Dutch HR-AI startups each chart their own course on algorithmic fairness

Recruitment software that scores, sorts or excludes candidates falls under the high-risk AI category in the EU AI Act, which entered into force on 1 August 2024. That means strict requirements for transparency, data quality, human oversight and the right of applicants to an explanation. Dutch HR-AI startups are aware of this, but the way they fulfil that responsibility differs considerably.

The core tension is familiar: selection tools are built to operate quickly and at scale, while demonstrable fairness demands slow-down, external scrutiny and openness about how an algorithm reaches its conclusions. What choices are the Dutch players in this field making?

Textkernel bets on explainability and external audits

Textkernel, founded in 2001 and since acquired for what the company describes as around €300 million, has been working on CV parsing and candidate matching for more than two decades. The company applies a so-called white-box approach to its Search & Match product: the system indicates, for each recommendation, which criteria were decisive, allowing a recruiter to interpret and adjust the outcome.

That is a deliberate product choice. Textkernel splits the matching process into two steps, document understanding and matching itself, and explicitly acknowledges that AI can pick up and amplify patterns from historical recruitment data. To limit this, the company uses an AI Fairness Checklist as part of a broader risk management framework. Models are trained on broad datasets and regularly audited for bias.

On the compliance front, Textkernel has already aligned with the New York City Automated Employment Decision Tools law, which requires annual bias audits and has been in effect since 1 January 2023. The company also states that it complies with ISO 27001, SOC II, GDPR and the EU AI Act. Monica Schoonhoven, Global Director of Customer Success at Textkernel, expects recruitment processes to be widely classified as high-risk. The EU AI Act's transparency provisions take effect on 2 August 2026.

Harver submits to external review of demographic outcomes

Harver, founded in Amsterdam in 2015 and funded with €32.2 million in a Series B round, builds predictive assessments for high-volume recruitment. The company has had an independent audit carried out by BABL AI for New York City Local Law 144, a law that requires automated decision tools used in hiring to undergo annual bias audits. The audit assessed the disparate impact of the Harver platform on demographic groups by race and gender, and concluded that the model met the audit criteria.

Harver operates an AI governance programme comprising regular internal audits, ethical guidelines, validation and reliability studies, and human oversight. The company also publishes an evaluation framework with seven criteria that HR teams, legal departments and talent leaders can use to assess AI tools on dimensions including transparency and fairness. This approach is partly aimed at the company's own clients: Harver also positions fairness as a reason to choose the platform.

Equalture makes fairness its core proposition

Equalture, founded in Rotterdam in 2017 and funded with €6.8 million in a Series A round, takes a different position. The company builds game-based assessments that measure behavioural traits and cognitive skills, with the explicit goal of reducing unconscious bias in the selection process. Whereas many recruitment tools use historical employee performance data for training, and can thereby reproduce existing demographic patterns, Equalture focuses on potential and behaviour in standardised situations.

Equalture emphasises avoiding proxy variables that indirectly correlate with gender, ethnic background or educational level. Whether and how the company conducts external audits of its models is not clear from publicly available information.

Other players focus on different stages of the recruitment process

RecruitRobin (Utrecht, founded 2019, €10 million via acquisition), Wonderkind (Amsterdam, founded 2016, €9.1 million Series A) and MrWork (Rotterdam) do not focus on assessing or selecting applicants, but on distributing job vacancies. RecruitRobin distributes vacancies across twelve job sites from a single platform; Wonderkind converts vacancies into social-media advertisements targeting young talent; MrWork automates candidate sourcing via AI.

Such distribution-oriented tools are also subject to provisions of the EU AI Act, particularly regarding targeted job advertisements that may exclude certain groups or target them disproportionately. How these companies are addressing this is poorly documented in public sources.

TestGorilla (Amsterdam, founded 2019, €81.2 million Series A in 2022) screens candidates through validated skills tests and positions this as an alternative to CV-based selection. The company emphasises that skills-based testing is less susceptible to bias than CV review, but does not, as far as is known, publish independent bias audits of its test batteries.

What the comparison reveals

The companies that explicitly focus on selection and assessment, Textkernel, Harver and Equalture, have the most visible approach when it comes to bias and transparency. Textkernel and Harver have already been assessed against New York legislation, which is structurally comparable to what the EU AI Act requires of high-risk systems. That gives them a head start in the documentation that European regulators will soon demand.

Equalture differentiates itself by linking the design of its assessments to fairness objectives, but the extent of external scrutiny is less transparent. For the other startups in this overview, their public communication on bias risks and EU AI Act compliance is limited, which does not mean they are not working on these issues internally, but does mean they are not yet deploying them as a competitive differentiator.

For the Dutch and European AI scene, this is a relevant signal. The EU AI Act turns compliance into a demonstrable obligation, not merely an internal promise. Startups that invest now in external audits, explainable models and documented risk frameworks are building something that will soon be a prerequisite for market access. Investors and public-sector clients will increasingly ask for it.