AI in HR means using AI to take routine load off the people team: answering policy questions, drafting adverts and letters, reading personnel files and summarising workforce data. Uses that shape decisions about people, such as shortlisting candidates, scoring performance or dismissals, are high-risk under the EU AI Act and trigger a right to object to automated decisions under Türkiye's data protection law (KVKK), so a human makes the final call.
Where does the HR team's week actually go?
Look at a typical week for an HR specialist in a mid-sized company. "How much annual leave do I have left?", "Can I return part-time after maternity leave?", "When does the travel allowance show up on my payslip?" The same questions arrive by email, phone and messaging apps, and the same answers are written again every month. Add job adverts, hundreds of CVs, expiring medical certificates and month-end timesheet checks, and the week is gone.
AI shows up in this picture with two very different faces. The first is a set of low-risk helpers that speed up repetitive, text-heavy work. The second is systems that produce judgements about people: who gets an interview, who is promoted, who is a "flight risk". Treating both groups as one project, with one level of care, is where most HR AI initiatives go wrong.
This article focuses on the HR function. If you are still deciding which process in the business to tackle first, start with our guide on where to start with AI. For examples in other departments, see our list of AI use cases in business.
The cost of starting without a plan
Employees are not waiting for HR to decide. They are already using AI, mostly without training or rules:
According to the 2024 Work Trend Index from Microsoft and LinkedIn, only 39% of people who use AI at work have received AI training from their company, while 66% of leaders say they would not hire someone without AI skills.
Türkiye's data protection authority, KVKK, has flagged the same gap. Its notice of 5 March 2026, Use of Generative AI Tools in Workplaces, warns that these tools are often used on the basis of individual preference rather than a corporate policy, which makes them hard to monitor. An HR specialist pasting candidate CVs or a salary table into a personal chatbot account is exactly that risk in practice. Nor are HR complaints hypothetical: the KVKK 2025 Annual Report records 244 complaints related to the human resources sector.
Meanwhile the skills clock is running. PwC's 2025 Global AI Jobs Barometer found that the skills employers seek in the most AI-exposed occupations are changing 66% faster. On the employee side, that change shows up as anxiety: in BCG's AI at Work 2025 survey, 46% of employees at organisations undergoing comprehensive AI-driven redesign worried about job security, against 34% at less advanced companies. Managing that anxiety is also HR's job; our article on human–AI collaboration at work shows how to split tasks between people and AI.
AI in HR use cases by risk level
The table below groups common HR scenarios by what the AI actually does and whether it produces a judgement about a person. The risk column is a prioritisation aid, not legal advice.
| Scenario | What the AI does | Risk level | Human role |
|---|---|---|---|
| HR policy and leave assistant | Answers from the staff handbook, leave procedure and benefits documents, citing the source | Low | Keeps documents current, closes unanswered questions |
| Adverts, announcements and letters | Drafts job adverts, rejection or offer emails from a role profile | Low | Reviews and sends |
| Personnel file extraction | Pulls dates and fields from scanned contracts, diplomas and certificates | Medium (may include special category data) | Confirms uncertain fields |
| Workforce data summaries | Summarises turnover, absence and overtime reports in plain language | Medium | Interprets and decides actions |
| CV screening and candidate ranking | Filters, scores or ranks applications | High (EU AI Act employment area) | Reviews every rejection with reasons |
| Performance and attrition scores | Generates scores or predictions from employee behaviour | High | Treats output as a signal, never a decision |
| Workplace emotion recognition | Infers emotions from face, voice or text | Unacceptable (banned in the EU) | Do not use |
The first two rows are the safest place to begin. A policy assistant follows the logic of an enterprise LLM built on RAG: it retrieves the relevant section of a document first, grounds its answer in it and shows the source. When the handbook does not cover a question it says so, which keeps AI hallucinations harmless in an HR setting.
Field evidence points the same way. In a 2023 NBER study of 5,179 customer support agents, an AI assistant delivered its largest gain, 34%, among novice and less experienced agents, and employee retention improved. It was measured in a different function, but it shows how much fast access to knowledge is worth during onboarding. We collect other field and experimental findings in our review of AI productivity impact research.
The high-risk zone: recruitment, CV screening and performance reviews
Under Annex III of the EU AI Act, systems used for recruitment and selection, filtering applications, decisions on promotion and termination, allocating tasks, and monitoring or evaluating performance are high-risk. According to the European Commission, the Digital Omnibus postponed the high-risk rules for employment systems to 2 December 2027.
Emotion recognition in the workplace, outside medical and safety purposes, has been prohibited under Article 5 since February 2025. Our guide to the EU AI Act for Turkish companies explains how firms with EU candidates, staff or customers are caught.
Companies operating only in Türkiye are not off the hook. Article 11 of Law No. 6698 (KVKK), Türkiye's equivalent of GDPR, gives individuals the right to object to an outcome against them that results solely from automated analysis of their data. A candidate rejected purely on an algorithm's score can challenge that result.
We cover lawful basis, notices and retention for candidate and staff data in employee personal data under KVKK, and AI-specific data questions in AI and KVKK.
Beyond the law there is a data problem. A model trained on past hiring decisions learns whatever bias those decisions contained, then applies it at scale. Move into this group only once low-risk uses are bedded in and your data inventory and governance are ready.
Six steps to introduce AI in HR
- Write the usage policy. Define which tools may be used with which data, and which candidate and employee data never goes into any external tool. Our company AI acceptable use policy guide gives you a template.
- Count the repeat questions. Log every question HR receives for a month and group them by topic. The ten most frequent topics are the first scope for a policy assistant.
- Consolidate the documents. Staff handbook, leave procedure, benefits, health and safety instructions and onboarding material should sit in one current source. Outdated versions mislead the assistant.
- Start with a low-risk pilot. Run a policy assistant or drafting support in one department, and review sources and unanswered questions every week.
- Measure the effect. Record question volume and response time before the pilot and compare afterwards. For the method, see our guide to measuring AI project ROI.
- Decide on high-risk scenarios separately. If CV screening or performance scoring is on the table, design the impact assessment, human approval step, appeal route and record keeping first.
None of this works without connecting AI to your existing HR systems. Leave balances, length of service and shift data are what let the assistant give a correct answer; we explain where that data comes from in time and attendance to payroll integration and choosing HR software.
How we deliver AI in HR at Digital Bridge
We start HR AI work with a discovery session. Together with your HR team we map the repetitive tasks that eat the most time, the documents behind them and the systems where the data lives today. Each scenario is classified against the table above, and anything high-risk stays out of the first scope.
The first pilot is usually an enterprise LLM assistant: the staff handbook, leave and benefits documents become a knowledge base and every answer cites its source. The assistant only answers from documents the person asking is authorised to see, so salary tables and other restricted HR files stay closed to others. Where data must not leave the organisation, we can run the assistant on open-source models on your own servers.
The second step connects the assistant to HR data. The HR and payroll software we build to your own rules, or whichever system you already use, is linked through system integration, so when someone asks how much leave they have left, the answer comes from the live balance. For tasks such as reading personnel files or summarising HR reports, our AI integration approach adds the module to your existing software.
The legal framework is built alongside the project. Through our data protection compliance service we settle privacy notices, the data inventory and which data may go into AI tools. For AI use in other functions, browse all of our artificial intelligence articles.
Next step
For one week, count the questions and requests HR receives by category, and note the five most repeated topics together with the documents they rely on. That list is enough to scope a first pilot. Get in touch with it and we will sort out together which scenario can start straight away and which needs legal groundwork first.