AI in HR: where it helps, from hiring to keeping good people

In short: AI in HR means using artificial intelligence for the repetitive reading, writing and summarising an HR team does: drafting job ads, summarising CVs, answering policy questions, checking attendance data and analysing why people leave. AI drafts and suggests; people still decide who gets hired and how performance is rated. In Vietnam, candidate and employee data has had its own rules under the Personal Data Protection Law since 1 January 2026, so the data rules should be in place before the tools are switched on.

What is AI in HR?

AI in HR is a set of uses that help an HR team handle text and people data faster. A large language model (LLM) can turn a manager's few lines into a job description, summarise a pile of CVs in one consistent format, or answer a new hire's “how many days of annual leave do I get?” from the company's own handbook.

HR software and AI do different jobs. An HR management system (HRM) stores records and calculates attendance and payroll with fixed formulas. AI handles the parts that need reading: CVs, emails, leave requests, review comments, exit reasons.

	HR software (HRM)	AI in HR
What it does	Stores records, calculates time and pay by rule	Reads, summarises, drafts, classifies, suggests
Input	Structured data: employee IDs, clock-in times, pay grades	Free text too: CVs, emails, requests, surveys
Output	Same result every time	A draft or suggestion that someone must check
Main risk	Wrong data entry	Misreading, bias, personal data leaks

AI only works well once HR is digitised. If attendance is still kept on paper there is nothing for AI to check, and if policies are not written down a policy chatbot has nothing to answer from. Companies at that stage should start with the digital transformation guide.

What HR teams use AI for

Most of the value sits in recruitment, internal Q&A and data summaries. The table ranks common uses by risk, so it is clear what can start now and what needs more control.

Task	What AI does	What people keep	Risk
Job ads and job descriptions	Drafts from the manager's brief	Approves requirements, pay, wording	Low
Interview scheduling	Suggests slots, drafts invitations	Confirms the schedule	Low
Policy chatbot	Answers from the approved handbook	Keeps documents current, handles hard cases	Medium
Attendance and leave checks	Flags missing punches, overlapping requests	Confirms with staff, closes the period	Medium
CV screening	Summarises, compares against criteria	Decides who moves forward	High
Performance reviews	Gathers KPI data, drafts comments	Sets the rating, holds the conversation	High
Attrition analysis	Points to groups at risk	Changes policy, never labels individuals	High

Recruitment. AI saves the most time on drafting and sorting: job ads, competency-based interview questions, invitation emails, and one-paragraph CV summaries in a fixed format (relevant experience, skills, gaps against the brief). The decision to reject a candidate should stay with a person.

Onboarding and HR chatbots. New hires ask the same questions in their first week: payday, how to request leave, who approves expenses. An internal chatbot answers from the handbook and quotes the source section, then hands anything personal (pay, discipline, health) to an HR officer. It works like an AI agent searching internal documents, and it is only as accurate as the documents behind it.

Attendance and payroll. Payroll should be calculated by software using formulas, not estimated by AI. AI helps one step earlier, by flagging days with a clock-in and no clock-out, leave that overlaps a business trip, or shifts over the internal overtime limit, so payroll staff work through exceptions instead of every row.

Performance with KPIs and OKRs. AI pulls KPI results from sales, project and service systems into one view, summarises a manager's notes and drafts feedback for the manager to rewrite. For OKRs it can make objectives measurable and spot duplicated key results across teams. The final rating and the conversation belong to the line manager.

People analytics. Combining hiring, attendance, review, pay and exit data answers management questions: after how many months do people tend to leave, which team runs long overtime, which roles take longest to fill. Use the results at group level to fix policy. Labelling individuals as “likely to quit” is often wrong and damages trust when it leaks. Choosing metrics and dashboards is covered in building a data-driven business.

Is AI CV screening fair?

AI screening is only as fair as the criteria and the human checks around it. Models learn from past data, so whatever bias sits in past hiring tends to carry over.

The best-known example is Amazon's experimental CV-rating tool. According to Reuters (2018, as republished by CNBC), it was trained on CVs submitted over a ten-year period, mostly from men, and learned to downgrade CVs containing the word “women's”; Amazon disbanded the team. A small company pasting CVs into a chat tool and asking for “the best candidate” is exposed to the same kind of inherited bias.

Before using AI for screening:

1. Write criteria from the job itself and leave out age, gender, hometown, marital status and photos.
2. Remove names, photos, birth years and addresses before CVs go into any AI tool. This reduces bias and the amount of personal data shared.
3. Ask the AI to assess each criterion with a reason. A score without a reason is not used.
4. After each round, read some of the lowest-ranked CVs to catch wrong rejections.
5. Compare pass rates across candidate groups and review the criteria if they diverge sharply.

Employee data and the law in Vietnam

Using AI on candidate and employee records is personal data processing. Vietnam's Personal Data Protection Law (No. 91/2025/QH15) took effect on 1 January 2026, together with Decree 356/2025/ND-CP, which replaced Decree 13/2023/ND-CP (VCCI). Article 25 of the law deals specifically with recruitment and employment.

