To eliminate knowledge silos in manufacturing is not to make experienced people unnecessary. It is to make decision conditions, exceptions and outcomes traceable so another qualified team member can reproduce routine work and escalate the hard cases correctly. This guide gives Japanese-managed factories in Thailand a practical 90-day path from scattered paper and Excel knowledge to governed, multilingual operations without taking unnecessary production risks.
Why eliminating manufacturing knowledge silos is urgent
“The sequence cannot be changed when the planner is absent.” “Only the line leader knows whether this defect is acceptable.” “The Excel total changes before every month-end meeting.” These are not failures of individual capability. They show that essential decisions sit outside the formal operating system. Even when a work instruction exists, the employee’s memory remains the real system if the applicable conditions, priority rules, exception authority and result records are disconnected.
This risk is amplified in a Japanese-owned factory in Thailand. Japanese expatriates, Thai supervisors, operators, equipment vendors and external maintenance teams may view the same process through different languages and accountabilities. Translating a Japanese instruction into Thai does not automatically transfer the reason behind a decision. When a product-mix change, employee turnover, equipment renewal and an audit arrive together, missing tacit knowledge can affect delivery, quality and safety at the same time.
The World Bank’s report released on 3 September 2026 says Thailand’s move toward higher-value growth depends on technology adoption, innovation, skills and stronger domestic value added. Its February 2026 Thailand Economic Monitor also observes that past manufacturing productivity gains have stalled or reversed. Neither finding means a single application will solve factory productivity. They do explain why converting operational decisions into reusable information is part of the productivity foundation.
The overview of Japan’s 2026 manufacturing white paper reports 54.8% for “continued employment of older workers” as a skills-transfer measure among the survey respondents. This is not a statistic for Thailand or for every manufacturer. The useful lesson is that retaining experts and turning their reasoning into an organizational asset are complementary. While experts are still present, the factory has its best opportunity to observe real exceptions and update standards with their knowledge.
Diagnose the symptoms before choosing a solution
“Knowledge dependency” is too broad to be a project scope. If management immediately concludes that the cause is a lack of manuals or training, the project may produce hundreds of PDF pages without changing a single decision. Start by separating visible symptoms and hidden causes.
| Symptom | Likely hidden cause | Evidence to inspect first |
|---|---|---|
| Only one planner can resequence production | Constraints and priorities are undefined | Schedule changes, shortages, setup time |
| Defect disposition varies by person | Specifications, boundary samples and authority are separated | Inspection records, photos, MRB approvals |
| Recovery time differs for the same alarm | Symptom, cause, action and outcome are not linked | Alarm history, maintenance log, parts history |
| Meeting totals differ by workbook | Source of truth, cut-off time, formula and edit authority are unclear | File versions, formulas, editor, timestamp |
| Paper reports cannot support improvement | Codes and record granularity are inconsistent | Free text, blanks, terminology, re-entry history |
| Meaning changes after translation | Controlled terminology and approved source text are absent | Source version, translation, approver, effective date |
The red flag is not “an expert is needed”; it is “the decision cannot be reproduced”
Expert judgment is normal in a complex factory. The problem is that the next qualified person cannot reconstruct why a decision was made, when the answer should have been different, who could approve the exception, and what happened after execution. Use four diagnostic questions:
- Can another employee verify the input conditions?
- Are the normal rule and exception conditions explicit?
- Is the authority boundary and escalation destination defined?
- Can the result be found and compared later?
If one element is missing, training alone is unlikely to create repeatability. When all four exist, experts spend less time answering routine questions and more time on complex improvement and coaching.
Choose one process: a small factory DX roadmap
A common mistake in a factory DX roadmap is to declare that every process will become paperless. The scope becomes too wide, the team spends months documenting the current state, and live-production constraints are overlooked. For a 90-day proof of concept, select one decision job using four lenses: business impact, dependency on a person, evidence availability and change risk.
Good candidates include daily sequence adjustment, material release, first-piece approval, first-line equipment troubleshooting, work-in-process location and reinspection disposition. Do not scope “the production department.” State what input is used, who makes which decision and at what moment.
Use a selection score to start the discussion
| Dimension | Example of a high score | Caution |
|---|---|---|
| Business impact | Direct link to late delivery or downtime | A monetary value is not mandatory yet |
| Personal dependency | Backup staff cannot explain the reasoning | Do not confuse dependency with task frequency |
| Evidence availability | Input, choice and result can be recorded in 90 days | Manual capture is acceptable before sensors |
| Change risk | Read-only or advisory pilot is possible | Exclude safety interlocks from the PoC |
For planning-related work, our guide to production planning simulation for factories in Thailand helps break an experienced planner’s intuition into demand, stock, capacity, labour and setup constraints. Before selecting software, review the typical pitfalls in production management system implementation.
