Navigating the Risks of the use of Digital Reporting and E-Invoicing in the Age of Advanced Fraud Detection

Paper from Madeleine Merckx

This paper critically examines the risks posed by automated fraud detection systems in the context of the EU’s VAT in the Digital Age (ViDA) initiative, which mandates digital reporting and e-invoicing across Member States by 1 July 2030.

The central concern: while these technologies promise better fraud detection, they also risk false accusations, bias, function creep, and lack of legal safeguards—especially when algorithmic predictions are treated as definitive proof of fraud.

Key Risks Identified

1. Misuse of Predictions

Fraud detection systems generate probabilities, not certainties.
Treating algorithmic predictions as proof of fraud can lead to false positives, harming compliant businesses.
Example: Poland’s STIR system can block bank accounts based on algorithmic suspicion without full transparency.

2. Data Quality & Model Accuracy

Models rely on historical data, which may be biased, incomplete, or outdated.
Without regular retraining and validation, systems can become inaccurate.
Supervised learning is good for known fraud patterns; unsupervised and active learning are needed to detect new ones.

3. Bias & Function Creep

Algorithms may unintentionally discriminate against certain groups (e.g., SMEs).
Function creep occurs when data collected for one purpose is used for another—e.g., linking VAT data to CESOP or TNA systems without safeguards.

4. Lack of Legal Protection

Many systems operate as black boxes, making it hard for taxpayers to understand or challenge decisions.
This undermines transparency, due process, and fundamental rights.

Technical Insights

Fraud detection has evolved from rule-based systems to AI-driven models using:
- Supervised learning (e.g., decision trees, neural networks)
- Unsupervised learning (e.g., anomaly detection, clustering)
- Active learning (combining both)
- Social Network Analysis (SNA) to detect group-level fraud like carousel schemes

️ Recommendations

Use fraud detection tools to support human decision-making, not replace it.
Ensure transparency, scientific rigor, and periodic audits of models.
Consider legal safeguards, including independent oversight and explainable AI (XAI).
Avoid over-enforcement and ensure systems distinguish between errors and fraud.

Conclusion

While digital reporting and e-invoicing offer powerful tools to combat VAT fraud, their success depends on responsible implementation. The authors urge caution, emphasizing that automated systems must be transparent, fair, and legally accountable to avoid harming honest taxpayers

Source Madeleine Merckx