HR teams in mid-sized companies often run on thin resources. One or two people handle recruitment, attendance, leave, contracts, BPJS, and employee questions about overtime rules. When a vacancy opens, hundreds of applications can arrive in a week, while routine admin still has to get done.
AI for HR can take some of this load, especially repetitive, text-heavy work. HR also holds the most sensitive data in the company, though, and its decisions affect people's lives directly. This article looks at where AI helps, where to be careful, and the ground rules worth agreeing on before you start.
Supporting application screening
Reading hundreds of CVs is exhausting, and the quality of judgement tends to slip after the hundredth one. AI can help by summarising each CV against criteria you have set, such as experience in a particular field, required certifications or language skills. The result is a short summary that makes candidates easier to compare.
The key word is support. AI prepares the material and the recruiter decides. Avoid any setup that rejects candidates automatically without a person reading the reason. A CV with an unusual format, a career gap for family reasons, or relevant experience from a different field can easily be missed if the decision is left entirely to a machine.
Write down your assessment criteria before using AI, and apply the same criteria to every candidate. It also makes the process easier to explain if a candidate asks.
Before CVs go to the AI, consider removing photos, dates of birth and marital status if they play no part in the assessment. Less personal data leaves the company, and there is less chance of the assessment being swayed by things unrelated to the job.
Drafting job descriptions
AI is good at turning a manager's rough notes into a draft job description. A cleaning services company hiring an area supervisor, for example, can provide a list of duties, the work area and the key requirements, and the AI turns them into a tidy job ad.
Check the result carefully. Make sure every requirement is genuinely needed for the job, and drop requirements that are irrelevant and may come across as discriminatory, such as age limits or appearance criteria unrelated to the work. Adjust the language to the people you want to reach. An ad for warehouse staff and one for a financial analyst should sound different.
Answering policy questions
Questions such as how much leave is left, how to claim for new glasses, or whether overtime on public holidays is paid differently reach HR again and again. An AI assistant that reads the company regulations, SOPs and employee handbook can answer them at any hour.
For answers people can trust, the assistant should use RAG: it retrieves the relevant part of the documents and answers from it, with a citation to the source. Employees can then check the quoted clause themselves. If the answer is not in the documents, the assistant should say so and point the employee to HR. Make sure the AI always reads the latest version of each document.
HR administration
Much HR admin is text that follows familiar patterns. AI can prepare drafts of employment reference letters, offer letters, internal announcements and onboarding checklists for new hires. It can also summarise exit interviews or employee surveys into the main themes for management to read. A clinic with several branches, for instance, can use AI to compile a monthly leave summary from attendance data, so branch heads no longer have to work it out by hand.
For documents with legal weight, such as employment contracts or warning letters, limit AI to helping with the draft. The final content still needs review by HR and, where needed, legal counsel.
Fairness and employee privacy
Employee and applicant data is personal data protected under UU PDP. Some of it is highly sensitive, such as health information, disciplinary records or family details. These are the ground rules worth agreeing on before AI is used in HR:
- Use company AI accounts whose data terms you have checked, and do not allow personal accounts for employee data.
- Keep health, religion and other sensitive data out of AI unless it is truly necessary and there is a clear basis.
- Mask NIK numbers, bank account numbers and addresses when the task does not need them.
- Decisions on hiring, promotion and termination are always made and signed off by people.
- Review screening results regularly to see whether any group of candidates keeps being filtered out without a clear reason.
- Tell applicants and employees when AI is used in a process that concerns them.
- Limit access so each person only sees the data within their remit.
These rules are easier to apply when management and employee representatives discuss them from the start.
AI on top of your HR data
If your employee data is already well organised, for example in Trellis through its Hana HR app, the next step is connecting AI to that data and those documents safely. Averoa builds AI that answers with sources, sets access per team, and masks personal data before it reaches the model. See our company AI page for details.
