AI workshop for HR departments: from scattered pilots to a single governance roadmap

A working day for your HR director, recruitment lead and data protection officer. Not a tool demo and not a prompt course, but a structured session in which you decide which AI is allowed in recruitment, assessment and employee experience, which falls under high-risk under the AI Act and which you are better off keeping out of the organisation. You go home with an AI roadmap, a DPIA template and a governance protocol — not a non-committal report.

AI Act art. 6 + Annex III DPIA for HR AI GDPR art. 22 Bias auditing Collective agreement & works council process
Plan the working day View programme
HIGH-RISK LIMITED

When an HR working day makes clear where AI may, must and cannot be used within your organisation

Most HR teams don't struggle with too little AI, but with too many separate initiatives that nobody has a full view of. A recruiter tests ChatGPT for job postings, talent acquisition runs a pilot with an interview AI, an external vendor offers an attrition model, and the DPO only gets involved once the contract is on the table. This workshop brings all those pieces together in a single day, at executive level, so you decide on the basis of AI Act classification and organisational interest rather than the enthusiasm of individual vendors.

The AI Act has been in force since 1 August 2024, and the provisions for high-risk AI systems will become enforceable in phases from August 2026. Recruitment, selection, assessment, promotion and the termination of employment relationships are explicitly listed as high-risk areas of use in Annex III of the Regulation. That means a large share of the HR AI on the market today (CV screening, interview scoring, predictive attrition, bias detection, candidate matching) falls under a conformity assessment as soon as you put it into production.

This working day is deliberately designed for the strategic layer: HR leadership, the CHRO, the talent acquisition lead, a representative of the works council's working group on terms of employment, and your data protection officer. It is not for HR staff who want to learn ChatGPT; for that we have a separate AI tools workshop for employees and an AI literacy training programme. On this day, you build the framework within which the entire HR chain will be allowed to work from tomorrow.

Which HR AI we classify on the day, and why that is the focus

We go through the seven use cases below one by one. For each case, you determine: does it fall under Annex III, does GDPR Art. 22 apply, is there an automated decision with legal effects, and, if you proceed, which human intervention, bias audit and explainability are required as a minimum.

📑

CV screening and candidate matching

AI-driven sorting or ranking of applicants falls under Annex III point 4(a). It is almost always high-risk, even if a recruiter formally takes the final decision. We discuss which matching criteria are permissible, when a four-fifths-rule bias audit is needed, and how to build fair-hiring principles into your procurement terms.

🎤

Interview AI and conversation analysis

Tools that score speech, facial expressions or language use during a job interview. Here you touch on Annex III, GDPR Art. 22 and, for biometric data, the special categories of Art. 9 GDPR all at once. We identify which vendors we see on the Dutch market, what substantiation they do and do not provide, and when your works council formally has a right of consent.

📞

Onboarding bots and HR helpdesk AI

A chatbot that guides new employees through their first weeks, or answers questions about leave and the collective labour agreement. Often limited-risk, provided no decision-making is attached to it. We put on paper where the line lies between informing and advising, how to prevent hallucinations on collective-agreement questions, and what logging you need for traceability.

📊

Predictive attrition and flight risk

Models that predict which employees will leave within six months. Annex III point 4(b): decisions on employment relationships, promotion and termination. The fact that the manager holds the conversation changes little if the model determines who gets one. We discuss which works council discussions and collective-agreement consultation steps apply.

💬

Sentiment analysis on employee surveys

Having an LLM summarise pulse surveys, exit interviews and open-text fields. The risk lies not in the technology but in pseudonymisation: in small teams, responses can be traced back to individuals, which makes them relevant under the GDPR. We draw up a protocol for minimum team size, opt-in and retention period, and for how to prevent a vendor from using your open answers for model training.

🎓

Skills mapping and NSE integration

Matching internal employees to vacancies, learning paths or succession plans based on skills extraction. Alongside the Dutch Higher Education Act and the NSE (national skills equivalence), this is a growing field. We look at whether a skills graph really belongs in-house, or whether you are better off building a vendor-independent layer that isn't tied to a single HR suite.

The classification table you fill in on the day

Not abstract legal theory, but a concrete matrix we work through per HR system. You leave with this table completed for your own organisation.

