“We need to get serious about automation.” We hear this more and more often from manufacturing plants across Thailand. But when the conversation actually starts, the first question is rarely “which robot should we buy.” It stops one step earlier, at “what should we be automating in the first place?” The growing number of plants searching for factory automation consulting support reflects exactly that: the entry-point decision is hard to settle inside the organisation alone. This article is not about selecting equipment. It is about the layer above that — how to choose which process to automate, and how to decide who should help you make that call.
What this article covers, and what it does not
There is no shortage of automation content available. Most of it describes categories of equipment or showcases finished installations. Yet what actually stalls a real plant is almost never equipment selection. It is the decision-making that has to happen before equipment enters the conversation at all.
Here is the scope of this article.
| Covered here | Not covered here |
|---|---|
| How to decide which process becomes the automation target | General explanations of robot and sensor categories |
| How to run a factory assessment that narrows the candidate list | Rankings of specific manufacturers or products |
| Where failures actually originate, and how to build a team that prevents them | Any assumption that automation automatically pays off |
| How to frame the payback numbers (the calculation framework) | A fixed claim that “payback takes X years” |
| Where to draw the line between in-house work and outsourced work | Named recommendations of specific suppliers |
| Strengths and blind spots by type of consulting provider | A single “correct” methodology |
For the equipment-specific considerations, we already have detailed articles published. If material handling turns out to be your target process, see how to introduce AGVs and AMRs, and what they cost. If you need machines working in the same space as people, see the cost and rollout path for collaborative robots. This article stays deliberately upstream of both, on the single question of which process to pick.
Why “what should we automate?” has become such a common question
Thailand’s investment climate is pointing towards automation
According to the Thailand Board of Investment (BOI), investment promotion applications in 2025 reached 3,370 cases worth THB 1.876 trillion, a 67% increase year on year (source: Nation Thailand). Within that total, automation-related applications from January to September 2025 were reported at 402 cases worth THB 37.652 billion, up 81% against the same period the previous year (source: Mahanakorn Partners). The reported focus areas are the transition to Industry 4.0, digital technology, automation and robotics, and renewable energy adoption. Automation has clearly emerged as an investment theme in its own right.
In the Eastern Economic Corridor (EEC), a robotics investment on the order of THB 10 billion has been approved, forming a humanoid robot component cluster in Chachoengsao province and expected to create over 1,000 high-skilled jobs (source: Mahanakorn Partners). Automation and robotics are also among the BOI’s priority sectors for 2026.
For a company weighing an automation investment, this is a favourable backdrop. But if the target process gets chosen because “everyone else is investing, so we should too,” you walk straight into the failure patterns described later in this article.
The labour environment is what makes postponement impossible
Thailand’s 2026 minimum wage is set by province at THB 337 to 400 per day, with the THB 400 upper end unchanged from 2025 (source: Thai Law Online / Trading Economics). Wages themselves are not spiking. The more fundamental change is on the supply side of labour. Population ageing is affecting the labour pool, and a shortage of skilled workers is becoming a bottleneck for manufacturing (source: Allied Thai).
In other words, this is not the simple equation of “labour got expensive, so replace it with machines.” It is a structural problem: it is becoming harder to reliably secure enough people, with consistent quality of work, in the specific hours you need them. That distinction matters a great deal later, when you sit down to build the payback case.
Production volume itself is not strong
At the same time, current output is not especially robust. The Manufacturing Production Index (MPI) for May 2026 was reported down 0.8% year on year, with automobile production at 123,276 units, down 11.4% year on year (source: Bangkok Shuho). The Ministry of Industry is reported to be forecasting full-year 2026 MPI growth of 1.0% to 2.0%.
Capital investment decisions are harder to make when volume is flat. The “we’re automating to increase output” argument becomes difficult to use. In this environment, selecting a process based on low variability and quality stability — both covered below — tends to be easier to justify, whether to local management or to headquarters.
Globally, robot adoption remains at a high level
According to the International Federation of Robotics (IFR) World Robotics 2025, global industrial robot installations in 2024 reached 542,076 units, more than double the level of a decade earlier and above 500,000 units for the fourth consecutive year. The regional share of new installations was Asia 74%, Europe 16%, and the Americas 9%. By country, China accounted for 295,000 units, or 54% of the world total; Japan 44,500 units (down 4% year on year); Korea 30,600 units (down 3%); and India 9,100 units (up 7%, an all-time high). Southeast Asia is projected to expand at an average annual growth rate of more than 8%.
