CompetentNL skills detection

Central Bureau of Statistics

CompetentNL is the Dutch standard for describing skills. CBS helps to keep CompetentNL up-to-date, independent and reliable. This is done by searching large-scale labour market data for skill descriptions; we call this skills detection. The results are shared with the organisation managing CompetentNL.

Last change on 26th of March 2026, at 15:28 (CET) | Publication Standard 1.0

• Publication category: Other algorithms
• Impact assessment: DPIA
• Status: In use

General information

Theme

• Work
• Education and Science

Begin date

09-2025

Contact information

[e-mail]

Link to publication website

https://competentnl.nl

Responsible use

Goal and impact

The algorithms are used within the collaboration CompetentNL, commissioned by the Ministry of Social Affairs and Employment (SZW) and the Ministry of Education, Culture and Science (OCW).

The algorithms were created to help CompetentNL. CompetentNL is a national standard for describing skills. The algorithms help process a lot of current data from the labour market. They search for descriptions of skills (identification) and organise them (classification).

Citizens and businesses use CompetentNL through digital services and platforms, such as the Learning Overview and the Skills Dashboard. These provide information on skills and career opportunities. The algorithms work in the background and do not interact directly with citizens and companies.

Considerations

Benefits: The algorithms help make CompetentNL better. They ensure that we use current and relevant skills from the labour market. This improves the quality of the data. It also ensures that everyone can identify with CompetentNL because we process the language of the labour market properly.

Disadvantages and risks: If the training data is outdated, it can affect the quality of the results. The algorithms work in the background and do not get direct feedback from users. This can make it difficult to find errors.

Justification: Using these algorithms is good because they improve the quality and relevance of CompetentNL. Human experts remain involved to check the results so we can mitigate risks.

Human intervention

The results of the algorithms are assessed by experts before entering CompetentNL. These experts check that the skills descriptions and classifications are relevant, accurate and of good quality. They can adjust or reject the results based on their knowledge.

This monitoring is important to find and improve errors and outdated information. The algorithms are also regularly evaluated and adjusted based on expert feedback and new developments in the labour market. This keeps the information up-to-date and reliable.

Risk management

Obsolescence of training data: We plan regular updates to the training data. This helps keep the models current. Experts constantly look at the results to see when we need to adjust something.

Bias and inaccuracies: The algorithm results are always checked by human experts before they are used. We improve language models and methods to reduce bias.

Dependence on the algorithms: The system is made so that results of the algorithms are not automatically adopted. Human control is always needed. Training and awareness of experts help to use the algorithms critically.

Limitations in interpretation and use: Results are not shown directly to end users without verification. Errors and imperfections can thus be found and corrected in time.

Legal basis

The algorithms are used according to the rules of the Central Statistics Act. They also follow the General Data Protection Regulation and Dutch rules on privacy. The process helps achieve the goals of the Ministry of Social Affairs and Employment and the Ministry of Education, Culture and Science. These ministries give the assignments. The use of algorithms makes the labour market and education better and fairer. The European Union has rules for AI, and they are followed. Data is processed with respect for people's rights. Experts make sure everything is fair and clear.

Links to legal bases

• Central Statistical Office Act: https://wetten.overheid.nl/jci1.3:c:BWBR0015926&z=2025-09-01&g=2025-09-01
• General Data Protection Regulation (AVG) (EU) 2016/679: https://eur-lex.europa.eu/legal-content/NL/TXT/?uri=CELEX%3A32016R0679
• AI regulation (Regulation (EU) 2021/0106): https://eur-lex.europa.eu/legal-content/NL/ALL/?uri=CELEX:52021PC0206

Elaboration on impact assessments

An Impact Assessment for human rights and algorithms has not been done. This is because the algorithms we use are only supportive and work in the background. They do not interact directly with users or make decisions about people. Therefore, the risk of human rights violations or discrimination is small.

The processing of data with these algorithms falls under CBS's general DPIA. In addition, another in-house DPIA was carried out in addition to the general DPIA. The in-house DPIA focused on what is not standard, namely AI processing and returns to UWV. This means privacy risks have been properly examined and mitigated. We pay attention to continuous monitoring and human control to spot and address potential risks in time.

Impact assessment

Data Protection Impact Assessment (DPIA): https://www.cbs.nl/-/media/cbs/over-ons/organisatie/standaardcbspia-2021-v20.pdf

Operations

Data

The algorithms use data from UWV. This is labour market data, such as job ads and profiles of job seekers. Experts from UWV provide annotations and labels to train the models. External language models, such as LaBSE, are also used for processing. The data helps to recognise and improve skills and competences within CompetentNL. All data processing follows privacy and security standards, and the data is anonymised where possible.

Links to data sources

LaBSE language model: https://huggingface.co/sentence-transformers/LaBSE

Technical design

The algorithms use smart language models to analyse labour market data. There are two algorithms: Identification and Classification. In Identification, the algorithm recognises important skills in labour market data, such as vacancies and profiles. It does this using an external language model and a self-created model to find relevant phrases. Experts check the results. In Classification, the algorithm compares the sentences found with existing skills in CompetentNL. The best matches are chosen and submitted to experts.