HR application AI Act GDPR art. 22 Bias audit required Works council consent
CV ranking and automated shortlisting High-risk Possibly Yes, at every model training Yes, art. 27 Works Councils Act (WOR)
Job ad text generation (LLM, no scoring) Minimal No No, but a house-style and inclusivity check No, provided no personnel data is involved
Interview AI with voice or video analysis High-risk Yes, almost always Yes, demographic subgroups Yes, consent required
Predictive attrition (flight risk) High-risk Yes, if used for promotion or retention Yes, annually Yes, art. 27 Works Councils Act (WOR)
Onboarding chatbot, informational Limited No, provided it makes no decisions No, but factual accuracy against the collective labour agreement (CAO) Sometimes, depending on scope
Sentiment analysis of open survey text Limited No, provided it is anonymised No, but framing validation Sometimes, for small teams
Engagement sentiment per individual High-risk Yes Yes Yes, consent required

How the day is structured

A day from 9:00 to 16:30 at your location or at Appfront. Four blocks, each with a concrete outcome. No lengthy theory sections: you receive the regulations in advance as reading material, so the day itself is about your organisation.

Four blocks, four deliverables

Each block ends with a concrete output you take with you. By the end of the day, your HR AI roadmap is no longer an intention but a document.

Inventory

We map all HR AI that is running now or sitting on the shelf: paid vendors, free tools employees use, and the vendor roadmaps of your HRIS. Output: shadow AI overview.

Classify

The seven use cases run through the AI Act matrix, with DPIA questions, GDPR art. 22 assessment and works council consent analysis. Output: a completed classification table for your situation.

Prioritise

Which HR AI delivers real value, which fits your risk appetite, and which is better outsourced to a specialist partner? Output: a phased HR AI roadmap with a go/no-go per case.

Document

We write the governance documents: an HR AI DPIA template, a vendor checklist with AI Act clauses, an incident escalation protocol, and a communication line to the works council and employees. Output: a governance pack ready for approval.

Who this day is for, and who it is not

A deliberately small group at a deliberately strategic level. We work at director level so the outcomes carry authority.

Yes: HR leadership and related roles

HR director or CHRO, head of talent acquisition, head of learning & development, a works council representative from the employment conditions working group, a data protection officer and, if you have one, the chief people officer. A maximum of eight participants, one table, no plenary format. Ideally also the finance controller who signs off the HR budget, as many decisions touch procurement.

No: HR staff and recruiters

Recruiters who want to learn ChatGPT for job ads, HR business partners who want a prompt library, or employees getting started with Copilot: for them we run the AI tools workshop for employees and the AI literacy training programme. Separate sessions, a different approach, a different pace. Mixing doesn't work, because the strategic work then doesn't get finished.

Yes: organisations that take the AI Act seriously

Employers with more than 200 staff, organisations with a works council and collective bargaining, and employers in healthcare, financial services, government or education where agency staffing or conformity assessments carry extra weight. Also suitable for staffing and recruitment agencies, which become high-risk providers themselves as soon as they offer AI screening to clients.

Not for: organisations that are only experimenting

If your organisation has not yet received a question from the works council, IT security or legal, and you simply want to explore whether AI could add something useful, a workshop is too heavy. In that case, start with a conversation about a lighter AI discovery workshop or a no-obligation introductory call. We are upfront about when an engagement does not fit — it saves everyone time.

Why Appfront, and not your HRIS vendor or a general consultancy

Workday, SAP SuccessFactors and Oracle each have their own AI modules and their own narrative about compliance. Understandable — they want to keep you on their stack. A large consultancy usually delivers a report, not a working solution. We are in a different conversation: we build custom AI ourselves, so we know the difference between what a vendor promises in the slide deck and what the system actually does in production.

Our role on the day is twofold: facilitator and second technical reader. If the vendor says their model is "not high-risk", we can hold the architecture up against Annex III and say specifically where it does fall under it. If the DPO says data minimisation is too restrictive, we sketch what a privacy-by-design architecture looks like. The goal is for you to stay in control yourself — not to become dependent on Appfront or any other party.

