Redesigning jobs around AI means breaking each process into tasks, setting an AI level from 0 to 5 for each, then rewriting duties and skill sets, following ETDA’s AI Job Redesign Guideline. This is for HR and executives, and gives the working steps and what to watch when AI is used to evaluate employees.
- The AIGC guideline analyses work at task level, where it is clearest what AI takes over and what people must still control.
- AI levels run from 0 (no automation) to 5 (full automation), but every level must name a person accountable for the outcome.
- Three deliverables: a Human–AI Collaboration Workflow, rewritten duties and job descriptions, and a skill set for upskilling and reskilling.
- The guideline says plainly that it must not be used to justify reducing headcount.
- AI that evaluates employee performance meets the PDPC's high-risk criteria: do a DPIA and give staff advance notice with an explanation.
Many Thai companies already use AI in HR work: drafting job ads, screening applications, summarising candidate profiles, answering staff questions about benefits. Yet the job descriptions of the people doing that work have not changed. Nobody has written down what they must now check, what they still decide, and what they are now accountable for.
That gap is exactly what the AI Job Redesign Guideline from the AI Governance Center (AIGC) at the Electronic Transactions Development Agency (ETDA) sets out to close. It is written for HR teams and executives, and it is guidance rather than law.
This article walks through the method step by step, then covers what the guideline leaves out: duties under the Personal Data Protection Act B.E. 2562 (2019) when AI is used to watch or assess employees, drawing on the PDPC's AI guideline.
Why staff resist AI tools the company has already bought
The AIGC guideline explains that choosing a project with a tool such as the AI use case canvas only tells you where AI should go and whether it pays. It does not tell you how the people already doing that work will carry on. Skip that question, and the guideline is blunt about the result.
If an organization only uses the AI use case canvas to select AI projects but does not go on to carry out AI Job Redesign, several gaps may arise, such as: employees not understanding how their duties have changed; some work being taken over by the system without an appropriate new role being designed for it; or there being no skills development plan to support the change — resulting in anxiety and resistance to change, even though the use case itself was well designed.
That paragraph belongs on the first slide HR shows management. It moves the conversation from which system to buy to whose job changes, and what that person needs to do the new job well.
The guideline looks at impact on three levels: task, job position and organisation. It starts at task level, because that is where it is clearest which steps become AI work, which need more human supervision, and how the order of work changes.
Which tasks go to AI and which stay with people?
The guideline sorts human–AI collaboration into three types: human-only, augmentation (people and AI together) and automation (AI in place of people). It then breaks these into six AI levels, 0 to 5, to be assigned task by task.
| Level | What the person does | HR example |
|---|---|---|
| 0 – No Automation | Decides and acts at every step | Talking to an employee about a sensitive matter such as a grievance (author's example) |
| 1 – Assistance | Uses the system's help but controls and acts personally | AI drafts a job ad from the job description; a person checks the wording and posts it (guideline example) |
| 2 – Partial Automation | Checks and confirms every result before proceeding | AI screens applications against criteria with a matching percentage; a person makes the final decision (guideline example) |
| 3 – Conditional Automation | Monitors, ready to take over when the system errs | Benefits chatbot that hands off to staff when a question falls outside scope (author's example) |
| 4 – High Automation | Oversees quality and reviews at the end of the task | Emailing candidate details to the committee after a person has confirmed the previous step (guideline example) |
| 5 – Full Automation | Policy-level oversight and accountability for outcomes | Automatic application status emails (author's example) |
A high level does not mean nobody is accountable. The guideline defines responsibility early on, in chapter 3.
Even where, for certain Tasks, the work is carried out by a system or artificial intelligence technology, ultimate responsibility for the outcome must still be clearly identified as resting with a specific individual, job position, or unit.
In practice, put the accountable job title on every row of the workflow, including rows at level 5. If a row can only say "the system", the redesign is not finished.
How does AI Job Redesign work, step by step?
The guideline frames the work as input, process and output. The input is an AI use case that has already been assessed for value and feasibility. The process has three main steps, which in time order look like this.
- Before you startPick one clear AI use case, identify the process, data and stakeholders, and form a team of five roles: executive, process owner, HR representative, IT and operational staff.Guideline 5.1–5.3
- Step 1Task analysis: break the process into sub-processes and tasks, and identify which AI capability could help with each.Chapter 6
- Step 2Stakeholder review: IT proposes the highest technically feasible AI level, the process owner and staff test it against real work, and the meeting agrees a level per task.Chapter 7
- Step 3Rewrite duties, update job descriptions and define the skill set for development plans.Chapter 8
- After go-liveOperational staff monitor AI quality in each task and report anomalies so the process and system can be improved.Guideline 5.2
Step 2 is where HR's role changes. The guideline casts HR as coordinator and facilitator of the meeting, not as the one who decides on the process owner's behalf.
