AI in HR: tools that work, pitfalls to avoid

AI is fundamentally changing HR

A single vacancy eats six weeks of lead time, a few thousand euros in ads, and an unhealthy number of emails along the lines of "Thursday 14:30 or Friday 10:00?". In a Dutch SME with one HR colleague serving 25 people, that hits hard. Recruitment, onboarding, leave, evaluations, contract changes: it stacks up until the strategic Friday-afternoon work ends up on the "someday" pile.

AI relieves that, if you know where to deploy it. Not by replacing the HR person (also not allowed, recruitment AI without human supervision is forbidden under the EU AI Act), but by removing the tedious in-between steps. Below: where AI delivers real value for Dutch HR teams in 2025, where it goes wrong the moment you get lazy, and how to start without giving the works council a stroke.

Where AI is already making a difference

Job postings that don't sound like every other LinkedIn post

Nine out of ten LinkedIn vacancies open with "Are you the driven professional who...". Not a vacancy, a template. AI fixes this if you brief it properly: feed your brand voice, two real customer stories and the actual profile into a Claude prompt and let it produce three variants. Textio does this more structured and flags male-coded language ("ambitious hunting team"), but for an SME with four vacancies a year, a good prompt library is enough. Time saving per vacancy: 60 to 80 percent, and response rates rise because your text finally stops sounding like everyone else.

Screening and selection (the dangerous one)

This is where the bias bomb sits. HireVue analyses video interviews, Pymetrics uses cognitive games, Greenhouse and Workday have CV scoring built in, Recruitee and Homerun bring up the rear in the Dutch landscape. Works, provided you accept that the tool assists and does not decide. The recruiter stays responsible for rejections and must be able to explain them. Anyone who gets lazy and rubber-stamps the shortlist will sooner or later get a GDPR complaint.

Intake and scheduling, the real gold

The biggest time savings sit here and almost no one talks about it. An intake bot that lets candidates ask about role, salary range and location 24/7. An agent that books interview slots itself based on two managers' calendars and a candidate. A WhatsApp flow that reschedules cancellations. Not high-risk, no works council drama, saves your HR colleague four hours a week. Build it with Cal.com plus a Claude agent or a custom GPT on your own documentation.

Onboarding

AI chatbots answer the first 100 questions a new employee has. From "where's the coffee machine?" to "how do I request leave?". Saves hours per week and gives newcomers a fast answer on weekends too. Notion AI on a decent handbook works fine, BambooHR's onboarding flows too.

Predicting turnover

Predictive analytics flag who is at risk of leaving. Not on gut feel, but on patterns: fewer Slack messages, lower response on feedback rounds, more sick days, lower survey engagement. Workday Skills Cloud, Visier, Gallup Q12. For an SME a simple spreadsheet with rules is often enough, and more honest, because you still understand what's happening inside it.

Leave administration and routine communication

Leave requests via WhatsApp that land in the system automatically. Welcome emails, first-day info, mentor introductions. Periodic management reports on absences and contracts. The silent time-wasters where AI tends to have the biggest impact, and which are not "high-risk" from an AI Act perspective at all.

The pitfalls

Bias in data

If you train AI on historical hiring data, you carry your own biases along for the ride. The famous example: Amazon developed an internal AI recruiter between 2014 and 2017, and in 2018 publicly disclosed that the project had been shut down because the system systematically disadvantaged women, simply because the training data contained mostly male historically-hired candidates. The AI learned "successful candidate = man" and discriminated accordingly.

This isn't solved. Every recruitment AI system carries this risk. The fix is a combination of transparent criteria you set upfront (rules you set yourself, not patterns the model invents), bias checks (monitoring dropout per group), and blind screening on name, photo and address in the first round.

Privacy and GDPR

Employee data is sensitive. Absences, evaluations, salary, conversation notes: these are GDPR categories you can't just dump into a cloud AI. Make sure your tools are GDPR-compliant, process EU-only data, and have standard data processing agreements in place. For the most sensitive work (psychological profiles, evaluation data), on-premise or an EU-private deployment is a serious option. Discussed per case, not the default.

AI Act high-risk classification

Recruitment AI falls under the "high-risk" category of the EU AI Act (Annex III). From 2 August 2026 that means mandatory transparency, human supervision, documentation, a fundamental rights impact assessment for public employers, and the right of objection for candidates. Not optional add-ons, legal requirements. Many existing AI recruitment tools do not yet comply. Ask every vendor explicitly for AI Act compliance documentation. If you get a vague email back, you know enough.

The human factor

AI screens and sorts, people decide. Nobody wants to be hired by an algorithm, and nobody wants to be rejected by an algorithm without explanation. Fixed rule: AI delivers the top-10 shortlist based on pre-defined criteria, the recruiter decides which 5 to call, the manager decides who gets hired. This is also the standard the AI Act prescribes.

Getting started with AI in HR: a practical step plan

Month 1: diagnosis and goals Which HR processes cost you the most time right now? Job postings? CV screening? Leave administration? Which deliver inconsistent quality (ask three new hires about their onboarding experience, you get three different stories)? Which are high-risk under the AI Act and which aren't? Set KPIs upfront: time per vacancy, time-to-hire, candidate satisfaction, shortlist diversity.

Months 2-3: pilot on a low-risk process Don't start with CV screening, that's high-risk and requires a works council process. Start with job postings, an onboarding bot or a scheduling agent. Tooling: Claude or ChatGPT with well-developed prompts and a vacancy template library. Workflow: AI drafts, HR reviews and adjusts. Measure KPIs weekly. If after six weeks it isn't going anywhere, stop and pick a different pilot.

Months 4-6: works council process for high-risk processes Only after the first pilots are running do you start with CV screening or turnover prediction. Here the works council must be involved in decision-making. Plan: information (1 session), demo (1 session), FAQ (1 round), consent request (formal documentation). Plan generously, a works council that feels ambushed will say no.

Month 7+: scale and integrate Roll out successful pilots to other departments or vacancies. Update your standard HR processes, not as standalone tools but as steps in workflows. Train new HR colleagues on AI tools as part of their onboarding.

How a typical engagement unfolds

I work in phases. First a short analysis of which HR processes deliver the most value and which fall under the works council process. Then a focused pilot on a low-risk process (vacancy content, onboarding checklist or scheduling bot). Only then do I tackle high-risk processes with the works council route. Between phases you evaluate whether to continue, so you don't get locked into a six-month commitment where nobody remembers what it was supposed to deliver. I work through a cost-benefit analysis upfront.

The bottom line

AI in HR works, provided you start with low-risk processes, build in GDPR and AI Act from day one, and communicate transparently with employees and candidates. Don't automate everything, discover step by step where AI adds value without losing the human touch. The win sits less in clever CV screening than everyone thinks, and much more in job postings that don't sound like all the others, a scheduling bot that doesn't email anyone about "Thursday or Friday", and an onboarding flow that also works on a Sunday evening.

Many firms start with job postings or an onboarding bot, and expand from there to leave administration and (after the works council process) CV support. Not a revolution, just more room for strategic HR work. That's where AI really makes a difference in 2025.

Want to know which HR processes deliver the most value? Take the free AI Scan and get a personalised analysis within 1 minute.