Returning to Decision 1 with specifics. Requirements differ sharply by inventory type, and forcing one method onto all of them guarantees friction somewhere.

Raw materials and purchased components revolve around reorder point systems and safety stock. Imported items have long and variable lead times, so you need a documented basis for the safety stock figure rather than a habit. Whether lot control and enforced FIFO management are required depends on your industry and your customers’ demands.

Work in process is the hardest thing to handle with an inventory system alone. Without a link to operation confirmations you cannot know WIP quantities, which puts this in production management or MES territory. It is also the first wall that plants hit after implementing a standalone inventory product. When someone says “we want to see WIP too”, read that as a signal to evaluate Type C.

Finished goods need allocation against shipping instructions and two-way lot tracing: from the customer back to the lot, and from a lot forward to every customer who received it. Because that capability determines how wide a recall has to be, design it together with your traceability requirements rather than after them. Our article on the cost of building a traceability system covers that ground.

Maintenance spares (MRO) are the most awkward inventory of all. Demand is sporadic so statistical reorder points do not work, unit values are high, turnover is low, and you cannot reduce holdings much because the cost of a stopped machine dwarfs the carrying cost. For this class, knowing where the part is usually beats knowing exactly how many there are, so bin location discipline and physical labelling take priority over quantity precision.

Indirect consumables invert the logic: they are frequently cheaper to manage loosely. Booking every glove, rag and screw in and out costs more in administration than the inventory is worth. Point-of-use storage with kanban replenishment and monthly expensing of purchases is the realistic answer. Declaring at project kick-off that “all items will be registered in the system” quietly removes your ability to make this call.

Tooling and moulds are about custody, not quantity. Who took it, and where is it now. RFID suits this use case well, though tag selection for metal surfaces and read-environment design carry their own difficulties.

The right-hand column is the most valuable one in practice. Unless you explicitly decide what you are not going to manage in the system, implementation always lands on the shop floor as additional data entry, and adoption suffers for reasons that have nothing to do with the software.

A 90-Day Rollout Roadmap for a Cloud Inventory or WMS Project

The first task in an inventory management system project is not selecting a system. It is counting physical stock. The three-phase, 90-day sequence below assumes Type A (cloud inventory management) or Type B (WMS).

If you are implementing Type C across the board, 90 days is not realistic and claiming otherwise damages your credibility. The workable framing is to run the inventory scope live first and treat these 90 days as stage one of a longer programme. Make clear to group headquarters early that a full ERP replacement is not a three-month exercise.

Phase 1 (days 1 to 30): count physical stock and establish the baseline

Tasks

• Count the target items. It does not need to be wall-to-wall; start with high-value items and known trouble items.
• Identify duplicate and obsolete records in the item master.
• Survey units of measure as they actually are: which items are ordered, issued and counted in which unit.
• Start weekly cycle counting and log every variance against the six root causes.
• Measure and record current inventory accuracy.

The goal of this phase is to hold real numbers before you choose a system. Once you know your accuracy percentage and the distribution of variance causes, the required feature set narrows on its own.

The classic mistake is skipping this phase and going straight to vendor selection. Without a baseline you cannot judge whether a proposal addresses your problem, and you have no way to prove improvement afterwards.

Phase 2 (days 31 to 60): define locations and clean up master data

Tasks

• Decide the bin numbering rules and physically label the racks.
• Clean up the item master, defining base units and conversion factors.
• Design the shop-floor workflow: how many terminals, and exactly where.
• Collect reporting requirements, including any format group headquarters needs.
• Document the variance correction flow: who investigates, who approves the adjustment.
• Build a multilingual glossary mapping each term across the group language, English and the local language.

This is the least glamorous and most decisive phase. Bin numbering rules are hard to change once labels are on racks, so allow for expansion. Aisle, rack, level and position is the standard four-tier form, but simplify it if the warehouse is small. Over-engineering the scheme so that relabelling never catches up is a worse failure than a scheme that is slightly too coarse.

The glossary is routinely dismissed as a nicety. Without it, the term used on the floor and the term used in the monthly report do not match, and every enquiry turns into a translation exercise. A table of 30 to 50 terms removes a surprising amount of post-go-live confusion, and it is the artefact your successor will thank you for.

Phase 3 (days 61 to 90): from pilot to go-live

Tasks

• Run a pilot limited to one warehouse or one item group.
• Keep the existing method running in parallel during the pilot and compare variances daily.
• Take improvement requests from the floor and explicitly split them into in-scope and out-of-scope.
• Train. Cut the words in work instructions and lead with photographs and screenshots.
• Set the go-live criteria in advance, for example inventory accuracy on pilot items at or above target for two consecutive weeks.
• Announce the go-live date and, just as importantly, the date the old method stops.

