One model, applied country by country
The African Revenue Intelligence Research Series is a 17-paper research programme applying a single, consistent methodology — the African Revenue Intelligence Model (ARIM) — to tax fraud detection, compliance analytics, and domestic revenue mobilisation across the African continent.
The series comprises one continental foundational paper (Paper 1), fifteen country-specific research papers (Papers 2–16), and one comparative cross-country synthesis paper (Paper 17). Every paper shares a common mathematical architecture, a shared 28-part academic paper structure, and strict continental research standards, while each country paper develops a genuinely country-specific research question, data architecture, and empirical simulation rather than a re-skinned duplicate.
The 17 Papers: One foundation, fifteen country studies, one comparative synthesis
The 17 papers span 15 sovereign tax administrations, capturing diverse fiscal environments: real-time invoice matching in East Africa, extractives transfer pricing in Southern and West Africa, dual-authority petroleum capex in Nigeria, multi-currency volatility in Zimbabwe, dual-calendar cycles in Ethiopia, and corporate rate cliffs in North Africa.
| # | Paper Title & Scope | Jurisdiction | Core Analytical Focus | Status |
|---|---|---|---|---|
| Paper 1 | An AI-Driven Revenue Intelligence Framework for Tax Fraud Detection and Revenue Mobilisation in Africa | Continental | Foundational ARIM mathematical model, 6 behavioural domains, graph analytics, and human decision chain | Published |
| Paper 2 | AI-Driven Revenue Intelligence for Tax Compliance in Kenya | Kenya (KRA) | eTIMS invoice-chain validation, real-time input-tax credit reconciliation, and M-Pesa mobile-money telemetry | Published |
| Paper 3 | AI-Driven Revenue Intelligence for Tax Compliance in South Africa | South Africa (SARS) | Corporate transfer pricing, extractive mineral valuation anomalies, and high-net-worth compliance networks | Published |
| Paper 4 | AI-Driven Revenue Intelligence for Tax Compliance in Nigeria | Nigeria (FIRS / NUPRC) | Dual-authority petroleum/corporate compliance, upstream capex verification, and non-oil VAT expansion | Published |
| Paper 5 | AI-Driven Revenue Intelligence for Tax Compliance in Ghana | Ghana (GRA) | Closing the 2.1% GDP VAT compliance gap, downstream fuel-marking reconciliations, and gold royalty tracking | Published |
| Paper 6 | AI-Driven Revenue Intelligence for Tax Compliance in Rwanda | Rwanda (RRA) | Electronic Billing Machine (EBM v2) invoice authenticity, automated input credit matching, and border trade telemetry | Published |
| Paper 7 | AI-Driven Revenue Intelligence for Tax Compliance in Tanzania | Tanzania (TRA) | Electronic Fiscal Device (EFD) telemetry, Dar es Salaam port transit corridor fraud, and mineral export tracking | Published |
| Paper 8 | AI-Driven Revenue Intelligence for Tax Compliance in Uganda | Uganda (URA) | EFRIS digital fiscal receipt validation, telecom airtime & mobile money excise leakages, and cross-border VAT reconciliation | Published |
| Paper 9 | AI-Driven Revenue Intelligence for Tax Compliance in Zambia | Zambia (ZRA) | Sliding-scale copper mineral royalties (LME price bands), mineral production verification, and mining input VAT refund audits | Published |
| Paper 10 | AI-Driven Revenue Intelligence for Tax Compliance in Zimbabwe | Zimbabwe (ZIMRA) | Multi-currency (ZiG vs. USD) volatility, bifurcated corporate tax returns, and parallel exchange rate audit normalization | Published |
| Paper 11 | AI-Driven Revenue Intelligence for Tax Compliance in Botswana | Botswana (BURS) | Pre-e-invoicing audit architecture, diamond value chain anomaly detection, and SACU common revenue pool verification | Published |
| Paper 12 | AI-Driven Revenue Intelligence for Tax Compliance in Namibia | Namibia (NamRA) | Sector-differentiated corporate tax tiers (mining vs. manufacturing vs. services) and SACU regional trade data sharing | Published |
| Paper 13 | AI-Driven Revenue Intelligence for Tax Compliance in Ethiopia | Ethiopia (MoR) | Dual-calendar temporal reconciliation (13-month Ge'ez vs. Gregorian), FX allocation tracking, and informal economy transition | Published |
| Paper 14 | AI-Driven Revenue Intelligence for Tax Compliance in Senegal | Senegal (DGID) | SENTAX digital platform integration, minimum alternate tax (Impôt Minimum Forfaitaire), and telecom billing surveillance | Published |
| Paper 15 | AI-Driven Revenue Intelligence for Tax Compliance in Côte d'Ivoire | Côte d'Ivoire (DGI) | Facture Normalisée Électronique (FNE), agricultural supply chains (cocoa & cashew), and cross-border West African transit | Published |
| Paper 16 | AI-Driven Revenue Intelligence for Tax Compliance in Morocco | Morocco (DGI) | Progressive corporate income tax rate cliffs (10% to 31%), Tanger Med PortNet customs telemetry, and offshore entity auditing | Published |
| Paper 17 | Comparative African Revenue Intelligence: A Cross-Country Framework for AI-Enabled Tax Compliance Across 15 African Tax Administrations | Comparative Synthesis | Multi-jurisdictional tax maturity index, cross-border transfer pricing synthesis, and continental capacity roadmap | Published |
Each country paper (Papers 2–16) targets approximately 12,000–18,000 words. The foundational (Paper 1) and comparative (Paper 17) papers target 15,000–20,000 words each.