To ensure that determining the AI Level is comprehensive, this step recommends conducting it in the form of a joint discussion among stakeholders, such as the responsible executives, directly involved operational staff, the information technology department, etc., with the Human Resources (HR) department acting as coordinator and facilitator, in order to summarize the results into a Human–AI Collaboration Workflow for use in the next step.
The documents the guideline asks you to bring to that meeting include the organisation's personal data protection policy and risk assessment criteria. Invite the data protection officer (DPO) from the start, rather than sending the result for comment after everything has been agreed.
The guideline's worked example: a recruiter
The guideline follows an entry-level recruiter throughout. The table condenses the role's duties before and after AI. The last two rows are duties that did not exist before.
| Before | After AI is introduced |
|---|---|
| Draft and post job ads personally | Have AI draft from the job description template, review the wording, choose channels and confirm posting |
| Read applications, check qualifications and decide the screening result | Have AI screen against criteria, review the matching percentage and results, and make the final decision personally |
| Summarise candidate data on the standard form | Have AI summarise, then verify accuracy and completeness before sending to the recruitment committee |
| — | Oversee automated sending of candidate data and track its status periodically |
| — | Ensure candidate data is used in line with the privacy policy and the organisation's AI governance rules |
Upskill or reskill, and what should training cover?
The guideline separates the two. Upskilling adds skills within the same line of work so the existing job is done better, for instance using AI to work faster. Reskilling means learning a new skill set because the role has changed significantly. Its recruiter example draws the line in a revealing place.
According to the example above, for the Recruiter position, using AI tools and writing commands (Prompts) to operate AI tools is considered Upskill, while skills in managing bias and complying with AI Governance guidelines are considered Reskill.
For HR, the message is that a one-day tool course is not enough. Bias and AI governance are treated as new skills that need a proper training plan. The guideline borrows the UK government's AI skills framework and groups skills under three headings. For the recruiter, they look like this.
- Technical skills: use screening and summarising tools correctly, write prompts that produce the right job ad, read and interpret the system's results.
- Responsible and ethical skills: challenge results that conflict with professional criteria, recognise factors that could make the system discriminate against some applicants, and handle candidate data in line with the PDPA.
- Non-technical skills: explain AI results to the recruitment committee, feed real-world problems back to the development team, and keep learning.
Managers have their own skill set, such as managing PDPA compliance in AI-supported processes and assigning responsibility for the impact of AI within the team. Supervisors therefore need training too, not just their teams.
How do you tell staff without triggering fear of job losses?
Start with what the guideline itself says in its chapter on scope and limitations. HR can read this out to staff word for word.
However, the content of this guideline is not intended to suggest that AI should be used in a manner that results in job losses, or to be used as a justification for supporting workforce reductions within an organization.
From the guideline, the author distils four things staff should hear clearly.
- Whose job changes, in which tasks, and at what AI level.
- What people must still do and still answer for. The guideline gives the example of routine workers moving up to decide or to vet what AI produces.
- What the company will teach, when, and who will teach it.
- Whether the AI system will use employees' own data, for what, and who sees the results.
The fourth point is the one HR most often forgets, and it is where personal data protection law comes in directly.
What does the PDPA expect when AI evaluates or monitors employees?
The PDPC's AI guideline lists AI used to evaluate employee performance among high-risk uses that should have a data protection impact assessment (DPIA) before they start. The PDPC text exists only in Thai, so an English gloss follows the quote.
หากมีการดำเนินการเก็บรวบรวม ใช้ หรือเปิดเผยข้อมูลส่วนบุคคลกับระบบ AI ที่มีลักษณะความเสี่ยงสูง (High Risk) และควรจัดทำ DPIA ได้แก่ การนำระบบ AI มาใช้ในกระบวนการตัดสินใจอัตโนมัติ (Automated Decision-Making) ที่มีผลกระทบทางกฎหมายหรือผลกระทบอย่างมีนัยสำคัญต่อบุคคล เช่น การใช้ AI ในการพิจารณาอนุมัติสินเชื่อ การคัดกรองการจ้างงาน หรือการประเมินผลการปฏิบัติงานของพนักงานWhere AI output feeds decisions with significant effects on an employee, such as selection, promotion or appraisal, section 5.4 of the same guideline asks the employer to say so in advance and explain the method actually used.
ผู้ควบคุมข้อมูลส่วนบุคคลควรแจ้งให้เจ้าของข้อมูลส่วนบุคคลทราบล่วงหน้าว่ามีการใช้การตัดสินใจอัตโนมัติหรือผลลัพธ์ของระบบที่มีอิทธิพลชี้ขาดต่อการตัดสินใจ พร้อมคำอธิบายขั้นตอนและหลักการที่ใช้จริงอย่างกระชับและเข้าใจได้เพียงพอให้เจ้าของข้อมูลส่วนบุคคลเข้าใจว่าข้อมูลใดของตนถูกใช้และมีผลต่อผลลัพธ์อย่างไร โดยไม่จำเป็นต้องเปิดเผยอัลกอริทึมหรือซอร์สโค้ด และไม่อาจอ้างความลับทางการค้าเพื่อปฏิเสธการอธิบายโดยสิ้นเชิงOn lawful basis, many employers will rely on legitimate interest under Section 24(5). The PDPC guideline expects a legitimate interest assessment (LIA) and flags a mistake HR teams often make.