A pilot that ends with “if it goes well we will switch over” produces permanent parallel running. Deciding the acceptance criteria and the shutdown date for the old process in advance is what actually gets a plant live.

The allocation is 30 + 30 + 30 = 90 days, but phase 2 master data work and rack labelling routinely exceed estimates. When that happens, extend phase 2 and protect phase 3. Compressing training and parallel running to hold a date lengthens the post-go-live disruption instead. A programme that slips from 90 days to 120 days is far cheaper than one that goes live on schedule and then produces unexplained variances for six months.

Ten Failure Patterns and How to Avoid Them

Patterns we see repeatedly on projects in the region.

1. Going live with master data unfinished. “We will fix it as we go” means transactions accumulate against wrong master records, and the cost of correction grows daily. Duplicate item records and unit definitions must be closed before go-live, without exception.

2. Loading book quantities as opening stock instead of counting. You destroy the baseline you need to demonstrate any improvement. High-value items must be counted before opening balances are loaded.

3. Starting with a single location called “Warehouse 1”. Adding bin locations later means doing it with live transactions in flight, which is more work than doing it up front. At minimum, fix the bin code structure before go-live.

4. Not buying enough handheld terminals. Buying one device to “try it out” creates queues, queues create deferred entry, and deferred entry is root cause 2 preserved in amber. Budget the real number plus spares from the start.

5. Designing to the reporting requirements of head office. Satisfying every analysis dimension regional HQ might want adds fields to shop-floor screens. For every field, ask who enters it, when, and on which screen. Any field without an identifiable owner will be blank.

6. Cutting over all items and all warehouses at once. You leave yourself nowhere to fall back to. Warehouse by warehouse, or item group by item group, usually finishes sooner in practice.

7. Having no variance correction flow. If nobody owns investigation and nobody is authorised to adjust the book, variance reports pile up unread. Set the value threshold too: below this amount the supervisor corrects it, above it the controller approves.

8. Localising the screens but skipping the glossary. A Thai or Vietnamese interface does not reduce communication cost if the floor and the office use different words for the same part.

9. Not naming an owner for reorder point reviews. Reorder points and safety stock drift as demand and lead times change. Without a named owner and a review cadence, the values fossilise at their go-live settings and the system starts recommending nonsense.

10. Assuming poor adoption is a training problem. When we are told “the operators are not entering data”, the cause is usually workflow or language rather than training. Workflow means the terminal is too far from where the material moves, so recording cannot be part of the physical task; the only fixes are more terminals or better placement, and more training will not help. Language means the on-screen wording differs from what the floor actually says, which the glossary fixes by aligning system labels to floor usage. If it is neither, there are probably too many mandatory fields, so go through them one at a time asking who enters this and when, and delete the ones nobody can answer for.

How to Build the Business Case for Regional HQ

At overseas sites, the obstacle to IT investment is rarely technical. It is explanatory. Building the pack in this order makes the decision discussion much easier.

1. Current inventory accuracy and the breakdown of variance causes. Use the measured figures from phase 1. “Accuracy is 95% today, and 40% of variances are late postings, 30% unit conversion errors” starts the conversation with facts instead of opinions.
2. The split between what the system fixes and what the operation fixes. The six-cause table works as a slide almost unchanged. Having it early resets the expectation that software solves everything.
3. Which of the four types you are choosing, and why. The trick is to document why you rejected the others. “We chose the ERP inventory module rather than a WMS because what group headquarters is asking about is the reconciliation between cost and stock” is a defensible sentence.
4. The five cost layers. Show all five, and separate internal effort from external spend. If you can explain why the total is 3 to 6 times the licence layer, you avoid the supplementary budget request that damages credibility later.
5. The benefit estimate and its assumptions. Working capital released by improved turnover, physical inventory hours saved, line stoppages avoided. State each assumption. Visible assumptions earn more trust than clean-looking numbers.
6. The 90-day roadmap and the acceptance criteria. What finishes when, and what conditions trigger go-live.

Statutory cost trends in Thailand are worth a slide of their own, because they change how the labour-saving argument lands. The Thai minimum wage rose to 400 baht per day for all sectors in Bangkok with effect from 1 July 2025, up from 372 baht, a difference of 28 baht or about 7.5%. Four provinces and one district in the manufacturing belt, including Chonburi and Rayong, had already moved to 400 baht from January 2025. Rates were held through 2026, leaving a national range of 337 to 400 baht per day (source: JETRO). A 28 baht daily difference works out to 728 baht per person per month on a 26-working-day basis.