We could not verify Thailand-specific installation figures at the time of writing, so this article does not state any. That restraint is worth adopting internally as well: if you try to win over your own management using a “look how many units are being installed in Thailand” number, you will be unable to answer when someone asks for the source. Use macro figures as background context, and anchor the actual decision in your own shop-floor data.
The typical dead end: process selection was skipped
“What should we automate?” is really a signal that selection has not started
When a request comes to us, the first thing we ask is why that particular process was nominated as a candidate. When a clear answer comes back, everything that follows moves smoothly. When the answer is “no particular reason, it just has the most people” or “another company did it,” it is usually faster in the end to step back and redo the process selection.
Projects that move forward with selection skipped tend to show a consistent set of symptoms.
| Symptom | The unresolved item behind it |
|---|---|
| Quotations were obtained, but nobody internally can decide | No agreed definition of what “good” means |
| The only benefit anyone can articulate is “fewer people” | No quantified baseline data exists |
| Effort is spread across several processes and nothing advances | No screening method for setting priorities |
| The shop floor pushes back and the project halts | The reasoning behind the selection was never shared with them |
| Headquarters will not approve | The payback assumptions cannot be explained |
| Each vendor proposes at a different level of detail | The scope of work has not been defined |
None of these are technical problems. They are simply missing inputs to a decision. The reassuring flip side is that once you assemble those inputs, the technical evaluation that follows can safely be handed to specialists in each field.
The process with the most people is not automatically the right answer
A process with many operators looks attractive because the headcount impact appears large. But a process with many people is usually a process with many elements that require a person. Frequent changeovers, high variation in the workpiece, inspection steps that involve judgement — these are exactly why the headcount is high in the first place.
The difficulty of automation is determined not by headcount but by how thoroughly fixed the work is. A process with only three operators can be an excellent candidate if it repeats nearly identical motions around the clock. Whether your organisation can make that shift in perspective is the first real fork in the road.
“Start small” degrades into a slogan very easily
“Let’s start small” is a sound principle. The problem is how often a small start leads nowhere. In most of those cases, the first theme was chosen only because it was small, with no technical continuity to the process that actually matters.
If you are going to start small, choose the theme that eliminates the single most uncertain element of the process you truly care about. That way the next investment decision has something to stand on. If your real target is bin picking, for instance, run a small-scale test of whether vision-based position recognition holds up — and nothing else. That is the right way to slice it.
The factory automation assessment: how to screen for the target process
This is the core of the article. When we run a factory assessment to narrow down the target process, we apply five screens in sequence. Applying them in order brings the candidate list down to a realistic number.
Screen 1: How many labour hours are actually being consumed
Start by producing actual labour hours per process. The critical word is actual, not standard. Standard time hides everything around the work: setup, rework, waiting for material, cleaning, and record-keeping. In practice it is not unusual for these peripheral activities to account for a far larger share than anyone expected.
Break labour hours down as follows.
| Category | What it includes | Suitability as an automation target |
|---|---|---|
| Primary work | Machining, assembly, inspection — direct value-adding work | High (but also technically demanding) |
| Peripheral work | Picking up material, repositioning, passing to the next process | Very high |
| Setup work | Die changes, jig changes, program switching | Moderate (depends on frequency) |
| Record-keeping | Entering results, filling in forms, logging inspections | High (often solvable on the system side) |
| Waiting and idle time | Waiting for the previous process, for material, for approval | Address through process design rather than automation |
| Abnormality handling | Recovering from minor stoppages, rework, defect disposition | Low (eliminate the root cause first) |
The two rows worth studying carefully are record-keeping and waiting. Where those two are large, there is likely room to improve through production management systems before any robot is installed. Confirm there is nothing to fix upstream of the machine before you move on to the equipment discussion.
Screen 2: The defect situation
A process with a high defect rate is a legitimate candidate, but it needs care. If the defects stem from variation in how operators work, automation helps. If they stem from variation in the material, from problems carried in by the upstream process, or from equipment ageing, automation will not reduce them. In fact, a machine that cannot handle inconsistent workpieces will stop, and your utilisation rate will fall.
Separate the causes as follows.
| Cause of defects | Effect of automation | What to tackle first |
|---|---|---|
| Variation between operators | Strong effect | Document the standard work |
| Variation within one operator across the shift | Strong effect | Identify fatigue factors |
| Variation in material or components | Limited effect | Revisit incoming acceptance criteria |
| Carried in from the upstream process | Weak effect | Countermeasures upstream |
| Change in equipment condition | Weak effect | Revisit the maintenance interval |
| Ambiguous judgement criteria | Depends on conditions | Quantify the pass/fail criteria |
Ambiguous judgement criteria can sometimes be automated with vision inspection. But that first requires defining what constitutes a defect, either numerically or through image samples. Skip that definition work and you end up with an inspection machine plus a human re-checking behind it, which does not reduce headcount at all.
