Verifiable digital tax trails — e‑invoices, electronic sales registers and consumer receipt incentives — reliably lift VAT revenue in many emerging economies, but only when administrations can act on the data; simple digitisation without new third‑party information produces modest, short‑lived gains.

Main Finding

Digital audit and financial information systems raise tax compliance and transparency in emerging economies most reliably when they produce verifiable third‑party information trails and when administrations have the capacity and political commitment to act on that data. Simple digitisation of processes without new cross‑checkable information yields smaller and less durable gains; continuous assurance and RegTech offer promise but their fiscal returns are conditional and less well‑established.

Key Points

• Strongest causal evidence: systems that create verifiable third‑party information (e.g., VAT paper trails, e‑invoicing, electronic sales registers, consumer receipt incentives) produce the largest and most durable compliance gains.
  • Example studies: Kleven et al. (2011) RCT on third‑party reporting (very low evasion); staggered/DiD and panel studies from Peru, Rwanda, China, Ethiopia, Brazil showing +5% or larger increases in reported sales/VAT liabilities in many settings.
• Digitisation without new information is fragile: e‑filing/portals that merely digitise filing reduce compliance costs but do not necessarily reduce underreporting.
• Displacement/substitution: when some margins become verifiable, misreporting can migrate to uncovered margins (inflated costs, transfer pricing, credit stocks), limiting net revenue gains unless coverage is comprehensive.
• Adoption and impact are conditional on institutional factors: administrative capacity, data quality, audit follow‑up, complementary enforcement, and political will drive whether technology converts into sustained revenue.
• Continuous auditing and analytics reframe assurance from episodic sampling to near‑real‑time population monitoring; adoption in emerging economies lags due to skills, infrastructure, governance constraints.
• Distributed ledger / RegTech offer conceptual advantages for data governance and traceability but currently rest more on conceptual and cross‑sectional foundations than on strong causal fiscal evidence.
• The policy lesson: design and sequencing matter — prioritize technologies that create verifiable, matchable data and build complementary capacity before layering sophisticated analytics.

Data & Methods (used in the review)

• Type of paper: narrative review synthesising empirical and conceptual work published 2006–2026 across public economics, accounting/assurance, and information systems literatures.
• Primary empirical methods surveyed:
  • Randomised controlled trials (e.g., Kleven et al. on third‑party reporting).
  • Quasi‑experimental approaches: staggered difference‑in‑differences, event study designs (e.g., e‑invoicing rollouts in Peru, China).
  • Administrative panel data analyses (e.g., electronic sales register evaluations in Ethiopia; Rwanda e‑invoicing).
  • Natural experiments and staggered rollouts (consumer incentive schemes in Brazil).
  • Cross‑sectional and conceptual studies for distributed ledger/RegTech and continuous assurance.
• Evidence grading (as summarized in the review):
  • Strong causal evidence: third‑party reporting effects, e‑invoicing, electronic sales registers, consumer receipt incentives.
  • Conditional/moderate evidence: electronic filing/payment platforms (effects vary by implementation and complementary enforcement).
  • Emerging/weak evidence: population‑scale continuous assurance and blockchain/RegTech for fiscal gains (promising conceptually but limited causal evaluation).
• Data sources typically: tax administration records, invoice logs, POS transmission data, audit outcomes, and sometimes taxpayer surveys.

Implications for AI Economics

• Where to apply AI effectively
  • Highest ROI: apply AI/ML to systems that already generate verifiable third‑party data (invoice matching, POS streams, inter‑firm purchase/sales ledgers). AI can scale matching, anomaly detection, risk scoring, and prioritised audit selection across population data.
  • Lower ROI if data is weak: AI models trained on poor‑quality, sparse, or manipulable data will produce fragile or misleading signals; investments in data infrastructure should precede advanced AI.
• Sequencing and capacity
  • Sequence reforms: (1) establish comprehensive, verifiable data capture (e‑invoicing, POS registers), (2) build data governance and audit workflows, (3) deploy AI analytics for continuous assurance and prioritised enforcement.
  • Capacity building is essential: AI adoption requires staff with data science skills, processes for model validation, and operational capacity to act on model outputs (audits, notifications).
• Behavioral and political economy considerations
  • AI changes the compliance equilibrium: higher detection probability via analytics may increase enforced compliance but can erode trust if perceived as opaque surveillance. Design transparency and taxpayer communication matter for preserving voluntary compliance.
  • Substitution/adversarial adaptation: expect evasion to move toward uncovered margins; AI systems must be continuously updated and combined with expanded data coverage to avoid yielding only marginal gains.
• Governance, privacy, and robustness
  • Data governance and privacy protections are prerequisites—RegTech, cryptographic techniques, and access controls can enable analytics while limiting misuse.
  • Model robustness and auditability: AI used for enforcement should be interpretable, auditable, and stress‑tested against adversarial manipulation of inputs.
• Research priorities for AI economics
  • Causal impact evaluations of AI analytics interventions (RCTs or credible quasi‑experiments) measuring revenue, taxpayer behavior, displacement effects, and welfare.
  • Cost‑benefit analyses comparing (a) extending third‑party data coverage vs (b) investing in AI on existing data, to guide investment sequencing.
  • Studies on how AI affects taxpayer trust, compliance morale, and political acceptability in low‑ and middle‑income contexts.
  • Work on algorithmic governance: governance architectures that balance effectiveness, explainability, and rights.
• Practical takeaways for policymakers and AI economists
  • Don’t treat AI as a substitute for building verifiable data trails and administrative capacity; it is a force multiplier when those foundations exist.
  • Prioritise simple, high‑impact data coverage reforms (e‑invoicing, POS capture, consumer receipt schemes) before complex ML deployments.
  • Embed rigorous evaluation and transparency requirements into AI deployments in tax administration to measure real fiscal returns and social effects.

If you want, I can (a) produce a one‑page visual checklist for policymakers on sequencing data vs AI investments, or (b) propose specific evaluation designs (RCTs / staggered rollouts) for testing AI audit tools in a tax administration.