Robot Implementation for SMEs: PoC, Safety, and Business Case for Thailand Plants

When a small or medium-sized manufacturer starts researching robot implementation, the first choice should not be a robot model. A Thailand plant with limited engineering, maintenance, staffing, and shutdown capacity should first select one stable, measurable task in one workcell—ideally one that can be isolated without stopping the entire factory. This guide is deliberately different from a broad introduction to industrial robots or a generic ROI article. It provides a decision path for a resource-constrained SME to run a 60–90-day proof of concept (PoC) and decide whether to scale, continue under conditions, or stop. It connects baseline measurement, make-or-buy boundaries, RFP requirements, factory and site acceptance tests, task-based safety, teaching skills, maintenance ownership, total cost of ownership, and scale/no-go gates.

Why an SME should not begin robot implementation with model selection

The Thailand Board of Investment (BOI) reported that it approved investment-promotion applications for 1,300 projects valued at approximately THB 1.31 trillion in the first half of 2026. It also reported 132 applications valued at approximately THB 17.2 billion under the Smart and Sustainable Industry initiative for machinery upgrades, digital technology, automation, and robotics. These figures are date-scoped indicators of industrial modernization activity. They do not prove that any individual robot project will be profitable, nor do they mean that BOI incentives apply automatically. Eligibility, timing, technical scope, documentation, and conditions involving domestic automation capability must be confirmed with BOI for the specific applicant and project.

For an SME, the constraint is rarely the purchase price alone. Product mix, fixture variation, material supply, changeovers, off-shift recovery, Thai-language procedures, employee turnover, and dependence on an external integrator interact. A robot can meet its own specification while the workcell fails because gripping, feeding, peripheral equipment, safeguarding, quality checks, and recovery were never integrated as one operating system.

NIST’s guidance for collaborative robot integration in small and medium-sized manufacturing operations starts with identifying the workcell best served by automation and progresses from quick, basic assessment methods to methods that take more effort but provide more accurate evaluation. NIST MEP likewise presents assessment, selection of priority opportunities, a business case aligned with strategy, connection to integrators, and rigorous measurement as a sequence. The first deliverable should therefore be a candidate-workcell comparison and a baseline—not a purchase specification.

Select one stable task and workcell first

The best first use case is not necessarily the task that looks most tiring. It is a task whose inputs, actions, outputs, and failures can be observed. Machine tending, consistent case packing, movement of known parts, simple dispensing, or feeding a fixed inspection position may be candidates. The application name is not enough: the team must inspect real parts, oil, dust, lighting, variants, fixtures, and cycle fluctuations at the gemba.

Example workcell-screening matrix

This is an internal decision aid, not a score prescribed by a standard. Change the importance of each factor to fit the plant.

Compare several candidates in the same format. Including a low-complexity/low-value task, a high-value/high-variation task, and a balanced task helps management and operations understand why the first workcell was selected.

Measure the current-state baseline before the PoC

The return on robot implementation comes from the difference from the current state, not the robot’s maximum catalog speed. Measure multiple shifts, representative variants, normal operation, and exceptions. A single “typical day” can hide changeovers, waiting, rework, cleaning, breaks, and breakdowns.

Record at least:

• actual good-part cycle-time distribution;
• runtime, planned downtime, unplanned downtime, and reason codes;
• work in process, material starvation, and upstream/downstream blocking;
• defects, reinspection, rework, scrap, and their causes;
• operator interventions, reasons, and duration;
• changeover, fixture replacement, cleaning, and startup verification time;
• exposure to approach hazards, awkward posture, heavy loads, heat, or sharp parts; and
• comparable utility, consumable, and maintenance effort data.

Document the measurement method, observation window, missing data, and exceptions. “A person takes 45 seconds and the robot takes 40” is not a valid decision model unless both figures include the same material delays, image retries, door movement, cleaning, and alarm recovery.