Build a decision information model, not another manual
Scanning paper into PDF or moving Excel to a shared drive does not remove dependency. The smallest useful unit is not a document; it is a decision record. At minimum, keep the following information together.
| Information | Example | Control point |
|---|---|---|
| Subject | Equipment, part, lot, operation | Use a master ID to prevent naming variation |
| Input condition | Stock, due date, measurement, alarm | Store timestamp and unit |
| Decision rule | Priority, threshold, combination | Keep readable text and structured logic |
| Exception | Out of specification, missing value, urgent order | Avoid one generic “Other” category |
| Authority | Operator, checker, approver | Define proxy approval and expiry |
| Execution | Selected action and actual change | Record who did what and when |
| Outcome | Pass/fail, elapsed time, loss, recurrence | Link to the decision with one ID |
| Basis | Standard, drawing, photo, prior case | Keep version, owner and effective date |
The model’s value is that reasoning and outcome stay connected. “Adjusted temperature” is not enough. Capture the alarm and measurement observed, which approved rule was applied, the new setting, the approving role, and the result on the next lot. The same evidence can then support operator training, similar-case search, standard revision and a future governed AI retrieval layer.
Reduce human error in manufacturing by changing controls
To reduce human error in manufacturing, “pay more attention” and permanent double checks are weak controls. Restrict input choices, fix measurement units, validate ranges, require evidence photos, route approvals, and verify the part or machine automatically. Design both error prevention and error detection.
More warnings are not always safer. When every screen displays a warning, operators learn to acknowledge every message. Each warning needs severity, reason, required response and reset authority. Review frequency and override rate. Automated data can also be wrong: missing sensors and mismatched master data require a safe response such as “stop the recommendation” or “request human confirmation.”
Recognize the limits of Excel production management
Excel is flexible and excellent for early experiments. The risk is not the tool itself; it is continuing to treat an uncontrolled workbook as the operational source of truth after several people, processes and languages depend on it.
The limits of Excel production management become visible when identical filenames multiply across email and shared folders, copied formulas differ, only the macro author can repair the file, changes after cut-off cannot be traced, or every workbook has a different item master.
Separate Excel’s role instead of banning it
Keep Excel for hypotheses, personal analysis and prototype templates. Move the operational source of truth toward a database, MES or production-management platform when several of these conditions apply:
- Multiple employees edit at the same time.
- Approval or access must differ by row or field.
- Version history must serve as audit evidence.
- Equipment or ERP data must be integrated automatically.
- Lot, process and result require unique identifiers.
- Offline synchronization conflicts must be governed.
Do not reproduce the old spreadsheet screen column for column. For every field, ask who enters it, when, from which source, in which unit, and for which decision. Remove unused fields and duplicate calculations. Separating operator input from management reporting makes both the screen and accountability simpler.
Move from paper-based factory operations without disrupting production
A paperless factory project can increase risk when every form is abolished at once. Paper may still serve as a fallback during a network outage, a physical identification aid, or evidence required by a customer. For each form, verify the purpose of capture, checking, approval, retention, retrieval and disposal.
| Stage | Action | Exit condition |
|---|---|---|
| 1. Observe | Follow one week of form movement and annotations | Actual users and exceptions are visible |
| 2. Classify | Separate source record, job aid and fallback | An owner decides whether each can retire |
| 3. Digitize minimum | Implement mandatory data and controlled codes | Reconciliation finds no missing requirement |
| 4. Parallel run | Compare paper and digital for a fixed period | Differences have explainable causes |
| 5. Cut over | Declare one source of truth | Training, authority and outage procedure work |
| 6. Retire | Stop obsolete forms and update retention rules | Old versions are removed from the workplace |
Parallel operation provides confidence, but without an exit condition it creates permanent double entry. Agree on the decision date and review discrepancies, missing entries, approval delays and failed synchronization before the process and quality owners sign off.
Design permissions, versions and audit evidence first
Putting standards in the cloud often creates one of two extremes: everybody can edit, or nobody owns updates. Define the document owner, content approver, translation approver, readers, effective date and next review date. Shop-floor devices should normally show only the current effective version. Retain obsolete versions for audit, but keep them out of normal search results.
ISO 10013:2021 provides guidance on documented information for quality management systems. It should not be presented as a new certification requirement. For the next edition of ISO 9001, the ISO committee page should be treated as under publication/expected as of September 2026. Do not implement unissued “ISO 9001:2026 requirements” as final facts. Base certification work on the currently valid standard, customer requirements and the company QMS, and confirm transition arrangements with the certification body.
Record the difference and its impact, not just a version number
Keep the reason, changed clauses, affected products and equipment, required training, approver and exact expiry of the previous version. Every translated version must share the same revision ID so that an updated Japanese original cannot leave the Thai instruction obsolete. Give emergency revisions an expiry and later decide whether to adopt them permanently.
