Artificial Intelligence
AI Fraud Detection Platform
- Client
- Ministère des Finances
- Year
- 2025
- Category
- Artificial Intelligence
- Tech stack
- PythonTensorFlowBigQuery
Development of a machine learning model for automatic tax fraud detection, cutting state revenue losses by 45%.
Context
The Ministry of Finance suspected recurring tax fraud patterns but had no tool to systematically detect them among the millions of filings processed each year.
Solution
Our Data team built a BigQuery data pipeline consolidating filing history, then trained a machine learning model (TensorFlow) able to identify risk profiles with a fraud probability score.
The model is embedded in an MLOps pipeline that guarantees traceability and auditability: every decision can be explained and justified to an auditor — a non-negotiable requirement in a public context.
The model is retrained monthly on fresh data to adapt to evolving fraud patterns.
Results
The platform cut detected tax-fraud revenue losses by 45%, while reducing agents’ manual review workload through priority targeting of at-risk cases.
Gallery