Screen 3: How little the process varies
Automation pairs well with processes that vary little. Variability comes in several forms.
| Type of variation | What to check | Impact when variation is high |
|---|---|---|
| Product mix | How many variants run per day | Changeover itself has to be automated separately |
| Volume | Month-to-month and day-to-day swings in output | Utilisation drops and payback stretches out |
| Workpiece dimensions and shape | Variation within the same part number | Jig and gripper design becomes far harder |
| Material presentation | Arranged in order, or randomly piled | Recognition technology is required, raising cost |
| Work sequence | Frequency of exception handling | Human intervention must be designed in |
| Equipment availability | How often adjacent processes stop | Buffer design becomes necessary |
High variation is not in itself a problem. The problem is fixing the equipment specification without ever having quantified the variation. A requirement such as “it must also handle the variant that runs twice a month” is a very common driver of equipment cost. Agreeing internally, in advance, that this particular variant will stay manual is often what keeps the specification inside a realistic envelope.
Screen 4: Safety and the working environment
Safety is hard to express in a payback calculation, yet it can dramatically change priorities. The following processes deserve to be moved up the list even when the financial benefit is modest.
- Processes with repetitive handling of heavy items, where back strain has become chronic
- Processes involving extended work in heat, dust, or organic solvent environments
- Processes such as pressing or cutting where people work close to pinch and entanglement hazards
- Processes involving work at height or on unstable footing
- Processes with a repeated history of near-miss reports
These also tend to be the processes hardest to staff. Work that attracts few applicants when you advertise is the work most exposed to the shift in labour supply. Evaluating a process on both safety and staffing difficulty makes the priority argument much easier to construct.
Screen 5: Suitability for 24/7 operation
Considering suitability for round-the-clock operation changes the payback story fundamentally. The same machine carries a completely different depreciation burden per operating hour on one shift versus three.
That said, running unattended overnight is never achieved by automating the machine alone. The following elements have to be in place before lights-out or minimally staffed night operation means anything.
| Element | What to verify | What to do if it is missing |
|---|---|---|
| Material supply | Can a full night’s material be loaded in advance | Consider stockers or automatic feeding |
| Finished goods discharge | Will the line stop once the output buffer is full | Size the discharge buffer properly |
| Stoppage and alerting | Who is notified, and how | Remote monitoring and notification |
| Quality assurance | How quality is guaranteed during unattended hours | In-line inspection and automated recording |
| Maintenance | Who handles minor overnight trouble | Define response procedures and standby coverage |
| Records and traceability | Whether you can reconstruct what was made and when | Integration with a production data collection system |
The bottom three rows — quality, maintenance, and records — cannot be solved by the machine alone. They have to be designed together with automated production data collection and traceability. If you intend to build your payback case on 24/7 operation, budget for these costs from the very beginning.
Consolidating the five screens onto one sheet
In practice, we build an evaluation sheet with candidate processes across the top and the five screens down the side. Each item is scored on a three- or five-point scale and weighted to produce a total. The aim is not mathematical precision. The aim is to leave behind a record that explains why this process was chosen.
The weighting depends on your situation. If securing people is your biggest challenge, weight labour hours and safety more heavily. If quality complaints are the problem, weight defects. Agreeing the weighting with senior management up front makes every subsequent investment decision faster.
Automation risk: where failures actually happen
The main cause of failure is not technical
Research in Japan on automation and AI adoption failures points to the main causes being concentrated not in technical limitations but in the implementation team, data readiness, and shop-floor engagement. Framing the problem correctly beforehand and validating in stages is what separates success from failure. The same research reports that many companies have not started at all because they do not know where to begin or cannot see the cost-benefit picture (source: Japan AI Adoption Support Association).
That finding matches what we see on the ground in Thailand. Our impression is that projects where the machine works but nobody uses it outnumber projects that failed because the machine would not run.
