Linking e‑invoicing, TaxPro Max and identity registries could let AI sharply improve Nigeria’s tax detection and revenue assurance; the technology’s payoff, however, hinges on better data, institutional capacity and safeguards against bias and over‑automation.

Main Finding

AI can materially strengthen tax compliance, digital detection, informal-economy monitoring, and revenue assurance in Nigeria — but only when deployed on top of reliable, identity-linked digital tax infrastructure (registration portals, e‑invoicing, payment systems) and embedded within clear governance, oversight, and service-oriented administration. Without good data quality, explainability and fairness safeguards, and human oversight, AI risks automating weak signals and producing unfair or legally vulnerable outcomes.

Key Points

• Purpose: Comprehensive literature and institutional review (2021–2026) assessing how AI can improve taxpayer visibility, detect noncompliance, reduce revenue leakage, and support tax administration in Nigeria.
• Digital tax architecture matters: TaxPro Max (central taxpayer portal), the FIRS Merchant/Buyer e‑invoicing solution, and identity linkage (CAC, NIN paths) create the enforceable traces that AI needs to work effectively.
• Principal AI use cases: risk scoring/classification, anomaly detection, network analytics (ownership/invoice chains), text analytics (unstructured declarations), geospatial analysis (clustered noncompliance), and automated decision‑support for audit selection and workflow.
• Informal economy: Extremely large and cash‑intensive (NBS: ~93% informal employment Q2 2024). Full formalization is unrealistic; AI’s value is in aggregating many weak observable signals (mobile payments, supplier/customer links, bank deposits, repeated phone numbers, social‑media selling patterns, location clusters) into actionable risk scores.
• Practical benefits: Earlier detection, prioritized audits, reduced manual workload, improved invoice validation and payment matching, and better revenue assurance when data are integrated and near‑real‑time.
• Key risks and constraints: data quality and linking failures; model explainability and fairness; privacy and due‑process concerns; weak validation and benchmarking; vendor lock‑in and overreliance on automation that can erode professional judgement.
• Recommended operational stance: hybrid model — combine AI detection with strengthened taxpayer services, human review of flagged cases, transparent procedures, iterative validation, and measurable performance indicators.

Data & Methods

• Type of paper: narrative / comprehensive review (synthesis of literature and institutional evidence).
• Time window and sources: studies, official materials, and comparative evidence published 2021–2026 with emphasis on Nigerian empirical work and FIRS technical documents (TaxPro Max, Merchant Buyer Solution, identity linkage notes), plus IMF/ OECD/other country case studies (e.g., Estonia, Indonesia, EU comparative findings).
• Data inputs described (what AI would need): historical tax returns, audit outcomes, payment histories, e‑invoices and QR validation data, corporate registry records (CAC), bank and third‑party payment data, customs declarations, unstructured text fields (invoices/declarations), geolocation/business mapping, and telecom/mobile payments traces.
• Methods mapped to use cases (from the review):
  • Machine‑learning classification — needs labeled historical outcomes for risk scoring and audit prioritization.
  • Anomaly detection — transactional volumes and peer benchmarks to flag outliers (duplicate claims, sudden drops/ spikes).
  • Network analytics — linked entities and invoice chains to detect VAT chains and circular trading.
  • Text analytics — unstructured narrative extraction to spot inconsistent explanations and deceptive descriptions.
  • Geospatial analysis — location and density data to find low‑filing zones and informal clusters.
  • Automated decision support — integrated datasets to rank and route cases for human follow‑up.
• Limitations of method/data noted: most recommendations depend on integrated, identity‑linked, real‑time data; the reviewed evidence is primarily descriptive and comparative rather than from controlled experimental deployments in Nigeria.

Implications for AI Economics

• Investment economics and ROI: The value of AI in tax administration hinges on preexisting digital infrastructure. Economic assessments must account for complementary investments (identity linkage, e‑invoicing, payment integration) not just AI software. Cost–benefit models should include administrative savings, additional revenue recovered, and transition costs (training, governance).
• Measurement and causal inference opportunities: The linked administrative datasets described (invoices, payments, registration) create microdata that can support rigorous impact evaluation (RCTs or difference‑in‑differences) of AI interventions on compliance, revenue, and taxpayer behavior.
• Informality and measurement bias: AI methods that rely on digital traces risk systematic selection bias against fully cash‑based operators. Economists should model how observable digital footprints correlate with true economic activity and adjust revenue and compliance forecasts for unobserved informality.
• Distributional and behavioral effects: AI‑driven enforcement can change taxpayer incentives (deterrence vs. evasion tactics). Models should consider behavioral responses, potential crowding‑out of trust, and heterogenous effects across firm size, sector, and regions.
• Labor and productivity implications: AI can shift tax officers away from routine processing to investigative work; economic analyses should quantify task reallocation, required upskilling, and potential long‑run productivity gains or losses from automation.
• Governance and welfare externalities: Privacy, fairness, and due‑process risks have welfare and legal‑compliance costs. Economic frameworks for AI in public administration should include these negative externalities and the fiscal/legal costs of remedial actions.
• Policy design guidance for economists advising governments:
  • Prioritize data integration and identity linkage as prerequisites for AI gains.
  • Design pilot evaluations with clear performance metrics (revenue recovered per audit hour, false positive rates, taxpayer satisfaction, distributional impacts).
  • Incorporate explainability and auditability into procurement and evaluation criteria.
  • Budget for ongoing model validation, human oversight, and remedies for errors to avoid costly litigation or loss of public trust.
• Research directions: quantifying the marginal revenue effect of specific AI tools; studying heterogeneous impacts across formal/informal sectors; exploring equilibrium responses of taxpayers; cost‑effectiveness comparisons of traditional audits vs. AI‑assisted targeting.

Summary conclusion: AI is promising for Nigeria’s tax system but is complementary to — not a substitute for — investment in integrated digital tax infrastructure, governance safeguards, human oversight, and evaluation frameworks. Economists should treat AI deployments as policy interventions that require rigorous measurement of benefits, costs, and distributional consequences.