evidoria

← Back to browse

Good practice Imported

FURS's eDIS VAT Risk-Scoring — Slovenia's Machine-Learning Models for Refund and Return Fraud Detection

Slovenia · Ljubljana · See the Slovenia profile · See the Ljubljana profile

Evidence: Descriptive / self-reported Top 75% 47/100 · Ask Evidence Copilot about this practice

Slovenia's FURS runs two ML risk-scoring models in its eDIS platform — one flags irregular VAT-refund claims since 2019, another scores every VAT return for fraud risk since 2022 — but a civil-society audit found no human-rights or data-protection impact assessment was ever done.

Details

Maturity
Established
Promoter
Finančna uprava Republike Slovenije (FURS) — Financial Administration of the Republic of Slovenia
Period
2019-present
Region (NUTS)
SI041
Keywords
tax administration, VAT compliance, fraud detection, risk analysis

Context

Slovenia's Financial Administration (FURS) operates the eDIS platform, which since 2019 has run a VAT-refund risk model (VAT-R) flagging irregular refund claims from foreign EU taxpayers, and since 2022 a VAT-control model that scores every domestic VAT return from 0 to 1 for fraud or error risk.

Objectives

Flagged cases are routed to human FURS staff for a final decision; the platform also supports AI-assisted document classification and a taxpayer-facing chatbot, developing roughly 1,000 candidate machine-learning models per risk area that are iteratively filtered down to a single production model.

Activities

The shared eDIS infrastructure underpinning these models cost €16.19 million. The OECD's 2026 Digital Government Scan of Slovenia documents the system's deployment and infrastructure investment as part of a broader government digitalisation review.

Conclusions

The independent civil-society Register of AI Use in the Slovenian Public Sector, maintained by the NGO Danes je nov dan, found that neither model has undergone a published human-rights or data-protection impact assessment, and that FURS has not disclosed training-data details for either model — a documented transparency gap that has persisted since deployment.

Implementation
Indicative cost
High (€500k–€5M)
Time to results
Long (> 3 years)
Staffing & skills
Finančna uprava Republike Slovenije (FURS) — operates and maintains both risk models, Human FURS staff make final decisions on flagged cases

Conditions for success

  • Shared eDIS infrastructure (€16.19 million) supporting multiple AI use cases including a taxpayer chatbot and document classification
  • Iterative model development — roughly 1,000 candidate models per risk area filtered to one production model

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Similar practices you may find useful