evidoria

← Back to browse

Good practice Imported

DUO's Discriminatory Fraud-Detection Algorithm — A National Cautionary Case in Automated Student-Grant Checks

Netherlands · Groningen · See the Netherlands profile

Evidence: Descriptive / self-reported Top 100% 7/100 · Ask Evidence Copilot about this practice

From 2012 to 2023, the Netherlands' DUO used an opaque risk-scoring algorithm to select students for financial-aid fraud checks. The Dutch DPA ruled it unlawfully discriminatory in 2024; compensation for thousands of affected students was announced in 2026.

21,500
Students selected for fraud checks partly on the algorithm's output (2013–2022)
>10,000
Students the government announced compensation for (2024–2026)
DUO's Discriminatory Fraud-Detection Algorithm — A National Cautionary Case in Automated Student-Grant Checks

Details

Maturity
Discontinued
Promoter
Dienst Uitvoering Onderwijs (DUO), Dutch Ministry of Education
Period
2012-2023
Keywords
financial aid, fraud detection, algorithmic governance, data protection

Context

From 2012 to 2023, the Dutch student-finance agency DUO used an internal, non-validated risk-scoring algorithm — based on factors such as the distance between a student's registered address and their parents' address, age and type of education — to select students living away from home for financial-aid fraud investigations. The model drew on the accumulated experience of DUO fraud investigators rather than a validated statistical method. Between 2013 and 2022, DUO selected 21,500 students for fraud checks partly on the algorithm's output.

Results

In July 2023, the education minister suspended the algorithm after investigative reporting exposed the practice, replacing it with random sampling. In 2024 the Dutch Data Protection Authority ruled the risk model unlawfully and indirectly discriminated against students with a migration background, a finding corroborated by Amnesty International's 2024 report 'Profiled without protection.' The Dutch government announced compensation for more than 10,000 affected students in late 2024 and again in 2026.

Implementation
Indicative cost
Medium (€50k–€500k)
Time to results
Long (> 3 years)
Staffing & skills
DUO fraud investigators / case workers

Conditions for success

  • independent bias/discrimination audit before deployment
  • a validated statistical model rather than investigator intuition alone
  • transparency about the risk-scoring criteria used

Common failure modes

  • opaque, non-audited scoring criteria used for over a decade
  • reliance on investigator experience rather than a validated model
  • indirect discrimination against students with a migration background

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful