AI in HR: where it genuinely helps, why hallucinations are unacceptable for leave and pay, what GDPR says about automated decisions, and when you need deterministic systems.
Translated from Romanian. This article covers Romanian labour and tax law; the Romanian version is canonical and is updated first.
The adoption of AI in everyday work has already moved past the experiment stage. In a study published by SD Worx in July 2026, 35% of employees in Romania use AI weekly — above the European average of 29% — and 7 in 10 report higher efficiency. In HR, the share of organisations investing in AI has risen to 50%, compared with 36% in 2025. The useful question is no longer “do we use AI?”, but where you let it decide and where you do not.
The discussion gets stuck because “AI” covers things with very different risk profiles:
Many products labelled “AI” are in fact combinations. The distinction matters because the legal risk differs radically between them.
Areas where an error is immediately visible and easy for a person to correct:
The common denominator: the output is a draft, not a decision, and a person validates it before it has any effect.
Now the part vendors gloss over. There are questions where an invented answer causes direct harm:
If a generative model answers “you have 7 days left” when there are in fact 3, the employee makes plans on a false figure and HR is left with the conflict. If it answers with the wrong sick leave percentage, the error ends up in the payroll.
The difference from the “safe” areas: here the answer is not a draft, it is a fact about the company’s data — and nobody checks it, precisely because the point of the automation was that it would no longer need checking.
For these questions, the right form is not a model that generates an answer, but a system that reads the real value from the database and returns it. That is not less “intelligent” — it is simply correct.
The legal aspect that is most often ignored. GDPR (Article 22) gives a person the right not to be subject to a decision based solely on automated processing which produces legal effects or significantly affects them. In HR, that touches directly on:
If you use AI in these areas, you need genuine human intervention (not a rubber-stamp approval), transparency towards the person concerned, and the ability to explain the logic. On top of that comes the minimisation principle: do not give an external system more data than it needs.
When an HR product promises an “AI assistant”, ask for an answer on three things:
| Type of task | Suitable approach |
|---|---|
| Drafting, summarising, translating | Generative AI, with human validation |
| Routing requests, classification | Classifier, with a confidence threshold |
| Balances, percentages, deadlines, figures | Deterministic, from the real data |
| Decisions affecting a person | A person, with AI at most as documented support |
Can AI calculate salaries? Payroll calculation has to be deterministic and verifiable — the same inputs, the same result, with every figure explainable. A generative model does not offer that guarantee. AI can help around the process (explanations, summaries), not with the calculation itself.
Is it legal to use AI in recruitment? Yes, with limits. GDPR restricts decisions based solely on automated processing that significantly affect the person. You need genuine human intervention, transparency and the ability to explain the logic.
Which HR data should not be sent to an external AI? Identification data (CNP), payroll data, health data (sick leave, fitness-for-work certificates) and any sensitive data. If a vendor sends them to a third-party API, you need a legal ground and a processing agreement.
What is the difference between a chatbot and a deterministic assistant? A generative chatbot composes an answer and can be plausibly wrong. A deterministic assistant recognises the intent, runs a query against the real data and returns an answer from a template — reproducible and verifiable.
Is AI worth it for a small company? Yes, if you apply it where it saves time without risk: drafting, summarising, routing requests. In an SME, though, the biggest benefit comes from deterministic automation — balances calculated on their own, alerts on deadlines, documents generated from data.
See also HR digitalisation in SMEs and GDPR in HR.
The internal assistant Modi in Workly was built on exactly the distinction above: it answers employees’ everyday questions — “how many days of leave do I have left?”, “what is my schedule today?”, “how many approvals are pending?” — with no external LLM anywhere in the answer flow.
The mechanism is deterministic: it recognises the intent in the text, runs a query against the employee’s real data (with that person’s access rights, not someone else’s) and composes the answer from a template in Romanian or English. The practical consequence: it cannot invent a balance, a percentage or a deadline — and the same input always gives the same answer, which means it can be checked and explained.
Just as important for GDPR: the data does not leave the platform for a third-party AI API. Personal identification numbers, salaries and medical documents stay encrypted in the system, with role-based access. And when Modi does not have a clear answer, it says so and points you to the source — instead of improvising.
Informational article, accurate at the date of publication. Not legal advice. The framework applicable to the use of AI in employment relationships is evolving (GDPR, the European rules on artificial intelligence) — check the current position or consult a specialist. The adoption figures quoted belong to the SD Worx study published in 2026.
How to spot exhaustion in your team using data you already have — overtime, untaken leave, absence patterns — and what a small business can do about it.
Compare the three payroll models: in-house team, SaaS software and full outsourcing. Criteria on cost, control, compliance risk and the right moment for each one.
A practical guide to internal reskilling in an SME: when it costs less than recruiting, what changes in the employment contract and what duties arise when the job duties change.
We'll show you how time tracking, payroll and the rest of the modules work for your company, in a short demo.
Request a demo