Set the make-or-buy boundary before procurement

The decision about which knowledge remains inside the plant has more influence on long-term downtime than the choice to purchase a particular arm. An SME need not build everything internally. However, outsourcing every decision creates delay and expense whenever a small product change or failure occurs.

Decisions the plant should own

• task purpose, quality criteria, and exclusions;
• normal, abnormal, and recovery scenarios;
• acceptance-test decisions and approval of open items;
• operating safety rules, training population, and change approval;
• access, backup, and change-record ownership; and
• operating KPIs and scale/stop decisions.

Specialist work that can be contracted

• detailed robot, gripper, fixture, vision, PLC, and safety design;
• specialized design and validation of risk-reduction measures;
• electrical, pneumatic, mechanical fabrication, installation, and calibration;
• software, HMI, data integration, and performance testing; and
• documentation, spares, training, warranty, and remote/on-site support setup.

Define the boundary through deliverables, not only a RACI chart. “The vendor handles teaching” is too vague. State who creates initial points, who approves production changes, what the plant may adjust, and who performs backup and restore tests before changes.

Design a 60–90-day PoC to create a decision

Sixty to ninety days is an illustrative planning horizon, not a mandatory standard or a promise. Adjust it for component lead times, shutdown windows, regulatory needs, and cell complexity. A PoC is not a period for producing a demonstration video; it is a controlled effort to collect the evidence required for the next investment decision.

Define the exit criteria before starting. Success is not “the robot moves.” Quality, safety, capability, recovery, training, maintenance, and total cost should each have a gate. A missed target may justify a conditional continuation if the cause, cost, owner, and retest are acceptable. Conversely, a technically working cell should be stopped if the plant cannot recover it or the updated business case exceeds its limits.

Use the RFP to compare maintenance and support, not only hardware

Require every bidder to respond to the same operating scenarios. Provide representative and boundary parts, quality requirements, cycle assumptions, layout, utilities, shifts, safety constraints, and data needs. Identify uncertain inputs explicitly and require bidders to list their assumptions.

Minimum RFP content

1. Products, variants, dimensions, mass, tolerances, surface conditions, and presentation.
2. Normal cycle, changeover, cleaning, start/end of shift, failure, and recovery scenarios.
3. Target capacity, quality, and availability measurement, including excluded time.
4. Scope of robot, tooling, fixture, vision, PLC, HMI, and safety components.
5. Buyer-supplied items, power, air, network, foundation, and delivery constraints.
6. Method for confirming risk assessment, applicable law, plant rules, and standards.
7. FAT/SAT cases, test parts, endurance runs, abnormal cases, and evidence.
8. Delivery of source files, credentials, licenses, backups, drawings, and bill of materials.
9. Operator, teaching, maintenance, safety, and administrator training by language.
10. Warranty, response, on-site arrival, remote support, spares, and obsolescence terms.

A claim of “24/7 support” is insufficient. Separate call reception from technical response; confirm supported languages, remote-access approval, response and restoration targets, exclusions, and pricing. If local stock is claimed, identify which parts, who owns them, allocation conditions, and transport time.

Make FAT and SAT contractual acceptance gates

The Factory Acceptance Test and Site Acceptance Test should control shipment, payment, and production release. If final payment is tied only to physical installation, unresolved performance, safety, documentation, and training issues may remain after the buyer’s leverage has declined.

FAT checks

• correspondence between approved drawings/BOM and the built cell;
• grip, placement, and quality with representative and boundary parts;
• normal cycle, sustained run, stop, and restart;
• expected failures such as missing/double-fed parts, sensor faults, and communication loss;
• evidence for guards, interlocks, emergency stops, and safety functions;
• HMI language, alarm messages, and recovery guidance;
• programs, parameters, backups, and restore instructions; and
• punch-list item, owner, due date, and retest condition.

Additional SAT checks

• actual power, air, network, lighting, climate, floor, and vibration conditions;
• upstream/downstream equipment, logistics, material flow, and human routes;
• capability, quality, changeover, and cleaning during real shifts;
• diagnosis and safe recovery by the plant’s operator and maintenance team;
• final risk reduction, marking, training, and access control; and
• as-built documentation, backup storage, and change-control activation.

