Nearly a third of DGFiP tax controls rest on mass-data analysis

AI system

System: DGFiP data-mining targeting of tax controls (program name not given in the sources)
Model: Not assessed
Release: Not assessed
Application developer: Direction générale des Finances publiques (DGFiP)
Role: Not assessed
Configuration: Not assessed
Inputs and outputs: Not assessed
Autonomy: Not assessed
Workflow: Mass cross-matching of declarative and third-party data to program a growing share of controls.
Human oversight: Not assessed
Relation to AI: The Cour describes risk analyses fed by mass data processing. It reserves the term artificial intelligence for the separate foncier innovant scheme. Whether the targeting uses machine learning is thus not established. DGFiP builds and runs the targeting itself. The Cour des comptes, an independent audit court, wrote the report.
AI contribution: Not assessed

Outcome identifiers: Not assessed
Benefit: nearly a third of DGFiP tax controls rest on mass-data risk analyzes (target 50% by 2027)
Category: Improved Organizational Efficiency
Status: Not assessed
Beneficiary groups: Organizations. Society.
Beneficiaries: Not assessed
Reach: Not assessed
Distribution and access: Not assessed
Duration: Not assessed
Expected?: Yes
Costs and harms: Not assessed

Studies: Not assessed
Study design: Not assessed
Measurement: Not assessed
Setting and sample: Not assessed
Measures: nearly a third of DGFiP tax controls rest on mass-data risk analyzes (target 50% by 2027)
Comparison: tax controls initiated by DGFiP agents' own analysis
Results: Reference: le rendement moyen à l’initiative des agents était près de cinq fois supérieur. Result: celui des contrôles résultant du data mining.
Additional value from AI: Not assessed
Limitations: Not assessed
Denominators: Not assessed
Materials for replication: Not assessed

Participants: Developer: Direction générale des Finances publiques (DGFiP). Operator: Direction générale des Finances publiques (DGFiP).
Jurisdictions: Developer: France. Operator: France.
Ownership and control: Not assessed
Dependencies: Not assessed
Cross-border connections: Not assessed
Geopolitical implications: Not assessed

Conditions for replication: The legal limit that cross-matching assists but never automatically characterizes fraud. Audit staff capacity to convert targeting into droits et pénalités, which the 2026 debate questioned.
Replication attempts: 1 follow-up entries. Benefit recurred: 1.
Later findings: 22 July 2026. Citing the Cour des comptes, deputy Mathilde Feld told the finance committee on 21 July 2026 that data-mining-derived controls were 44% of DGFiP controls in 2023, up from nearly a third, while yielding under 14% of the sums recovered.
Availability: Not assessed
Institutional decision: Not assessed
Grounds: Not assessed
Action and result: Not assessed
Success criterion: Not assessed

Assessed benefits: Organizations. Society.
Depth: Not assessed
Reasons: Organizations: Sources give the share of controls programmed by data mining and a contested yield share, not an outcome attributable to the system. Society: No recovered amount is attributed to data mining by the Cour. The 2026 figure is a yield share stated in debate.
Observed or intended: Not assessed
Rubric: Not assessed
Assessor and date: Not assessed
Alternative assessments: Not assessed
Fill: Independently studied

Kind: Not assessed
Related entries: Not assessed
Field: Tax compliance. Audit targeting and fraud detection.
Purpose: Target tax audits by cross-matching mass data to detect declarative inconsistencies.
Dates: Date basis unassessed: 2023 (precision unassessed). Updated: 21 October 2026.
Stage: Not assessed
Version: SHA-256: b75dc7ffc8ce…