AI in HR means software that reads, drafts, matches, answers or predicts across the employee lifecycle: recruiting, onboarding, HR service, learning, performance and workforce planning. It is useful where work is repetitive and text-heavy (screening, scheduling, HR questions, drafting) and risky where it feeds decisions about people. In the EU, AI that selects candidates or evaluates, promotes, dismisses or monitors employees is classed as high-risk (Regulation (EU) 2024/1689, Annex III no. 4); the obligations for such systems apply from 2 December 2027. In German companies with a works council, it usually has a say before you start. This guide maps use cases, legal triggers and a sensible order to begin. It is general information, not legal advice.
Where AI is used in HR today
Vendor guides list anywhere from seven to twenty use cases. They fall into six areas, and the area decides how much risk and how much regulation comes with it.
- Recruiting: writing job ads, searching and matching candidates, screening CVs, scheduling, interviews by chat or voice. This is the most mature area. See our guide on the AI recruiter.
- Onboarding and HR service: assistants that answer standard questions on leave, payslips, policies and equipment, and prepare onboarding steps. Details in 12 HR helpdesk use cases.
- Learning and skills: skill maps, development suggestions, learning paths.
- Performance and feedback: drafting review texts, preparing 1:1s, summarising feedback.
- People analytics and planning: attrition signals, skill gaps, headcount scenarios.
- Agents: assistants that carry out multi-step HR tasks on their own.
Adoption figures in vendor surveys range widely, because "uses AI" can mean anything from a writing assistant to an automated screening. We therefore quote no adoption percentage here.
What AI does well in HR, and where it fails
It works for language tasks with a human check: first drafts of job ads, summaries of long feedback, answers from a policy library, shortlists that a recruiter reviews. It fails when it is treated as an oracle: scoring "potential", predicting who will quit, or ranking people on thin data. Historical data carries historical bias, and a score looks more objective than it is. The rule we use: AI prepares, a person decides, and the person must be able to see why.
AI in HR and the law: what applies to which use case
Three bodies of law matter in the EU, and in Germany a fourth: the AI Regulation, data protection (GDPR), equal treatment (AGG) and co-determination (BetrVG). The table links each use case to the trigger that usually applies. "Likely" means: check it for your concrete tool.
| Use case | EU AI Act | GDPR / AGG | Works council (BetrVG) |
|---|---|---|---|
| Job ad targeting, CV filtering, candidate evaluation | High-risk, Annex III no. 4 (a) | Art. 22 GDPR if solely automated; AGG §§ 1, 7 | Likely § 95 (2a), § 99, § 87 (1) no. 6 |
| Decisions on terms of work, promotion, dismissal, task allocation by personal traits | High-risk, Annex III no. 4 (b) | Art. 22 GDPR, DPIA (Art. 35) | Likely § 87 (1) no. 6, § 94, § 95 |
| Monitoring and evaluating performance or behaviour | High-risk, Annex III no. 4 (b) | DPIA, proportionality | § 87 (1) no. 6 (co-determination, enforceable) |
| Emotion recognition of employees | Prohibited at work (Art. 5), with narrow exceptions | – | – |
| HR chatbot answering policy questions | Transparency: people must know they talk to AI (Art. 50) | Personal data in questions: Art. 5, 6 | Often § 90 (information), § 87 if usage is logged per person |
| Drafting texts, summaries for HR staff, no personal ranking | Usually minimal risk | No personal data into public tools | § 90 (1) no. 3 (information on AI planning) |
Source for the classification: Annex III no. 4 covers (a) "recruitment or selection of natural persons", including targeted job ads, filtering applications and evaluating candidates, and (b) systems that "make decisions affecting terms of work-related relationships", promotion or termination, "allocate tasks based on individual behaviour or personal traits" or "monitor and evaluate the performance and behaviour". Most HR guides cover only (a). Point (b) is what reaches performance and people-analytics tools, and it is the part many HR teams have not looked at.
When do the high-risk duties apply?
The Digital Omnibus, Regulation (EU) 2026/1744, moved the application of the high-risk requirements for Annex III systems from 2 August 2026 to 2 December 2027. The ban on emotion recognition at work has applied since 2 February 2025, and the transparency duty in Art. 50 has applied since 2 August 2026. As an employer you are a "deployer": you must use the system as instructed, assign human oversight, keep logs and inform workers and their representatives before you put a high-risk system into use at work (Art. 26). Do not wait until 2027 for the preparation: the works council process takes months.