After the day, do you want to build your own compliant HR AI that is not locked in to Workday AI or another closed ecosystem? Then we can continue that work as your implementation partner. See our pages on AI development in the Netherlands and interim AI tech lead for what that means in practice. No forced follow-on — only if the match and the business case are there.

Practical details

Location and duration

A full working day from 9:00 to 16:30, at your office or ours. Lunch and coffee are part of the set-up — the work happens on the day itself, not in plenary presentations. Maximum group size: eight participants.

Preparation

Two weeks beforehand: a short intake call, reading material on AI Act Art. 6 and Annex III, a questionnaire for participants about your current HR AI, and — if available — a vendor overview from your procurement or IT organisation.

What you take away

A completed classification table, a draft HR AI roadmap, a DPIA template for HR AI, a vendor checklist with AI Act clauses, a works council communication draft, and a follow-up call after four weeks to review progress.

Frequently asked questions about the AI workshop for HR

Our HRIS vendor says their AI modules do not fall under the AI Act — is that correct?
A vendor can say that, but the classification follows from how the tool is used in your organisation, not from the product description. Annex III ties high-risk to the area of use (recruitment, evaluation, promotion, termination of employment). If you use an AI module for any of those, it falls under high-risk — regardless of what the vendor claims in its marketing. On the day, we test this against the tool's actual data flows and outputs, not against the vendor's disclaimer.
Does the AI Act already apply, or only later?
The Regulation entered into force on 1 August 2024. Prohibited practices (Art. 5) apply from February 2025, transparency obligations from August 2025, and the high-risk obligations for Annex III systems from August 2026. The HR AI decisions you take now on procurement and rollout must already account for this: a six-year vendor contract without an AI Act clause is a visible risk in 2026 or 2027.
Does the works council have a formal right of consent for HR AI?
Under Art. 27(1)(l) of the Dutch Works Councils Act (WOR), the works council has a right of consent on schemes for staff appraisal, staff monitoring systems and privacy protection. An AI system for CV screening, interview scoring or attrition prediction falls under this in almost all cases. Delaying or introducing it without the council's knowledge can be reversed afterwards, so it is better to involve the works council early. This is a standard part of the governance output of the day.
What is the difference between this workshop and AI literacy training?
AI literacy (mandatory under AI Act Art. 4 from February 2025) is about basic knowledge for all employees who work with AI. This workshop is about governance and classification decisions for the entire HR chain. In the end you need both: employees with literacy, and leadership with a roadmap. We can deliver both: see AI literacy training for the broader programme.
Will you also do the DPIA for us, or only provide the template?
On the workshop day we deliver the template plus a completed example DPIA for one of your use cases. The full DPIA for each HR AI system is a follow-up step. Your own DPO can carry it out using the template, or we can take it on as a separate assignment. The choice is yours; we do not push the DPIA as an upsell.
Do you work with fixed vendor choices, or are you vendor-independent?
Vendor-independent. We build ourselves, so we have no partner deals with HRIS suppliers or HR AI tools that steer our advice. On the workshop day we discuss the pros and cons of Workday AI, SuccessFactors, Eightfold, HiPeople, Pymetrics, Beamery and open-source alternatives on factual grounds, not out of any commercial interest on our side.
What if we want to build in-house after the workshop instead of choosing a vendor?
That is a legitimate outcome, especially for organisations that see strategic value in owning their HR data or that want to avoid vendor lock-in. We build this kind of custom software: candidate matching with explainable models, your own LLM infrastructure for onboarding bots, or a privacy-by-design skills graph. Our technical approach is described on the page about AI development. If an implementation request follows after the workshop, we quote for it as a separate project.
Do you also work in the public sector and healthcare, where the rules are even stricter?
Yes. For municipalities and government organisations we also apply the Dutch Algorithm Act, the algorithm register and the broader Code Goed Digitaal Openbaar Bestuur (Code for Good Digital Public Governance). In healthcare, the GDPR special-categories discussion applies on top, plus sector rules from works council advisers in the Hospitals collective agreement (cao Ziekenhuizen). See our pages AI for municipalities and government and AI in the healthcare sector for sector-specific work.

Want to schedule a workshop day for your HR leadership?

Send us a message. We will schedule a short intake beforehand to agree the scope, the right participants and the preparation materials, and then set a date.

Schedule an introduction

Edit content