Breaking down where failure originates
Mapped along the flow of a project, the failure points look like this.
| Stage | Common failure | What to do first to prevent it |
|---|---|---|
| Problem definition | The objective becomes “to automate” | Set quantified target values first |
| Process selection | Chosen purely on headcount | Evaluate with the five screens and keep the record |
| Data readiness | No baseline performance data exists | Measure one to two months of it, manually if necessary |
| Specification | Trying to cover every exception variant | Separate machine scope from manual scope |
| Implementation team | The owner is doing this on top of another job | Allocate the hours and document the responsibilities |
| Shop-floor engagement | The floor is informed only after decisions are made | Involve key floor personnel from the selection stage |
| Commissioning | No commissioning period in the schedule | Allow for the ramp-up to full utilisation |
| Sustained use | Only one person can operate it | Include training and manuals in the deliverables |
The two most frequently overlooked items on this list are data readiness and sustained use.
Without data, you cannot measure the effect
To explain the effect of automation, the state before installation has to exist as numbers. Yet plenty of plants still keep results only on paper daily reports, or transcribed into Excel but never aggregated. Automate from that starting point and the strongest post-installation statement you can make is “it feels better.”
Once the target process is decided, we recommend capturing at least the following for one to two months.
- Actual labour hours per process, including peripheral work
- Production quantity by day and by part number
- Number of defects and a breakdown by defect type
- Equipment downtime and the reason for each stoppage
- Number of changeovers and the time taken per changeover
Manual recording is perfectly acceptable. What matters is that it is captured in a form you can compare against on the same basis after installation. On systematising this kind of data collection itself, our article on labour-saving automation examples and cost-effectiveness works through how to think about the return.
Failure patterns that are particularly common in Thailand
On top of the failures shared with plants anywhere, manufacturing sites in Thailand tend to show the following patterns.
| Pattern | What happens | Direction of the countermeasure |
|---|---|---|
| The specification is set by headquarters overseas | The specification does not match how the work is actually done locally | Base the specification on local process measurements |
| Operating instructions are lost to the language barrier | Local operators never fully master the equipment | Make Thai-language manuals and HMI screens a delivery condition |
| No maintenance staff who can work in the vendor’s language | Every breakdown waits on a call overseas | Write the local support arrangement into the contract |
| The person who owns the equipment changes jobs | Operational know-how disappears | Document procedures and train multiple people |
| Spare parts are not available locally | Downtime drags on | Check local parts availability during selection |
| Differences in power, air, and ambient conditions | The machine does not perform as specified | Provide measured local conditions as the design basis |
Turnover of the responsible engineer is a risk that plants in some other countries rarely have to plan for. Documentation and multi-person training as protection against key-person dependency are best written into the contract as deliverables. Wording it concretely — “operating manuals to be produced in Thai and delivered in both printed and electronic form” — leaves far less room for later disagreement.
Capital investment planning: how to frame the numbers
“Labour cost savings” alone will not carry the argument
The most common payback story is “we save X operators’ wages, so it pays back in Y years.” That story is frequently not enough on its own.
Thailand’s 2026 minimum wage is THB 337 to 400 per day by province, with the upper end unchanged from 2025 (source: Thai Law Online / Trading Economics). Actual pay is often higher than the minimum, but wage levels remain comparatively low. Trying to recover equipment cost through labour savings alone therefore produces a long payback period on paper.
The answer is to build up the benefit from several separate components.
| Benefit component | How to quantify it | Cautions |
|---|---|---|
| Direct labour | Headcount reduced × annual cost per person (including social contributions) | Treat redeployment separately from genuine reduction |
| Overtime and holiday work | Hours reduced × premium rate | Be careful annualising a peak-season-only effect |
| Defects and rework | Defects avoided × loss per unit | Include labour, not just material cost |
| Extended operating hours | Additional units × contribution margin | Only valid if the demand exists |
| Shorter changeover | Time saved × hourly rate for that process | Verify the actual changeover frequency |
| Automated quality records | Recording and aggregation hours × rate | Include the hours spent supporting audits |
| Reduced accident risk | Difficult to quantify | Present separately as a qualitative assessment |
| Recruitment and training cost | Cost per hire × reduction in annual hires | Requires actual turnover data |
The note about redeployment matters. If the people freed up by automation are moved to other processes, labour cost in cash-flow terms does not fall at all. In that case it is more accurate to reframe the benefit as “how much more output the same headcount can produce.”
How to treat BOI incentives in the calculation
For capital investment in Thailand, whether BOI incentives apply has a large effect on the payback period. The BOI grants strong incentives to the manufacture of robots and automation machinery and to system integration businesses, and incentives are reported to be granted even when existing machinery is replaced with more efficient machinery (source: BOI investment promotion policy materials).