Social security changed as well. Under a ministerial regulation gazetted on 12 December 2025, the maximum wage base for social security contributions rose from 15,000 baht to 17,500 baht per month from 1 January 2026, with the floor at 1,650 baht. That is an increase of 2,500 baht in the base, about 16.7%. The monthly contribution ceiling rose from 750 baht to 875 baht for each of employer and employee. Both figures represent 5% of the respective base (750 divided by 15,000 equals 5%, and 875 divided by 17,500 equals 5%), so the contribution rate itself is unchanged. The base is scheduled to rise further to 20,000 baht from January 2029 and 23,000 baht from January 2032 (source: JETRO).

The employer-side increase is therefore 125 baht per person per month, or 1,500 baht per person per year. With 200 affected employees, that is an additional 300,000 baht per year. One caveat prevents a common overstatement: the full increase only applies to employees earning 17,500 baht per month or more. A worker on 400 baht per day earns 10,400 baht over 26 working days, which does not even reach the previous 15,000 baht ceiling. This increase therefore bites mainly at administrative, engineering and management grades, so do not compute it across your direct headcount.

There is an instructive coincidence in those numbers. The 300,000 baht annual increase calculated above is almost exactly the 300,000 baht cap on the depa 200% deduction. One is a recurring increase in fixed cost; the other is a one-off allowance available to a narrow set of companies for a limited period. That contrast is another reason not to place a tax incentive at the centre of an investment decision.

In an environment of rising statutory labour cost, a proposal that adds headcount to manage inventory is a hard sell. The corollary is that a proposal built on reducing counting hours and data entry hours lands better now than it would have two years ago, provided your baseline numbers are real.

Frequently Asked Questions

How much does an inventory management system cost?

Looking at licence fees alone will mislead you. Judge the total across the five layers set out above (licence, hardware, master data, integration, first-year operation), and expect the total to run 3 to 6 times the licence layer. As a Japanese market reference point, published ranges describe cloud products at zero to a few tens of thousands of Japanese yen to set up plus roughly JPY 20,000 to JPY 50,000 per month, and on-premise deployments at JPY 1 million to JPY 10 million up front plus roughly JPY 100,000 to JPY 400,000 per year (source: System Kanji). Those are Japanese yen figures for the Japanese market. Your effective cost in Thailand or Vietnam depends on language support requirements, whether the vendor has a local delivery team, and how much of the master data and integration work you outsource, so build the local number from local quotations and mark any currency conversion as approximate. When comparing quotations, first align which of the five layers each one covers.

What is the difference between a WMS and an inventory management system?

Coverage. A WMS optimises work inside the warehouse: location management, receiving inspection, picking instructions, enforced FIFO, and task confirmation on handheld terminals. A general inventory management system covers receipts, issues and balances but does not direct warehouse work. A useful rule of thumb: if the dominant symptoms are “we cannot find it” and “old lots get stranded”, lean WMS; if the dominant symptom is “the quantity itself is wrong”, a plain inventory system is often enough. Neither handles work in process or operation confirmations, so if you need those, you are looking at a production management system.

When is the right time to move from Excel to an inventory management system?

When three or more of the eight signs listed above apply. Three of them justify moving quickly: multiple versions of the same file, a month-end close taking three days or more, and reorder points that exist only in one person’s memory. Conversely, with a small SKU count, one stable owner and a single site, staying on Excel can genuinely be the lower total cost option. If you do decide to move, do not try to reproduce your Excel layout in the new system. Excel is designed around one person seeing everything; a system is designed around several people each seeing their part, and the structures are not compatible.

Can inventory discrepancy be reduced to zero?

Realistically, no. Benchmarks put 97% inventory accuracy as the minimum acceptable level and 99% or above as best-in-class, while surveys of plants running scanner-based WMS operations report 95% to 98% in practice (source: Factory AI). With 5,000 managed items, even 99% accuracy leaves 50 items that do not match. The target should not be zero but a state where every variance has an explainable cause. Once causes are known, you can prioritise by value and attack the expensive ones. Reducing the number of unexplained variances is worth more than chasing decimal places.

Should we choose cloud or on-premise inventory management?

