Tax audits in France are undergoing a quiet but profound transformation: that of digitalisation and artificial intelligence. Long seen as a matter of on-site visits, paper-based accounts and ‘old-fashioned’ checks, they now rely on algorithms capable of analysing millions of data points in a matter of seconds. By 2025, more than half of tax audits will be triggered by alerts generated by data mining tools. The tax authorities, pioneers in Europe, are thus taking a new step forward in the fight against tax fraud.

Data mining at the heart of the DGFiP’s strategy

Le programme GALAXIE, développé par la DGFiP, recoupe automatiquement les données fiscales, sociales, bancaires et patrimoniales. Déclarations de TVA, comptes annuels, flux bancaires, registres immobiliers ou encore informations échangées dans le cadre de l’OCDE (CRS, DAC6, FATCA) : tout est analysé.

In 2024, this system enabled the identification of more than 60 % of external audit files. The most common anomalies relate to:

• discrepancies between the declared turnover and bank payments,

• organisations with unusual financial flows with overseas entities,

• taxpayers whose lifestyle appears out of step with their declared income.

Artificial intelligence for fraud detection

Since 2022, the DGFiP has incorporated AI tools capable of assessing a taxpayer’s tax risk. The algorithm assigns a risk score based on a range of criteria (sector of activity, volume of international transactions, frequency of anomalies).

A striking example: the use of AI to detect undeclared swimming pools using aerial photographs from the IGN has led to thousands of cases being regularised. But AI is no longer limited to this type of symbolic inspection: it now targets cryptocurrencies, e-commerce and complex cross-border flows.

A more efficient but more intrusive administration

For taxpayers, this revolution has two aspects:

• Increased efficiency: audits are more targeted and less random. A ‘clean’ taxpayer is therefore less likely to be audited unnecessarily.

• Enhanced scrutiny: the sources of data are vast, extending far beyond simple tax returns. Bank transactions, international transfers and even certain public data can trigger an audit.

This development raises issues relating to data protection and transparency of algorithms: can a tax audit initiated on the basis of an automated score – the exact criteria for which the taxpayer is unaware – be challenged?

How should one prepare for tax data mining?

In 2025, the best defence is still to plan ahead:

• Check that the various tax returns (VAT, corporation tax, income tax, property tax) are consistent with one another.

• Document cross-border flows and unusual transactions.

• Establish an internal monitoring system to detect and rectify anomalies promptly.

• Develop a communication strategy in the event of an audit based on ‘algorithmic’ indicators.

For international groups, transparency has become essential: a lack of documentation or clear explanation now raises suspicion.

Conclusion

Tax audits in the age of artificial intelligence are no longer merely an administrative formality. They are becoming a strategic and technological process, in which data is central to the selection process. For high-net-worth individuals and businesses alike, the rule is simple: any inconsistency will be detected sooner or later.

At Qualifisc, we help our clients secure their data flows, anticipate risk indicators and prepare their defence in this new digital landscape.