Each test item needs a method, condition, sample, acceptance result, record, signature, and response to failure. Retain distributions and stop reasons rather than only averages. A few cycles under ideal conditions are not adequate evidence.

A collaborative robot label does not determine workcell safety

Collaborative features do not automatically make a complete application safe. Sharp gripped parts, dropped objects, crushing against a fixture, process hazards from welding or machining, unexpected startup, and stored energy during maintenance remain application hazards.

ISO 10218-1:2025 addresses the industrial robot as partly completed machinery. ISO 10218-2:2025 addresses integration of industrial robot applications and robot cells across design, integration, commissioning, operation, maintenance, decommissioning, and disposal. A project must use the purchased standards, applicable Thai requirements, plant rules, and manufacturer instructions with competent safety specialists. A public ISO overview page is not enough to determine conformity.

ISO/DTR 20218-3 is a Draft Technical Report intended to explain the second edition of ISO 10218-2. As of September 2026, the cited ISO page lists it as under development at the approval stage. It must not be described as a published normative standard or used alone as contractual evidence of conformity. Recheck its publication status when a decision is made.

The U.S. OSHA technical manual on industrial robot safety is a useful practical reference for task-based risk assessment and hazards during installation, programming, testing, operation, and maintenance. It is a U.S. source, not Thai law and not a substitute for Thailand-specific legal, authority, insurance, or customer requirements.

Where machinery is made available on the EU market or put into service in the EU, Regulation (EU) 2023/1230 may also be relevant on a conditional basis. It is not a blanket legal requirement for every robot project in a Thai factory. The Regulation generally applies from 20 January 2027, with specified provisions applying earlier, so the economic operator, product scope, application date, conformity route, and technical documentation should be confirmed for the particular EU-facing project.

Task-based risk assessment sequence

1. List normal operation, loading, unloading, changeover, cleaning, jam clearing, teaching, maintenance, troubleshooting, recovery, and decommissioning.
2. Identify who approaches, where, why, and which energy or hazard can reach them.
3. Include reasonably foreseeable misuse, sensor faults, fixture movement, wrong parts, and restart behavior.
4. Consider inherently safe design, engineering safeguards, administrative control, and training in that order.
5. Record residual risk, competence, inspection, validation, and reassessment after change.

Frequent cell entry is a design input, not just an operator problem. External replenishment, visible diagnostics, and recovery from outside the safeguarded space may reduce both exposure and downtime.

Do not let robot teaching become one employee’s private skill

Robot teaching is more than editing points. Stable operation requires understanding frames, tooling, speed, collision potential, product quality, backups, safe modes, and change approval. Accept training through practical performance, not attendance.

Provide material in the languages used on the floor, including Thai, and align HMI labels, alarms, SOPs, and component names. Keep a searchable alarm list, recovery flow, backup location, and contacts—not only videos. Include a SAT scenario in which an off-shift team recovers without the original project engineer.

Assign an owner for robot maintenance and support

A support contract does not inspect daily cable wear, loose fixtures, dirty sensors, unusual sounds, or air leaks. Before production release, add the cell to the plant’s maintenance system with an asset record, inspection plan, spare strategy, access roles, backups, and escalation contacts.

Maintenance handover package

• final drawings, BOM, equipment models, serials, and warranty start dates;
• complete PLC, robot, HMI, vision, and safety-configuration backups;
• tested restore procedure and required software;
• inspection tasks based on manufacturer recommendations and operating conditions;
• recommended spares, long-lead items, consumables, compatibility, and storage;
• alarm list, first-line diagnosis, stop conditions, and escalation;
• remote-access request, approval, logging, and termination process; and
• version history and authority to deploy changes.

Do not generalize a maintenance interval or spare quantity. Use manufacturer instructions, environment, criticality, procurement lead time, and redundancy. A low-cost component may deserve local stock if it can stop production and takes a long time to obtain. An expensive part should not be stocked without a risk-based case.