Corporate income tax exemption is available for up to eight years, and in designated EEC zones the exemption period may be extended under certain conditions, with an additional 50% reduction for three to five years after the exemption period ends (source: BOI / JETRO).
However, eligibility varies with industry, the nature of the investment, location, and application timing. It would be risky to read this article and conclude that your project qualifies. Always confirm directly with the relevant authority or the BOI. In practice, the safest approach is to model the payback period twice — once with incentives and once without — and confirm that the case still holds without them.
How to read the payback period
There is no universal standard for what payback period is acceptable. The judgement depends on what the investment is meant to achieve.
| Nature of the investment | How the acceptable payback period tends to be viewed | Notes |
|---|---|---|
| Pure labour reduction | Usually expected to be short | The wage level caps the size of the benefit |
| Defect suppression | Depends on the size of the losses | Base it on actual escaped-defect complaints |
| Capacity expansion | Depends on confidence in the demand forecast | Whether firm orders back it up is decisive |
| Safety measures | May take priority over payback period | Share the qualitative judgement with management |
| Preparing for labour scarcity | Does not fit a payback calculation well | Reinforce with actual recruitment difficulty data |
| Moving to 24/7 operation | Improves through the extended operating hours | Build in the added cost of night coverage |
From the plant’s perspective, presenting these separately by objective moves the discussion along better than collapsing them into a single number. “As a labour-reduction case it is X years, but on safety grounds this process is the top priority” is much closer to reality.
How to view the cost of the validation stage
A small-scale proof of concept often sits ahead of full deployment. As a benchmark within Japan, PoC costs are cited at roughly JPY 500,000 to 3,000,000 over two to three months (source: Japan AI Adoption Support Association).
That is a domestic Japanese benchmark and nothing more. Engineering rates in Thailand, local procurement conditions, and import lead times all differ, so these figures cannot be transposed directly onto a Thai project. Treat them as a rough sense of magnitude and obtain individual quotations for the actual amount.
At the validation stage, the following points matter more than the cost figure.
- Whether the pass/fail criteria for “success” were agreed before the work started
- Whether the data and samples used are genuinely representative of real production
- Whether you have decided what you will learn from a failed validation and how you will carry it forward
- Where ownership of the validation output — data and programs — sits
Pass/fail criteria in particular should be agreed in writing before validation begins. Proceed with vague criteria and you land in the worst outcome: it ran, but nobody can decide whether to deploy it.
Production engineering outsourcing: drawing the in-house line
Do not outsource everything
Even when you bring in external help for the automation study, certain functions belong in-house. Here is how we draw the line.
| Area | Fit for in-house | Fit for external help | Reason |
|---|---|---|---|
| Defining the current problem | High | Low | Requires knowing both management intent and shop-floor reality |
| Process measurement and data collection | High | Medium | A capability you need continuously, so it should stay |
| Final selection of the target process | High | Medium | It is an accountable decision |
| Assessing technical feasibility | Low | High | Requires broad technology knowledge |
| Detailed equipment design | Low | High | Highly specialised and infrequently needed |
| Control and software development | Low | High | Specialist engineers are hard to retain |
| Installation and commissioning | Low | High | Requires a large temporary workload |
| Day-to-day operation and maintenance | High | Low | Directly tied to downtime, so internal response wins |
| Measuring results and improving | High | Medium | An ongoing activity |
The message of this table is that problem definition, final selection, and operation and maintenance should stay in-house. Outsource beyond that point and you may be able to install equipment, but you will no longer be able to make the next investment decision on your own.
At the same time, insisting on doing detailed design and control development internally carries a heavy cost in securing and retaining specialists. If you only run one or two projects a year, bringing in outside capability is the rational choice.
When the production engineering function is thin
At local subsidiaries in Thailand, it is common to find one or two dedicated production engineers, or a manufacturing manager covering the role on top of their own job. Advancing automation from that position requires one of two things.
- Ring-fence a dedicated person’s hours for a defined period, explicitly removing other duties
- Bring in an external party to act as your production engineering function
For the second option, you need a partner who sits on the buyer’s side of the table rather than an equipment manufacturer. Consult an equipment manufacturer and the answer will inevitably lean towards the solutions that manufacturer is good at. That is not wrong in itself, but once equipment is presumed from the process selection stage onward, genuine comparison is no longer possible.