Does the works council have to agree?
Often yes. Four provisions matter:
- § 87 (1) no. 6 BetrVG: co-determination on technical systems "designed to monitor" behaviour or performance. It is enforceable: if the parties do not agree, the conciliation board decides (§ 87 (2)). Courts read "designed" broadly; a system that can log who did what is typically enough.
- § 90 (1) no. 3: you must inform the works council in good time about planned work procedures, including the use of AI.
- § 95 (2a): selection guidelines need consent, and this applies when AI is used to draw them up. § 94 does the same for appraisal principles.
- § 80 (3): the works council may call in an expert on AI at your cost where it needs one.
More on the recruiting side: Does the works council have to approve AI in recruiting?.
Can AI decide on its own?
Art. 22 GDPR gives people the right not to be subject to a decision based solely on automated processing that has legal or similarly significant effects. The Court of Justice held in the SCHUFA case (C-634/21, 7 December 2023) that a score can itself count as such a decision where a third party relies heavily on it. For HR this means: a ranking that the recruiter merely signs off may still be a decision. Build real review, not a rubber stamp.
Discrimination and the burden of proof
Under § 22 AGG, if an applicant proves indications of disadvantage because of a protected ground, the employer has to prove that no violation occurred. With an opaque score, that is hard. Choose tools that show the criteria behind a rating, keep records and spot-check outcomes by group. Using a vendor does not move the responsibility.
Where to start: a sensible order
- Low risk, high frequency: drafting, summaries, HR service answers from your own policies, interview scheduling. Few legal triggers, quick benefit.
- Recruiting support with human review: sourcing, CV screening against must-have criteria, structured interviews. High-risk class, so document criteria, oversight and logs from day one.
- Only after the basics are in place: performance, potential and attrition models. They fall under Annex III no. 4 (b), touch § 87 BetrVG and rest on the weakest data.
In our conversations with HR and recruiting teams, data protection is usually a checkpoint, not a blocker, as long as hosting, retention, deletion and consent are answered concretely. Bring those answers to the first meeting, and the works council conversation gets shorter.
What does it cost? A worked example
Two models are common. With credits, you pay for the AI work. At Sprad, screening one application with CV check (2 credits), candidate updates (9) and a voice interview (30) costs 41 credits; at the list price of 0.10 € per credit that is 4.10 €, and 3.69 € at the Atlas Base price of 0.09 € (source: sprad.io/pricing, status 29 October 2026). For 500 applications that is roughly 1,845 to 2,050 €. For people development, the Growth module is priced from 5 € per employee and month plus usage; for 200 employees that is 1,000 € per month as a platform fee. Compare this with the hours your HR team spends on the same tasks, and with your current tools. Prices change, so check the page before you plan.
FAQ
What is the difference between AI in HR and HR automation?
Automation follows fixed rules (send a reminder after three days). AI interprets language or data (read a CV, answer a free-text question, summarise feedback). Many products combine both.
Is AI in HR allowed in Germany?
Yes, subject to data protection, equal treatment and, in companies with a works council, co-determination. Some uses, such as emotion recognition of employees, are prohibited.
Do I need a data protection impact assessment?
For systems that systematically evaluate people, usually yes (Art. 35 GDPR). Clarify it with your data protection officer before the pilot.
Which AI HR tools exist?
Platforms for recruiting, HR service, learning and performance. Sprad covers recruiting as an AI recruiting platform with AI sourcing, CV screening and voice interviews and a talent network. Sprad is itself a vendor, so use our views as one input and run your own tests.
As of 8 November 2026. General information, not legal advice. Legal references checked against the official texts. Sprad is a vendor of recruiting and HR software; have the specific use reviewed by your data protection officer and employment counsel.
Jürgen Ulbrich
Jürgen Ulbrich has more than a decade of experience in developing and leading high-performing teams and companies. As an expert in employee referral programs as well as feedback and performance processes, Jürgen has helped over 100 organizations optimize their talent acquisition and development strategies.
