evidoria

← Back to browse

Good practice Imported

AGT's AI Cross-Referencing System — Angola's Tax Authority Flags 15,000 Companies for Fraudulent Zero-Revenue Declarations

Angola · Luanda · See the Angola profile · See the Luanda profile

Evidence: Descriptive / self-reported Top 67% 53/100 · Ask Evidence Copilot about this practice

Angola's tax authority AGT used AI to cross-reference e-invoices and import records, flagging 15,000 of ~40,000 firms that filed zero-revenue returns despite real trading activity. Notices went out in June 2026 — no independent accuracy audit has been published yet.

~40,000
Zero-revenue Industrial Tax declarations received (Apr–May 2026)
~15,000
Companies flagged for real activity despite zero-revenue filing (2026)
~300 billion kwanzas
AGT revenue, prior year
~430 billion kwanzas
AGT revenue, current year
AGT's AI Cross-Referencing System — Angola's Tax Authority Flags 15,000 Companies for Fraudulent Zero-Revenue Declarations

Details

Maturity
Scaling
Promoter
Administração Geral Tributária (AGT)
Period
2024-2026
Keywords
taxation, fraud detection, revenue administration, compliance

Context

Angola's General Tax Administration (AGT — Administração Geral Tributária) began testing artificial-intelligence tools in 2024 to close gaps in a tax base where a large share of registered companies report no taxable income.

Objectives

The system aims to cross-reference electronic invoices, import records and other commercial data in real time to flag inconsistencies between what a company declares and what it actually appears to be doing.

Activities

In April–May 2026, AGT received roughly 40,000 Industrial Tax declarations reporting zero revenue. Applying the AI system to this pool, AGT's Board Chairman José Leiria announced that about 15,000 of these companies showed evidence of real commercial activity despite claiming none, and AGT sent correction notices to the flagged companies in June 2026.

Results

Some flagged companies voluntarily amended their filings. Overall AGT revenue rose from roughly 300 billion kwanzas to 430 billion kwanzas year-on-year, though that increase reflects a broader package of reforms (taxpayer-ID regularisation, new market-based collection posts, expanded electronic invoicing) alongside the AI screening, not the AI system in isolation.

Conclusions

No independent audit of the tool's false-positive rate, or of how many flagged companies were ultimately found non-compliant, has been published, and no algorithm documentation, appeal process, or bias assessment has been made public.

Implementation
Indicative cost
Medium (€50k–€500k)
Time to results
Medium (1–3 years)
Staffing & skills
Administração Geral Tributária (AGT) under Board Chairman José Leiria

Conditions for success

  • Real-time cross-referencing of e-invoicing, import and commercial records
  • Complementary reforms: taxpayer-ID regularisation, new market-based collection posts, expanded electronic invoicing
  • Public communication of results to encourage voluntary correction

Common failure modes

  • No independent audit of the false-positive rate or eventual compliance outcomes published
  • No algorithm documentation, appeal process, or bias assessment made public

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.

Attachments

Similar practices you may find useful