Understand how the work is packaged
FA system integration work is packaged into stages. You can award different stages to different parties, or award all of them to one.
| Stage | Main content | Deliverables | How to think about bundling versus splitting |
|---|---|---|---|
| Concept design | Target process selection, comparison of methods, budgetary cost | Concept drawings, comparison study, budgetary quotation | High value in splitting this out to compare multiple options |
| Detailed design | Mechanical design, electrical design, control specification | Detailed drawings, bill of materials, control specification | More efficient to award together with fabrication |
| Fabrication and procurement | Parts procurement, machining, assembly, in-house trial run | The machine itself, trial run records | Typically the same party as detailed design |
| Installation | Delivery, positioning, wiring, piping | Installation completion report | Select on local execution capability |
| Commissioning | Adjustment, pilot production, performance verification | Acceptance records, performance data | Write the assumed commissioning period into the contract |
| Training and handover | Operator training, maintenance training, document handover | Manuals, drawings, programs | Fix the deliverables list in advance |
The stage most often neglected is the last one. Leave training and handover vague in the contract and you end up with a delivered machine but no program source code in your hands, or drawings that are not the as-built version — both of which will hurt you later during modification or maintenance.
The deliverables worth confirming at the time of order are as follows.
- Mechanical drawings (final version, reflecting any modifications)
- Electrical drawings and panel wiring diagrams
- Control program (source, commented) and the scope of the licence granted to you
- Operating and maintenance manuals in a language your operators can read
- Bill of materials (model numbers, manufacturers, local availability)
- Performance data at acceptance, together with the measurement conditions
How to choose a factory automation consulting partner
Each type of provider has a different profile
“Automation consulting” is offered by very different kinds of organisation. Each has strengths and limits.
| Type | Strengths | Points to watch |
|---|---|---|
| Equipment manufacturers | Fast, concrete proposals built around their own products | Solutions tend to converge on what their products cover |
| Trading companies and distributors | Can compare products across multiple manufacturers | Engagement with process design itself may be limited |
| Management consulting firms | Connect to management KPIs, structure the investment case | Technical feasibility still has to be verified separately |
| System integrators | Can design through to the link between equipment and IT systems | Areas of strength vary widely from firm to firm |
| Production engineering service firms | Build specifications from the buyer’s side | Without fabrication capability, a separate order is needed |
| Universities and public institutions | Verification of technical validity, insight into emerging technology | Implementation as production equipment is often out of scope |
Which type suits you depends on where you currently stand. If the target process is undecided, choose a partner who will not presume the equipment, so the study stays broad. If both the process and the method are already settled, going straight to a partner with strong implementation capability will get you there faster.
What to confirm in quotations and contracts
Confirming the following before you select a partner will prevent a great deal of misalignment later.
| Item to confirm | What to ask specifically |
|---|---|
| Scope of work | Through concept design, through detailed design, or through fabrication and installation |
| Assumptions | Which part numbers, volumes, operating hours, and material conditions the proposal assumes |
| Handling of changes | How additional costs are treated if the assumptions change |
| Deliverables | What will be delivered, in what format, in what language |
| Schedule | Duration of each stage, and what happens if it slips |
| Acceptance criteria | What constitutes completion, what performance is guaranteed, and under what measurement conditions |
| Warranty and support | Warranty period, coverage hours, availability of local response, and cost |
| Intellectual property | Ownership of control programs and data, and whether reuse is permitted |
| Track record | Experience with similar processes at a similar scale |
| Team | Who will be assigned, whether they will be on site, and how language is handled |
Assumptions and acceptance criteria in particular must be committed to writing. Projects that proceed with these two left vague are exactly the ones that descend into disputes about who said what during commissioning.
Running a competitive bid properly
If you are collecting quotations from several firms, they are not comparable unless every firm receives the same conditions. At a minimum, give all of them identical information on the following.
- The current state of the target process: work content, cycle times, staffing
- Target part numbers, with quantity, dimensions, and the range of shape variation for each
- Required takt time, or annual production volume
- Installation footprint and constraints: ceiling height, floor loading, interference with existing equipment
- Available power, compressed air, and other utility conditions
- Operating hours, and any planned future change to them
- How the equipment connects to upstream and downstream processes
- Quality standards and the items requiring inspection
- Desired delivery date and the reason behind it (fiscal year end, peak season, and so on)
Package all of this into a single RFP and the differences between proposals become visible as differences in approach. Without it, each firm quotes on different assumptions, leaving you no basis for judgement beyond price.