Evaluate robot implementation ROI through TCO

A payback calculation based only on robot price hides tooling, integration, safety, shutdown, training, maintenance, and product changes. Update assumptions before the PoC, after FAT, after SAT, and after stabilization.

TCO cost categories

• robot, controller, gripper, fixture, vision, and safety devices;
• PLC/HMI, cabinet, wiring, air, network, foundation, and delivery;
• design, fabrication, integration, programming, risk assessment, tests, and documents;
• shutdown, trial material, scrap, rework, and ramp-up support;
• training, interpretation, travel, software, remote access, and cybersecurity;
• preventive maintenance, failure response, spares, consumables, calibration, and updates; and
• new variants, tooling changes, reteaching, reassessment, and disposal.

Benefit categories

• increased good-output capacity at equivalent quality;
• reduced defects, reinspection, rework, and material loss;
• reassignment of people from dangerous/repetitive work to higher-value activity;
• greater resilience to staffing or demand variation;
• operational, stop, quality, and traceability data; and
• capability to meet customer or new-product requirements.

Avoid assuming that every labor hour disappears. Material handling, supervision, quality, and maintenance may remain or shift. Build base, conservative, and upside scenarios, then test sensitivity to volume, downtime, scrap, support cost, and demand. Apply the company’s finance and tax policies with appropriate specialists.

For the broader investment model, see our robot implementation ROI guide for Thailand. This article focuses on replacing estimates with PoC evidence.

Do not put an unconfirmed BOI incentive into the business case as cash

BOI’s current Smart and Sustainable Industry material describes measures concerning machinery upgrades, automation and robotics, and digital technologies, including conditions related to links with the domestic automation industry. The summary is not an eligibility decision for a particular company.

Confirm the applicant, activity, equipment, application and order timing, domestic value, technical scope, supporting evidence, and accounting/tax treatment against current announcements and BOI guidance. Model the case both with and without an incentive. That shows whether the project is robust to eligibility, approval, timing, or policy changes.

Define Scale, Hold, and No-go criteria before the PoC

Do not scale because the first cell looks impressive. Approve the decision matrix when the PoC starts, then fill it with evidence. The quantitative thresholds should be defined by the plant and aligned with the purchase specification.

Hold is a managed condition, not a hidden failure. Assign the open action, owner, cost, and retest date. A No-go should still transfer the baseline, RFP, test, and risk-assessment learning to the next candidate.

Common SME robot implementation failures

Selecting the most visible task

A showcase task may have too much variability and human intervention. Select on measurable value and learning potential.

Treating the robot manufacturer as the complete system boundary

Cell performance and safety depend on tooling, fixtures, feeding, peripherals, controls, and operations. Read our industrial robot implementation guide based on ISO 10218:2025 for the broader integration lifecycle.

Accepting the fastest cycle

A short run under ideal conditions excludes cleaning, changeovers, starvation, recovery, and quality variation. Test representative shifts and failures.

Concentrating teaching knowledge in one person

Leave, resignation, or transfer can stop the cell. Split access by role, verify multiple people through practical tasks, and standardize change and restore.

Adding safety after installation

Late guarding or safety controls may change access, cycle, and maintenance space. Start with tasks and access reasons during concept design.

Making an incentive the only reason the project works

Eligibility and timing require confirmation. Build a non-incentive scenario or an approval gate before purchase.

Frequently asked questions

Which task should an SME automate first?

Begin with one workcell whose inputs, actions, outputs, and failures are observable, whose variation is manageable, and whose stop can be isolated. Compare candidates on baseline, safety boundary, maintainability, and reuse—not only task difficulty.

How should an SME calculate robot implementation ROI?

Compare full TCO—including integration, safety, shutdown, training, maintenance, spares, product changes, and disposal—with benefits measured at equivalent quality. Update the model with PoC evidence and use several scenarios.

Can robot teaching be left to the integrator?