Three criteria decide it. First, plant network reliability: on an industrial estate with unstable connectivity, shop-floor work must not stop when the link drops, which a cloud product can handle if the device caches transactions locally. Second, the number of interfaces: if you need many connections to production equipment and finance systems, on-premise sometimes assembles more naturally. Third, local IT capability: with nobody on site who can administer a server, cloud is the safer option. On cost, applying the Japanese market reference ranges over five years gives roughly JPY 1.2 million to JPY 3 million for cloud subscriptions plus setup, against JPY 1.5 million to JPY 12 million for on-premise, a factor of four at the top end and near parity at the bottom.

Can a foreign-owned plant in Thailand use the depa 200% deduction?

Usually not. Royal Decree No. 802 (B.E. 2569), gazetted on 6 February 2026, allows a 200% deduction on software, hardware, smart devices and digital services registered in depa’s Thailand Digital Catalog, but eligibility is limited to SMEs with paid-up capital of THB 5 million or less and revenue of THB 30 million or less (source: Mahanakorn Partners). Most manufacturing subsidiaries of foreign groups exceed both thresholds. The cap is also only THB 300,000, which is small against an inventory system project, general-purpose laptops and desktops are excluded, and the spending deadline is 31 December 2027. Check with a tax adviser if you think you may qualify, but do not make the incentive a precondition for the investment. In terms of scale, the BOI Smart and Sustainable Industry measure, with 132 applications worth THB 17.158 billion in the first half of 2026, is the more realistic avenue to explore.

Can we reduce the number of physical inventory counts?

Changing their character is more practical than reducing their frequency. Annual or quarterly wall-to-wall counts are usually hard to drop because they are driven by accounting requirements, but adding routine cycle counting reduces the volume of variance that the full count uncovers. Because the purpose of cycle counting is identifying causes rather than correcting numbers (source: Factory AI), counting a small set of items once a week is already worthwhile. It is also worth pricing the count itself: 20 people over 2 days is 40 person-days, which at Thailand’s statutory minimum of 400 baht per day is 16,000 baht per count and 64,000 baht across four counts a year. Real wages, overtime premiums, supervisory attendance and the cost of halting shipments are all excluded, so treat it as a floor.

Summary

What an inventory management system comparison needs is not a product list but a set of measuring sticks. To recap the three from this article.

First, the positioning of the four types. Cloud inventory management, WMS, the inventory module of a production management system or ERP, and custom build separate along two axes: coverage and customisation depth. Wider coverage is not better. The task is matching where your problem lives to where each type is strong.

Second, the five cost layers. Split spend into licence, hardware, master data, integration and first-year operation, and the pattern that total cost runs 3 to 6 times the licence layer becomes visible. Align which layers each quotation covers before you compare totals, and treat published price ranges from other markets as structural references rather than local rates.

Third, the six root causes of inventory discrepancy. Unit conversion errors, missing location definitions and deferred issue postings can be structurally suppressed by software. Informal emergency withdrawals, unbooked scrap and unverified receiving quantities only respond to rules and ownership. Documenting that split before you select a product is what separates projects that deliver from projects that just add data entry.

On sequence: counting physical stock comes before selecting software. Measure current inventory accuracy and the distribution of variance causes in the first phase of the 90-day roadmap and your required feature set narrows by itself. The projects that struggle later are almost always the ones that skipped those 30 days.

The regional picture also argues against a single group-wide template. Thailand’s Manufacturing Production Index fell 3.10% year on year in June 2026 with average second-quarter capacity utilisation at 57.47% (source: Xinhua), while Vietnam’s manufacturing PMI held above the 50 mark at 51.8 in June with first-half manufacturing value added up 9.97% (source: VietnamPlus). Two sites in the same group, in the same year, facing opposite pressures. In the contracting Thai plant, visibility of slow-moving stock usually comes first; in the expanding Vietnamese plant, stockout prevention does. Plan for both rather than averaging them.

Used properly, an inventory management system will structurally remove about half of your inventory discrepancy. The other half is a question of shop-floor rules and accountability. A plan that keeps those two things separate is not a pessimistic plan; in our experience it is the fastest route to a result you can report with confidence.

If you are working through how to measure your own inventory accuracy, how to classify your variances, or how to size the five cost layers against your own conditions, you are welcome to raise it with us through the contact form. We are based in Bangkok and work with manufacturers across Thailand and ASEAN on factory IT and production management systems, including the inventory functions of our PEGASUS production management platform, but the most useful starting point is usually a joint look at which of the four types actually fits your situation. Treat it as help organising the decision, not as a sales conversation.