Learning from smaller-company automation examples: how to phase the scale
Do not attempt everything at once
At smaller sites, or at sites tackling automation for the first time, phasing the work raises the success rate far more than building a large line from the outset. There is a well-established way to phase it.
| Phase | Content | What this phase gives you |
|---|---|---|
| Phase 0 | Get to a state where you can capture actual performance data | A baseline for measuring effect; identification of the real problems |
| Phase 1 | Assist a single process (transport, feeding, discharge) | Experience operating equipment; internal consensus |
| Phase 2 | Automate a single process (machining, assembly, inspection) | Quality stability; redeployment of people |
| Phase 3 | Connect processes (automated transport, data linkage) | Less idle time; shorter lead time |
| Phase 4 | Extend operating hours (minimally staffed nights and weekends) | Better return on the investment already made |
| Phase 5 | Whole-line and multi-line rollout | Efficiency of horizontal deployment through standardisation |
Do not skip Phase 0. Move to Phase 1 without data and you will be unable to explain the effect, which means you will not get approval for the Phase 2 investment.
Starting points that pay off even at small scale
The following are relatively easy to start with a small investment and tend to lead naturally into the next step.
- Transport between processes (automating trolley pushing)
- Feeding material and components, and discharging finished goods
- Automating inspection records and production data entry
- Assisting with lifting and turning heavy items
- Downstream work with highly repetitive motion, such as packing and palletising
If you are considering automated transport, our article on introducing AGVs and AMRs sets out the decision criteria and how to think about cost. For the wider picture, see our article on factory automation in Thailand, which covers current conditions and how to proceed.
Design the first machine for horizontal deployment
If you operate several sites or several lines, whether the first machine is built in a form that can be replicated determines what every subsequent machine costs.
| Decide this for horizontal deployment | What happens if you do not |
|---|---|
| Standard manufacturer and model for control devices | Spare parts differ by site and inventory balloons |
| Program structure and naming conventions | A different engineer cannot modify it |
| HMI screen layout and language | Training has to be rebuilt from scratch at each site |
| Data output format and fields | Results cannot be compared across sites |
| Safety design philosophy | Safety levels vary from site to site |
| Drawing and manual formats | Document control fragments across sites |
At the first-machine stage this looks like unnecessary overhead. It comes back as a difference in engineering cost and commissioning duration for every machine after it. Data output format and fields in particular are worth fixing early, because unifying them later means modifying installations at every site.
Putting the whole path on a timeline
Rearranging everything above along a timeline gives the following. The durations are general guides only and will shift with the difficulty of the target process and the strength of your internal organisation.
| Phase | Main activities | Roles required internally | What external help can cover |
|---|---|---|---|
| Preparation | Clarify objectives, form the team | Management, project owner | Designing the approach |
| Baseline assessment | Process measurement, data collection | Manufacturing, production engineering | Measurement method design, analysis |
| Process selection | Evaluation using the five screens | Production engineering, manufacturing, quality | Designing the criteria, third-party perspective |
| Concept | Method comparison, budgetary cost, benefit estimate | Production engineering, finance | Technical feasibility assessment |
| Investment decision | Internal approval, BOI and related checks | Management, finance, legal | Support in preparing documentation |
| Implementation | Design, fabrication, installation, commissioning | Production engineering, manufacturing, maintenance | Design, fabrication, installation, commissioning |
| Sustained use | Training, procedures, measuring results | Manufacturing, maintenance, quality | Providing the training programme |
| Next step | Verifying results, selecting the next process | Everyone | Review and proposal of the next case |
The column worth studying is “roles required internally.” Manufacturing and production engineering appear in nearly every phase. Outsourcing does not reduce your internal workload to zero. Build a schedule without allowing for those hours and the project will stall partway through.
Frequently asked questions
How much does factory automation consulting cost?
It varies enormously with scope, so no single figure applies. Supporting process measurement and target selection is an order of magnitude different from covering concept design through to fabrication and installation. Decide first how much you want to entrust to an outside party, then collect quotations from several firms on identical conditions. For small-scale validation, one Japanese study cites PoC costs of roughly JPY 500,000 to 3,000,000 over two to three months (source: Japan AI Adoption Support Association). Note that this is a domestic Japanese benchmark and cannot be transposed onto Thai engineering rates and procurement conditions.
How long does a factory automation assessment take?
It depends on the scope and on how much data you already have. There is a large difference between a plant where performance data can already be aggregated and one that has to start measuring from scratch. If you are starting from measurement, we recommend capturing at least one to two months of data so that production variability is properly reflected. Conversely, if you are only narrowing the candidate list using data you already hold, it can sometimes be organised in a much shorter period.
Can we talk to you before the target process is decided?
