Datamining score finds more payment errors in French family benefits

The Cour des comptes found that desk checks picked by the family branch's data-mining score found overpayments in 26% of files in 2019, up from 6% in 2012.

  • 2 an organization

    Desk checks picked by the score found overpayments in 26% of files in 2019 against 6% in 2012, and data mining detected €218.5 million of overpayments that year.

  • Not rated a person

    The checks found €87.2 million in arrears owed to beneficiaries in 2019, but the Cour gives neither the number of households nor what each received.

Depth runs from 1, a convenience or time saved, to 5, a life or a livelihood preserved.

Status
Reported benefit.
Evidence

foundstudiedreportedclaimed

Two independent audit reports support the benefit.

Compared with

Desk checks picked by the score in 2012, its first year of use. Measured.

Without the system
6 % et 5 % respectivement en 2012
With it
26 % de dossiers comportant des indus et 22 % des rappels détectés en 2019
The system’s part
Not isolated. The Cour compares hit rates in 2012 and 2019, and the score chose the checks in both years. The Cour compares no checks chosen without the score. It writes that the score depends on each file's similarity to a model built from the yearly survey of payment accuracy.
Limits
Not yet checked.
Costs and harms
Not yet checked.
Later
Not yet checked.

Download data Cite Suggest a correction

Benefit type
Improved Task Accuracy
System
Data-mining risk score of the family branch of French social security (le datamining)
Made by
Caisse nationale des allocations familiales (Cnaf)
Used by
Caisses d'allocations familiales (Caf), nationwide
Where
France (national)
Stage
Expanding
Date
Kind
Aggregate result
Broad benefit
Analyzing Data or Information, Organizing, Planning, and Prioritizing Work
Collections
Scrutiny and findings
Last revised
31 October 2026

Sources

  1. Audit report, by authors with no tie to the system.

    Cour des comptes. La lutte contre les fraudes aux prestations sociales. Communication à la commission des affaires sociales de l’Assemblée nationale, September 2020, pp. 48-49.

    Les contrôles effectués par les Caf sur des situations d’allocataires sélectionnées par le datamining (soit 233 733 contrôles sur pièces et 111 774 contrôles sur place en 2019) conduisent à détecter des fréquences de plus en plus élevées d’indus et de rappels

  2. Audit report, by authors with no tie to the system.

    Cour des comptes (France). Sécurité sociale 2025: Rapport sur l'application des lois de financement de la sécurité sociale (RALFSS 2025), chapter on benefits paid in error (indus). Deuxième partie, chapter on indus in the benefit branches, passage on the family branch's controls (around pp. 300-312)

    Par ailleurs, les contrôles liés à l'analyse des données de masse ont détecté 67 M€ d'indus (49 M€ en 2019).

Details

AI system

Workflow

The score ranked each beneficiary's file by its similarity to a model built from the yearly survey of payment accuracy. Each Caf checked the files that the score selected, on paper or on site.

Money

Kind of value

Overpayments detected in 2019 through checks that data mining selected

Amount

218,5 M€

Recipients and payers

The family benefits branch

Follow-up

Conditions for replication

The annual survey must keep supplying data to rebuild the model. The variables must respect the rules on personal data, a point that the Cour raised for nationality and place of residence.

Entry details

Sector

Social benefits (family and housing).

Purpose

The Cnaf built the data-mining score to pick the beneficiary files most likely to contain overpayments or underpayments for checks.

Not yet checked: model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, currency and prices, period and unit, realization, gross or net, included costs, comparison, method, return and payback, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.

Cite this record

Peter H. Vartanian. (2026). Datamining score finds more payment errors in French family benefits (AIBD-0686). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0686.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)

BibTeX Data Field definitions