Early positive trials on low-risk VAT reclaims encourage full automation
Italy’s Agenzia delle Entrate tax authority is introducing artificial intelligence to speed up VAT refund processing and enhance fraud detection, while keeping final decision-making under human supervision.
This follows recent Italian AI VAT chatbot adoption. This follows other tax authorities adopting AI.
Tax authorities report to Parliament
During an October parliamentary hearing, officials explained that AI is being embedded into the agency’s risk analysis and compliance systems to improve efficiency and accuracy. The technology is applied only at the pre-assessment stage — when refund claims are screened for inconsistencies — and does not influence audit or enforcement decisions.
The new system enables faster collection, organisation, and evaluation of large volumes of taxpayer and transactional data, helping auditors identify potential anomalies more effectively. The goal is not to replace human staff, but to strengthen their ability to focus on complex or high-risk cases.
Machine Learning consuming various data sources
To achieve the speeded up times to refund, VAT-relevant data flows into the Revenue Agency’s central tax data warehouse — including:
-
e-invoices (from the Sistema di Interscambio),
-
Cross-border transaction reports,
-
customs import/export declarations, and
-
banking and withholding-tax data.
AI and machine-learning models first normalise and cross-reference these datasets, cleaning duplicates, tagging counterparties, and aligning data structures for further analysis.
VAT anomaly and risk scoring (Pre-assessment phase)
Refund requests are scored automatically for risk before any human review.
Algorithms use a mix of:
-
Pattern recognition (e.g. sudden spikes in refund claims vs turnover, chain-transaction patterns, mismatched VAT ID activity),
-
Network analysis (to flag connections to known high-risk traders or carousel chains),
-
Historical profiling (based on past refund reliability, sector averages, and invoice density),
-
Textual AI for classifying invoice narratives or identifying unusual tax code usage.
The models don’t decide outcomes — they rank claims by likelihood of error or fraud, feeding dashboards used by auditors.
Cutting VAT recovery processing times
By 2026, low-risk VAT refunds are expected to be processed automatically, cutting waiting times and releasing liquidity to compliant businesses more quickly. Time saved through automation will be redirected to deeper investigations. Early data shows progress already: the average refund time fell to 67 days by September 2025, ahead of the Finance Ministry’s 70-day target for 2025.
Officials highlighted that Italy’s approach to AI remains cautious. Unlike some other EU countries that use automated scraping of online or social-media data, Italy’s Revenue Agency avoids such practices to uphold strict privacy and data-protection standards. Any future alignment with wider European digital-tax frameworks, they said, would need to preserve these safeguards.