Yes — and honestly, that is the better time to talk. Once the equipment specification has hardened, there is little room left to rearrange the assumptions, and genuine comparison becomes impossible. If you can describe the situation as a problem — “we cannot find enough people,” “defects are not coming down,” “we want to run overnight” — that is enough to start narrowing down the target process together.
What payback period should we expect from automation?
There is no universal standard. The period your organisation will accept depends on whether the objective is labour reduction, quality stability, capacity expansion, or safety. Whether BOI incentives apply also changes the tax burden and therefore the calculated payback. In practice, the safest approach is to model both cases — with and without incentives — and confirm the investment is still acceptable without them. Because eligibility varies by industry and conditions, always confirm with the relevant authority or the BOI directly.
What is the single biggest cause of automation failure?
Research in Japan points to the main causes of failure being concentrated not in technical limitations but in the implementation team, data readiness, and shop-floor engagement (source: Japan AI Adoption Support Association). From what we have seen at plants in Thailand, our impression is the same: projects where the machine works but has fallen out of use outnumber projects that stopped because the machine would not run. In particular, without baseline data from before installation you cannot explain the effect, and without that explanation there is no next investment. Involving key shop-floor people from the process selection stage, and agreeing how the effect will be measured before you start, is the practical prevention.
Is automation realistic for a smaller site?
Small scale is not a disqualifier. But rather than building a large line from the outset, phasing up from assisting a single process is far more realistic. Transport, material feeding, finished-goods discharge, and automating production data entry are all starting points that require relatively little investment while leading naturally into the next phase. The important thing is to choose that first machine in a form that leads to the next step.
What changes if we assume 24/7 operation?
The payback calculation may improve, but the machine on its own is no longer sufficient. Overnight material supply, finished-goods buffering, stoppage detection and alerting, quality assurance during unattended hours, a response arrangement for minor trouble, and automatic recording of production results all have to be designed together. If you intend to base your payback case on 24/7 operation, include these additional costs from the start.
Can BOI incentives be applied to automation investment?
The BOI grants strong incentives to the manufacture of robots and automation machinery and to system integration businesses, and incentives are reported to be granted even where existing machinery is replaced with more efficient machinery (source: BOI investment promotion policy materials). Corporate income tax exemption is available for up to eight years, and in designated EEC zones the exemption period may be extended under certain conditions, with an additional 50% reduction for three to five years after the exemption ends (source: BOI / JETRO). Eligibility varies with industry, the nature of the investment, location, and application timing, so always confirm directly with the relevant authority or the BOI.
Summary
When an automation discussion starts from “what should we automate,” the target process selection has usually been skipped. Before moving on to equipment, evaluate the candidates against five screens — labour hours, defects, low variability, safety, and suitability for 24/7 operation — and leave behind a record that explains why you chose what you chose. That record is what makes every later investment decision faster.
The main cause of failure is reported to be concentrated not in technology but in the implementation team, data readiness, and shop-floor engagement. Capturing baseline data before installation, involving the shop floor from the selection stage, and including training and documentation in the deliverables are what determine whether the equipment keeps being used. At plants in Thailand, add language support, a local maintenance arrangement, and preparation for personnel turnover to that design.
Payback numbers rarely hold up on labour cost savings alone. Build them up separately from defect reduction, extended operating hours, reduced record-keeping effort, and lower recruitment and training cost — and model BOI incentives both ways, with and without, so the decision stands on stable ground.
Finally, you do not need to outsource everything. Keep problem definition, the final selection of the target process, and day-to-day operation and measurement in-house, and bring in outside capability for technical feasibility assessment, detailed design, fabrication, and commissioning. Draw that line clearly and you will be able to make the second and third investment decisions yourself.
You are welcome to talk to us before the target process is decided — in fact, the conversation is usually more productive before the equipment specification hardens, because the assumptions can still be worked through together. Describing it as a problem is enough: “we cannot find enough people,” “defects are not coming down,” “we want to run overnight.” Feel free to get in touch through our contact form, and we will work through your situation and where it makes sense to start.
References
- Nation Thailand (BOI investment applications, 2025)
- Mahanakorn Partners (automation applications and EEC robotics investment)
- BOI Thailand Investment Promotion Policy for Automation and Robotics Industries (Japanese PDF)
- BOI Automation
- JETRO: Incentives for Foreign Investment in Thailand
- IFR World Robotics 2025
- Thai Law Online (minimum wage 2026)
- Bangkok Shuho (MPI and automobile production, May 2026)
- Allied Thai (Thailand economy 2026)
- Japan AI Adoption Support Association (AI in manufacturing: failures and costs)
