Responsible AI Platform

AFAS HR under the EU AI Act: how a Dutch software vendor handles Annex III

ยทยท8 min read

AFAS Software is a Dutch software vendor with a strong market position in NL SMEs and (semi-)public sector. Where Workday and SAP loudly promote their AI roadmap, AFAS chooses a more restrained path: AI as support where it adds value, no autonomous decision-making, and a public vendor statement that puts the human decision-maker first. For Dutch compliance leads that is a refreshing line, but it does not absolve the employer of an independent check.

This analysis describes the current AFAS HR AI presence (Profit HR, Insite, OutSite), places it against Annex III point 4, and ends with vendor questions that are concrete for Dutch AFAS users.

What AFAS publicly offers around AI in HR

AFAS publishes via blog, product pages and annual SoftwareUpdate sessions:

  • Profit HR & Payroll โ€” core system for administration, payroll, absence
  • AFAS Insite โ€” employee portal with workflow and self-service
  • AFAS Pocket โ€” mobile employee app
  • OutSite โ€” application portal (candidates apply, recruiters process in Profit)
  • AI assistance โ€” recent rollout of AI features for recruiter productivity (generating job ad text, candidate summaries), largely generative
  • Document AI / OCR โ€” automatic processing of invoices, certificates, ID documents
  • No integrated candidate matching or scoring โ€” AFAS does not publicly offer this as a core feature

AFAS's position is notable: the vendor emphasizes that customers "retain control" and that AI features do not autonomously assess candidates or workers. This is vendor positioning, not a waiver of AI Act classification.

The seven checks applied to AFAS

1. Does the AI rank or score candidates?

Not in the standard AFAS modules. Profit HR does no candidate ranking, and OutSite is a receiving portal. For AFAS in baseline configuration: likely outside 4(a).

2. Does the AI optimize who sees a vacancy?

AFAS does no sourcing or distribution via AI itself. If you work via integrations with external job boards (Indeed, LinkedIn), that targeting falls under the classification of those external platforms.

3. Is CV parsing really only parsing?

Document AI extracts fields from submitted CVs for administration. No skills inference to matching. This remains parsing as a rule.

4. Is the chatbot logistical or selective?

AFAS Insite Chat (where configured) supports employee requests, leave issues, expense claims. Logistical. For candidates: AFAS does not offer a standard screening chatbot.

5. Does the assessment tool measure behavior or performance?

AFAS does not offer psychometric assessments. Performance modules in Profit are classical workflows without AI scoring. Customers who deploy external assessments via integrations assess those separately.

6. Does the system continue post-hire?

Profit handles HR matters, payroll, absence. Classical administration here, no AI assessment of workers. For users of AFAS Beoordelen and Functioneren: workflow tools, no AI scoring.

7. Can you substantiate the vendor claim?

AFAS does not publish enterprise-style Model Cards or AI Fact Sheets, but does have an AI positioning via blog and Customer Service. For compliance that means: request written confirmation that no candidate or worker decisions are supported by scoring or ranking models in the modules you use.

The classification call

AFAS in baseline configuration likely sits outside Annex III point 4 high-risk. That is a significantly lower risk profile compared to Workday, SAP or HireVue. But:

  • AI assistance features that AFAS rolls out for recruiter productivity must be checked per release
  • Integrations with external ATS or assessment platforms fall under the classification of those platforms
  • Generative AI for job ad text is GPAI use (not 4(a)) but falls under Article 50 transparency requirements if you publish generated content
  • For (semi-)public sector employers FRIA obligation remains once you deploy a high-risk AI system, even if it does not come via AFAS

Vendor due diligence for AFAS

1

Document 'outside Annex III point 4' in your AI register

'Outside high-risk' is also a classification decision you document. An AI register with "AFAS HR โ€” no high-risk AI usage established" with reference to vendor confirmation is defensible.

2

Keep integrations separate for classification

External ATS or assessment vendors that integrate via AFAS have their own classification. AFAS outside Annex III point 4 does not mean an integrated HireVue is too.

3

Schedule a vendor check at every AFAS major release

AFAS rolls out new features annually via SoftwareUpdate. Schedule a fixed compliance check on AI features at every major release to do reclassification on time.

Practical tool

Test your AI against the Article 6(3) filter

Interactive self-assessment, updated for the Commission guidelines of 19 May 2026. 9 steps, personal report with reasoning, vendor questions and next steps.

Start the Annex III Classifier 20265-8 minutes ยท Article 6(3) filter built in

Frequently asked questions about AFAS and the AI Act

Practical questions for Dutch AFAS users under the EU AI Act.

What to do now

For AFAS users in the Netherlands the sensible approach: request written vendor confirmation, document a classification decision ("outside Annex III point 4 in current configuration"), and build a fixed compliance check into every AFAS release. That is proportional work for a proportional risk profile, and it does not absolve you of Article 4 AI literacy for your HR team.

For the rest of your HR stack (external ATS, assessment, sourcing) the heavier routes apply: see the HR AI hub and the HR AI Evidence Pack.

โš–๏ธ Referenced Legislation