Treat offline and multilingual work as core requirements
Factories have weak Wi-Fi zones, metal structures, shared terminals, gloves and night shifts. A long online-only form will not survive reality. Keep the minimum necessary work package and master data on the device. Attach timestamp, device, operator and synchronization status to offline records. When two devices change the same lot, avoid a simple last-write-wins rule; use field-level conflict rules or supervisor review.
Multilingual operation begins with terminology control, not unlimited free translation. Give parts, operations, machines, defects, actions and statuses stable codes; switch only the display between Japanese, English, Thai and, where required, Vietnamese. Ask actual operators to read short production phrases. Replace textbook wording with language that has one operational meaning.
The OECD Skills Strategy Thailand highlights adult learning, labour-market relevance, flexible learning that reduces time barriers, and better skills-information systems. In a factory, this supports short guidance, photos and prior cases available in the work screen rather than relying only on long classroom sessions. The ILO publication on transformational change likewise provides a useful workforce-transition perspective. Digitalization is more likely to last when employees participate in designing the new work.
A 90-day factory DX roadmap
The purpose of a 90-day PoC is not to complete the whole factory. It is to test whether one decision becomes more reproducible, usable and controlled. Every quantity below—including case counts, pilot cohort and improvement thresholds—is an example target that must be adjusted to the factory’s baseline and risk.
Days 1–15: diagnose and establish the baseline
- Select one decision and define its start, finish and owner.
- Collect about 20–30 prior cases as an example; adjust to actual event frequency.
- Measure decision time, clarification contacts, rework, downtime and recurrence.
- Use the same interview questions in Japanese and Thai.
- Define the safety and quality boundaries that always require human approval.
Days 16–30: model the decision and govern the standard
- Define subject, input, rule, exception, authority, execution, outcome and basis.
- Create the terminology set and master IDs.
- Define current, obsolete and emergency version handling.
- Design responses to missing input, abnormal values and synchronization conflicts.
- Prepare normal, exception and outage acceptance-test scenarios.
Days 31–60: run a read-only or advisory pilot
- Begin with prior-case search, checklists and recommended-action display.
- Do not allow the system to change equipment directly.
- Pilot with approximately two to five users as an example; adjust by shift and frequency.
- Test offline mode, shared devices, night shift and every required language.
- Record why users did not use the system, not only successful transactions.
Days 61–90: compare and decide expand, redesign or stop
- Compare with the baseline using the same KPI definitions.
- Review differences between expert and backup decisions.
- Classify wrong suggestions, obsolete standards and permission violations.
- Consider only approved, low-risk actions for later automation.
- Decide whether to expand, redesign the same process or stop.
Ninety days is also an example. Equipment modification, customer approval or regulatory review may require more time. High-frequency form work may produce enough evidence sooner. Comparable evidence matters more than protecting a calendar promise.
KPI and ROI: do not count saved minutes twice
ROI based only on saved hours multiplied by labour cost is often overstated. Distinguish whether time was actually converted into lower overtime, less outsourcing, additional output or improvement work. Monetize quality or downtime avoidance only when both probability and unit-cost evidence are available.
| KPI | Example definition | Validation question |
|---|---|---|
| Decision reproduction rate | Share matching an approved conclusion under the same conditions | Who approved the answer set? |
| First-pass completion | Share completed without another clarification | Are abandoned records excluded? |
| Exception escalation rate | Defined exceptions correctly returned to a person | Lower is not automatically better |
| Time to valid standard | Time from search start to opening the current version | Is the same start point used? |
| Rework rate | Share requiring re-entry, reinspection or reapproval | Is the process boundary fixed? |
| Obsolete-version use | Count of expired documents actually used | Are offline devices included? |
| Adoption rate | Share of target decisions using the new method | Is forced login excluded? |
Make the ROI equation transparent
An illustrative structure is:
Annual benefit = value of time actually redeployed + avoided rework cost + avoided downtime loss + reduced audit preparation − incremental operating cost
Publish the assumption, source, equation and sensitivity. “Five minutes saved per case” has different credibility if it is an observed median rather than a self-reported estimate. Show low and high transaction volumes and confirm the business case does not work only under its most optimistic assumptions.
A depa Thailand article published on 23 April 2025 contains the figures 70%, 87.33% and 69.33% for the respondents and questions described in that article. Those values must not be relabelled as a Thailand-wide factory DX adoption rate or as a return forecast. Use the plant’s own baseline for an investment decision; external percentages provide context, not a guarantee.
RFP requirements and acceptance criteria
An RFP should describe operational ownership and failure behavior, not only a screen-function list. A checkbox for “multilingual” or “offline” does not define the level required on the shop floor.
