Domain Atlas / Public benefits & eligibility
Rotterdam welfare-fraud risk model
Explore this deployment in the PAN Lab ↗
In the PAN Lab, the readouts of this case's model organization carry a shaded evidence band whose width follows the least-established class among the modeling inputs the readings rest on.
The least-established modeling input behind the derived readings of this case's model organization is assumed: “This models the machine-learning welfare-fraud risk-tool pattern documented in the Rotterdam case file — not a reconstruction of the actual model or its inputs.” Evidence base: 5 assumed · 1 published baseline.
Independent scrutiny of Rotterdam's welfare-fraud risk model — a 2021 municipal audit followed by a 2023 journalistic investigation that obtained the model itself — documented scores skewed against already-vulnerable groups, and the city suspended the system's use.[5]
What happened
Rotterdam used a machine-learning model (originally built with Accenture) to score welfare recipients for fraud-investigation priority. The city's own audit office, the Rekenkamer Rotterdam, flagged the ethical risks in a 2021 report — proxy discrimination from variables like language skill — and the city paused the model. In 2023, journalists at Lighthouse Reports and partners obtained the model and its training data, and their analysis documented in detail how risk scores skewed against already-vulnerable groups — women, parents, people with limited Dutch.
The sociotechnical reading
Rotterdam is the Atlas's clearest case of the audit-as-actor: the control that finally bound the system came from investigative access to the model itself, which is exactly what vendor opacity usually prevents. It also illustrates selection harms that live outside output accuracy — the model decided who got investigated, and investigation is itself a burden. Governance that only measures "was the flag correct" misses the question "who bears the process". The equity-measurement patterns in the Practice Library exist for this distinction.
The concepts used in this reading are defined in the Field Guide; the governance responses live in the Practice Library. The model organization for this case can be stress-tested in the PAN Lab.