Advanced changes can remain contracted, but the plant should own safe start/stop, standard recovery, backup, approved minor changes, and escalation. Define permissions and demonstrate competence in acceptance tests.

Does a collaborative robot eliminate the need for guarding?

There is no universal answer. Assess the complete application, including gripped parts, fixtures, process hazards, speed, contact, routes, faults, and maintenance tasks. Select safeguards against applicable requirements and competent risk assessment.

What belongs in a robot maintenance support agreement?

Separate reception, technical response, supported language, on-site arrival, remote access, exclusions, warranty, spares, software, backups, staff replacement, and obsolescence. Also assign internal daily inspection and first-line diagnosis.

Must a PoC finish in 60–90 days?

No. This is an illustrative decision horizon. Long-lead parts, shutdown work, process complexity, and safety validation can extend it. The essential controls are agreed deliverables, gates, corrective action, and a decision date.

Does BOI support apply automatically to a robot project?

No. Current BOI measures may be relevant, but eligibility and benefits depend on the applicant, activity, technical scope, timing, conditions, and approval. Confirm the current rules directly and keep the amount conditional in the business case.

Conclusion

Successful robot implementation for an SME means starting small enough to collect reliable evidence. Select one stable task, measure the current state, define what the plant and integrator own, and place requirements, FAT, SAT, safety, training, maintenance, and TCO gates inside a 60–90-day PoC plan. The goal is not merely motion. It is a cell that operators can run and recover safely, that meets measured quality and capacity at an acceptable total cost, and that creates reusable assets for the next workcell. The ability to choose Hold or No-go protects scarce resources.

TOMAS TECH can support a Thailand plant from candidate-workcell screening and baseline measurement through PoC requirements, RFP/FAT/SAT design, safety and maintenance responsibility, and TCO updates—even before a robot model or integrator has been selected. To discuss a narrowly scoped first workcell, use our contact page.

References

1. Thailand Board of Investment, “Thailand Secures $43.6bn 1H 2026 Investment Surge…” — first-half 2026 application/approval and Smart and Sustainable Industry figures.

https://www.boi.go.th/un/boi_event_detail?language=de&module=news&topic_id=139075

1. Thailand Board of Investment, “Smart and Sustainable Industry” — current measure overview; confirm project eligibility directly with BOI.

https://www.boi.go.th/index.php?language=en&page=smart_sustainable

1. ISO, ISO 10218-1:2025, *Robotics — Safety requirements — Part 1: Industrial robots*.

https://www.iso.org/standard/73933.html

1. ISO, ISO 10218-2:2025, *Robotics — Safety requirements — Part 2: Industrial robot applications and robot cells*.

https://www.iso.org/standard/73934.html

1. ISO, ISO/DTR 20218-3, *Robotics — Safety design for industrial robot systems — Part 3* — under development at approval stage as of September 2026.

https://www.iso.org/standard/90680.html

1. NIST, *Best Practices for the Integration of Collaborative Robots into Workcells Within Small and Medium-Sized Manufacturing Operations*.

https://www.nist.gov/publications/best-practices-integration-collaborative-robots-workcells-within-small-and-medium-sized

1. NIST, *Performance of Emerging Technologies for Robotics* — updated April 23, 2026.

https://www.nist.gov/programs-projects/performance-emerging-technologies-robotics

1. NIST MEP, *Robotics and Manufacturing Automation*.

https://www.nist.gov/mep/robotics-and-manufacturing-automation

1. U.S. OSHA, *OTM Section IV, Chapter 4: Industrial Robots and Robot System Safety* — practical U.S. reference, not Thai law.

https://www.osha.gov/otm/section-4-safety-hazards/chapter-4

1. EUR-Lex, *Regulation (EU) 2023/1230 on machinery* — consolidated version of July 27, 2026; check scope and application dates only for machinery made available or put into service in the EU.

https://eur-lex.europa.eu/eli/reg/2023/1230/2026-07-27/eng/pdf