The same 28 parts, every time
To ensure academic reproducibility, legal transparency, and comparability across jurisdictions, every paper in the 17-part programme follows an identical 28-section academic structure:
This standardized taxonomy allows revenue commissioners, fiscal policymakers, and data science teams to compare how identical algorithmic components (such as GNN-based entity resolution or SHAP explainability) behave across distinct regulatory regimes.
Research standards applied across the series
Every paper in the series is bound by seven strict, non-negotiable research protocols defined across the series architecture:
Authoritative, Current Sources
Data and legal citations must derive exclusively from verified statutory authorities, revenue authorities (KRA, SARS, FIRS, GRA, RRA, TRA, URA, ZRA, ZIMRA, BURS, NamRA, MoR, DGID, DGI), finance ministries, central banks, national statistical offices, ATAF, IMF, World Bank, OECD, AfDB, and peer-reviewed literature. No statistics, datasets, or citations are fabricated.
Rigorous Typological Distinctions
Every typology analysis strictly distinguishes legitimate tax planning, aggressive tax planning/avoidance, unlawful tax evasion, deliberate fraud, administrative error, and normal commercial friction. No compliant taxpayer is conflated with a fraudulent actor.
Hypothetical Simulation & Entity Protection
Every synthetic dataset and empirical case study is explicitly labelled hypothetical. No synthetic taxpayer is a real commercial enterprise, and no hypothetical case constitutes an allegation against an identifiable business or individual.
Realistic Data Availability Classification
No model assumes a dataset exists. Every information source is explicitly classified as Essential, Desirable, Optional, or Potentially Unavailable, and the mathematical and operational consequence of a missing dataset is documented in the technical appendix.
Prioritisation Signal, Never Proof of Fraud
Every AI risk score is exclusively an audit selection and triage signal. No risk score constitutes a legal finding of liability or fraud. Audit selection, field investigation, administrative assessment, tax appeal, and legal prosecution remain distinct, human-decided stages.
No League Tables or Country Rankings
The series strictly rejects competitive league tables or value judgments asserting that one administration is superior to another. Paper 17 evaluates capacity through multi-dimensional maturity matrices, celebrating contextual adaptation rather than ranking.
Institutional Capacity Realism
Data availability, digital infrastructure, and institutional audit capacity limitations are explicitly identified rather than assuming uniform, frictionless capacity across the continent.
Six domains, one shared mathematical architecture
Every paper in the series scores a taxpayer using the exact same underlying mathematical formulation, adapted with country- or sector-specific variables rather than inventing a disconnected model:
Paper 1 defines each domain in full; country papers instantiate those same six domains with locally available variables:
Tax behaviour
Patterns within filed returns, adapted to local tax regimes — such as corporate rate-cliff clustering in Morocco (Paper 16) or minimum tax filings in Senegal (Paper 14).
Transaction behaviour
Real-time invoice flows and payment streams, including eTIMS discrepancy matching in Kenya (Paper 2), EBM v2 in Rwanda (Paper 6), and FNE in Côte d'Ivoire (Paper 15).
Network behaviour
Graph analytics tracking circular trading rings, beneficial ownership clusters, and high-net-worth entity structures in South Africa (Paper 3).
Sector behaviour
Sector-calibrated benchmarks — such as LME sliding-scale copper royalties in Zambia (Paper 9) and upstream capex claims in Nigeria (Paper 4).
Historical compliance
Longitudinal audit adjustments, appeal history, and payment latency — reconciled across dual-calendar fiscal years in Ethiopia (Paper 13).
The decision chain never changes
Every paper enforces the exact same non-negotiable legal separation between computational risk scoring and administrative legal enforcement:
A machine learning risk score is exclusively a case-selection and prioritisation signal. It indicates to an audit team where evidence may be missing or contradictory. The substantive legal finding of tax liability, assessment, and legal enforcement remains entirely human-decided at every gate.