Minimum RFP content
- Target process, target decision and explicit exclusions.
- Users, roles, delegated approval and segregation of duties.
- Inputs, source of truth, refresh frequency, unit and identifier.
- Normal rules, exceptions, stop conditions and escalation.
- Screens and terminology required in Japanese, English, Thai and Vietnamese.
- Offline data scope, encryption, synchronization and conflict handling.
- Version control, audit log, retention and export.
- ERP, MES and equipment interface, including failure isolation.
- Performance, recovery, backup and support hours.
- Training, operational handover and exit conditions for configuration and data.
Accept repeatability, not just “the feature works”
| Test area | Example acceptance criterion |
|---|---|
| Normal decision | Approved scenario shows the expected result and reason |
| Exception | Missing, abnormal or unauthorized input stops and alerts the named role |
| Multilingual | Every language displays the content under the same revision ID |
| Offline | Save, resend, duplicate prevention and conflict resolution work on device |
| Version | The effective version changes on time and preserves history |
| Audit | Decisions, approvals and changes can be exported by identifier |
| Migration | Master count, blanks, duplicates and code mappings reconcile |
| Outage | ERP failure produces a safe state and later reconciliation |
Do not accept a demonstration using only sample screens. Test representative parts, night shift, shared terminals, a slow connection and actual exceptions. Keep inputs, expected results, actual results, differences, corrective action and retest evidence.
After go-live: standardization is an operating process
Go-live is not the finish line. The standard owner reviews expiry, quality teams analyze exceptions, IT monitors permissions and interfaces, and production supervisors investigate non-use. A monthly review should see KPI, unresolved exceptions, obsolete-version access, translation backlog, training completion and master-data mismatches together.
Instead of asking an expert to “write down everything,” observe a real decision and ask: Which value did you look at? What was different from normal? Who would you consult? How do you know the outcome was acceptable? Let a facilitator structure the record while the expert validates content. Recognize knowledge contribution as improvement work, not extra clerical work.
Frequently asked questions
What does eliminating knowledge silos in manufacturing aim to achieve?
It does not aim to make people interchangeable. Routine decisions should be reproducible by another qualified employee; exceptions should reach the correct expert; and outcomes should improve the next version of the standard.
Where should a factory start to reduce human error in manufacturing?
Start with one high-severity or high-frequency decision for which inputs and results are observable. Prefer choice restriction, unit control, range validation, approval, automatic matching and safe stop conditions over reminders alone.
How do we recognize the limits of Excel production management?
Simultaneous editing, field-level permissions, audit versions, automated ERP or equipment links, unique lot IDs and governed offline conflicts are strong signs. Keep Excel for analysis while moving the operational source of truth to a controlled platform.
Should paper-based factory operations be stopped immediately?
No. Separate paper used as the source record, job aid or outage fallback. Compare paper and digital for a fixed period, resolve discrepancies, and then declare one source of truth with an approved outage procedure.
What should a 90-day factory DX roadmap complete?
It should prove that one decision model, authority design, version control, multilingual/offline workflow, KPI and acceptance test work in the real environment. All sample counts, cohort sizes and improvement targets in this guide require adjustment.
Must we implement ISO 9001:2026 now?
As of September 2026, treat the next edition as under publication/expected on the official ISO page, not as final requirements. Work to the valid standard and company QMS, then confirm formal transition guidance with the certification body.
Conclusion: make decisions reproducible without removing expertise
Eliminating manufacturing knowledge silos is not a project to remove people. It connects conditions, exceptions, authority, version, execution and outcome so the team can reproduce normal work and route difficult cases to experts. Keep paper and Excel where they have a defined role, establish one operational source of truth, design offline and multilingual work from the beginning, and test one process with a controlled 90-day PoC. Measure against the factory’s own baseline, not an external case-study percentage.
If your factory in Thailand is still selecting a process, documenting the current state, or drafting PoC, RFP and acceptance criteria, TOMAS TECH can help define a practical starting point around your live operation and existing systems. Contact us to discuss the planning stage.
References
- METI: 2026 White Paper on Manufacturing Industries
- METI: 2026 manufacturing white paper overview PDF
- World Bank: World Bank Group Report Identifies New Engines for Thailand’s Next Phase of Growth
- World Bank: Thailand Economic Monitor, February 2026
- OECD: OECD Skills Strategy Thailand
- ILO: Navigating transformational changes and transitions
- ISO: publication status of the next ISO 9001 edition
- ISO: ISO 10013:2021
- depa Thailand: article published 23 April 2025
This article reflects public information available on 7 September 2026. Confirm the application of standards, certification, law, employment rules and personal-data obligations with qualified advisers and relevant authorities in Thailand.