ความคาดหมายตามสมควรของเจ้าของข้อมูลส่วนบุคคลเป็นปัจจัยสำคัญ โดยการระบุไว้ในประกาศความเป็นส่วนตัวเพียงอย่างเดียวไม่เพียงพอที่จะถือว่าเจ้าของข้อมูลส่วนบุคคลคาดหมายได้In HR terms, a line in the employee privacy notice saying the company may use technology to analyse performance is not sufficient. The real context matters, including the employment relationship and whether staff know which of their data is being used. The same section reminds readers that where this basis is used, the data subject has a right to object under Section 32(1).
This is performance evaluation that affects pay. Give advance notice, explain the criteria, and let a supervisor with real authority amend the score.
This matches level 2 in the AIGC guideline, but it is still employment screening, which the PDPC treats as high risk, so a DPIA and periodic bias checks are advisable.
Still tell staff they are talking to an automated system, limit access to data, and do not keep conversations longer than needed.
One more point to remember: a supervisor who approves every AI result without looking further is not reviewing it. Section 5.4 of the PDPC guideline requires reviewers to weigh other factors, to have authority to reverse the result, and to be trained on automation bias. That training belongs in the development plan for supervisors.
Once the redesign is done, add the new processing activities to the record of processing activities (RoPA). Section 5.1 of the PDPC guideline says that using data beyond its original purpose requires notice of the new purpose and an updated RoPA. See our guides on choosing a lawful basis and writing a privacy notice.
HR checklist before AI goes live in real work
- Pick one clear use case and form the five-role team the AIGC guideline describes.
- Break the process into tasks and agree an AI level from 0 to 5 for each in a joint meeting.
- Put the accountable job title on every row of the workflow.
- Rewrite job descriptions to add checking, supervising and data stewardship duties.
- Build upskilling and reskilling plans by job level, including supervisors.
- If AI touches selection, appraisal or monitoring of staff, have the DPO complete a DPIA and LIA before launch.
- Update privacy notices for employees and applicants to say what AI is used for.
- Give staff a way to request human review and to contest results, and keep statistics on overturned outcomes.
AIPDPA includes a workplace AI policy with a staff quiz, and a DPIA workspace where HR and the DPO can work together; see it in the demo account. If you have no policy yet, start with our generative AI policy guide, and read the executive view in what the board should ask before approving AI.
Frequently asked questions
Is the AI Job Redesign Guideline binding on private companies?
No. It is guidance from the AI Governance Center (AIGC) at ETDA, to be adapted as each organisation sees fit. Duties under the Personal Data Protection Act that arise when AI uses employee data still apply in the normal way.
Do we have to redesign every role at once?
No. The guideline recommends starting with one clear AI use case and analysing it mainly at task level, where change is most visible and closest to daily work, then extending to job positions and the wider organisation.
Can a small company with a one-person HR team follow it?
Yes. The guideline says some roles can be held by the same person; in an application screening use case, the executive, process owner and HR may be one individual. What should not be skipped is listening to the people who actually do the work before agreeing AI levels.
Do we need employee consent to use AI for performance evaluation?
The PDPC guideline does not say consent is always required and treats consent as a last resort. If you rely on legitimate interest you need an LIA. AI-based performance evaluation is treated as high risk, so a DPIA should come first, and staff should be told in advance with an explanation.
If a supervisor approves every AI result, does that count as human review?
Not if they simply confirm it. Section 5.4 of the PDPC guideline says rubber-stamping is not meaningful review. The reviewer must consider other information, have authority to reverse the outcome and be trained on automation bias.
Law and sources
- AI Job Redesign Guideline — AI Governance Center (AIGC), ETDA (guidance, not law) — 3 · 4.1 · 4.5 · 5.2 · 7.1 · 8.3 · 8.4
- PDPC Guideline on Personal Data Protection in the Development and Use of AI, version 1.0 (guidance, not law; Thai only) — 4.5 · 5.1 · 5.4 · 6.2
Text in the "The text" boxes is copied word for word from AIPDPA's legal knowledge base, transcribed from the Royal Gazette and official documents; every edit is checked by the system against that base, character by character. English wording is an unofficial translation; the Thai text prevails. This article is general information, not legal advice for a particular case.
Do it in the system
Who stands behind this guide?
Written by the AIPDPA legal team · Content owner: Phuwara Krobtaku, Attorney-at-law · Thai Barrister-at-Law · technology & intellectual property counsel
Lawyer's licence no. 477/2558 · Thai Barrister-at-Law · LL.M. in Intellectual Property & IT Law (Gold Medal)