Stage	What Article 25 requires (summary)	What to do when AI is involved
Recruitment	Ask only for information needed to recruit, use it for that purpose, with the candidate's consent; delete it if the candidate is not hired unless agreed otherwise	Say in the job notice that AI tools may help review applications; schedule deletion, including copies held in AI tools
Employment	Keep data for the period set by law or agreement, in line with labour law	Set permissions so chatbots only read what the asking user may see
Monitoring technology	Only lawful measures that employees clearly know about; no use of unlawfully collected data	Give written notice of face-recognition attendance or computer monitoring; never let AI quietly read staff email or chat
Termination	Delete the data unless an agreement or law says otherwise	Check the HR system, AI tools and stray spreadsheets

Decree 356 lists health status and biometric data as sensitive personal data. Its Article 41 lets small enterprises and start-ups opt out of certain obligations for five years from the law's effective date, but not if they directly process sensitive data. A company that clocks staff in by face recognition should not assume it is exempt.

The Law on Artificial Intelligence (No. 134/2025/QH15), in force since 1 March 2026, states that AI must not replace human authority and responsibility, and requires deployers of high-risk systems to keep human intervention possible (English text). The Prime Minister issues the high-risk list, so ask vendors of screening or appraisal tools how their product is classified. Broader rules are in AI governance and data security.

Legal points were checked in November 2026 and are for reference only. Take legal advice before introducing biometric attendance or employee monitoring.

Ready-to-use prompts for HR

These work with ChatGPT, Gemini, Copilot or whatever AI tool the company already uses. Only paste anonymised data, and only into accounts the company controls.

You are a recruiter at a [industry] company in [city]. Write a job ad for
[role] from these points: [main duties], [must-haves], [nice-to-haves],
[salary range], [benefits]. Plain, accurate tone. No requirements about age,
gender or appearance. Separate must-haves from nice-to-haves. Max 300 words.

Below are the criteria for [role] and [number] CVs with names, photos,
birth years and addresses removed. For each CV, add one table row: CV code,
years of relevant experience, each criterion met/not met with a quote from
the CV, and one follow-up interview question. Do not rank and do not
recommend hiring. If the CV does not say, write "unclear".

The second prompt deliberately stops the AI from ranking. It lays out the facts and leaves the comparison to the recruiter.

Where to start

1. Pick a low-risk task. Job ads or a policy Q&A bot. Mistakes there are easy to spot and cost nobody a job.
2. Clean up documents and data. The handbook and pay policy must be current with one owner, and records should live in one system. The business management software guide has selection criteria.
3. Write one page of AI rules for HR. Which tools and accounts, which data never goes in (payroll, health records, non-anonymised CVs), and who approves output before it leaves the team. A short internal session usually beats a long course at this stage.
4. Pilot and measure. Record a baseline first: time to write a job ad, days from posting to first interview, share of chatbot answers that need no human.
5. Then extend to CV summaries, attendance checks and attrition analysis, with the fairness and data steps above. Staff will worry about being scored or replaced, so leadership should say plainly what AI is and is not used for. See change management in digital transformation.

Hold off if records are still on paper, if you hire only a few people a quarter, if nobody owns checking the output, or if the real aim is to watch staff without telling them.

Common mistakes

• Pasting CVs or payroll into personal, free AI accounts, where the data may be used for training depending on the service terms.
• Letting AI reject candidates on its own, so nobody can explain a rejection and bias goes unnoticed.
• Running a chatbot on an outdated handbook, which then gives wrong answers with full confidence.
• Giving individuals an “attrition risk” score, which changes how managers treat them.
• Deleting a rejected candidate from the HR system but leaving the CV in an AI chat history.

FAQ

Will AI replace the HR team? AI takes over repetitive drafting, look-ups and data checks. Interviews, reviews, employee relations and decisions about people still need HR staff and line managers.

Does a small company need dedicated software to use AI in hiring? Not at first. For a few roles a quarter, a company-managed AI account and a set of prompts is enough. Recruitment software with AI makes sense once reading applications by hand becomes the bottleneck.

Is it legal to read CVs with ChatGPT in Vietnam? Article 25 of the Personal Data Protection Law requires candidate data to be used for recruitment with the candidate's consent. The safer route is to disclose AI use in the application process, anonymise CVs first and use a business account that does not train on your data.

Does face-recognition attendance need employee consent? Facial data is biometric and counts as sensitive under Decree 356/2025/ND-CP, and Article 25 requires employees to clearly know about monitoring technology. Use written notice and consent, define where data is stored and when it is deleted, and get legal advice on the details.

How should we measure AI in HR? Tie metrics to the task given to AI: time from posting to interview, hours spent drafting each month, share of internal questions answered correctly, attendance errors caught before payroll closes. Measure before and after in the same way.

---

HAT Solutions advises HR teams on applying AI: choosing what to automate first, building an internal policy chatbot on your own documents and setting the employee-data rules that go with it. Talk to us about your HR workload. For the wider picture, read AI for business and AI tools for the office.