Evaluating AI-Driven Tax Compliance in Taxation Reform Economies

Thesis Abstract

The rapid digitization of tax administration and the deployment of artificial intelligence (AI) tools in compliance monitoring address growing revenue gaps and administrative inefficiencies in economies undergoing taxation reform. However, the effectiveness of AI-driven tax compliance initiatives remains contested due to concerns about accuracy, fairness, data quality, and impact on voluntary compliance. This study aims to evaluate the effectiveness, acceptance, and equity implications of AI-based compliance systems within taxation reform economies, with a view to informing policy design and institutional implementation. The specific objectives are (i) to assess the impact of AI-enabled risk-based auditing and real-time data analytics on tax collection performance; (ii) to examine taxpayer perceptions of AI-driven compliance mechanisms and their influence on voluntary compliance; (iii) to analyze the distributional and fairness implications of AI decisions across taxpayer groups; (iv) to identify organizational and technical determinants of successful AI integration in tax administration; and (v) to formulate actionable recommendations for governance, transparency, and capacity-building. The study adopts a mixed-methods research design anchored in a realist evaluation and theory-informed analysis. The quantitative component involves a cross-sectional survey of 1,200 taxpayers and 120 tax officials across three jurisdictions undergoing tax reform, complemented by administrative data on assessed taxes, detected non-compliance instances, and audit outcomes for a five-year window (2019–2023). The qualitative component includes 40 in-depth interviews with senior tax administrators, AI engineers, and frontline auditors, as well as 20 focus groups with taxpayers representing small, medium, and large enterprises. Data collection instruments comprise structured questionnaires, semi-structured interview guides, and documentary evidence from policy documents and system logs. Validity and reliability are ensured through pilot testing, expert panel reviews, triangulation, and confirmatory factor analysis for constructs related to perceptions of AI fairness, trust, and perceived effectiveness. Data analysis utilizes a combination of regression-based impact assessment (difference-in-differences where feasible, panel data techniques if longitudinal elements are retained), propensity score matching to mitigate selection bias, and robustness checks. The qualitative data are analyzed through thematic analysis guided by the Technology Acceptance Model (TAM) and the Bounded Rationality framework, with coding cross-validated by multiple researchers. A conceptual model integrating TAM, Social Exchange Theory, and Fairness in AI decision-making is employed to interpret results and to map pathways from AI deployment to compliance outcomes. Key expected findings include (a) AI-driven risk scoring and automated anomaly detection positively correlate with improved revenue collection and reduced time-to-audit, conditional on data quality and integration with human audit processes; (b) taxpayer trust and perceived fairness significantly mediate the relationship between AI use and voluntary compliance, with transparency about decision rules amplifying acceptance; (c) disparities in AI outcomes emerge across small vs. large taxpayers, sectors, and regions, driven by data sparsity, digital literacy, and historical enforcement patterns; (d) organizational readiness, including governance structures, cross-functional teams, and data governance maturity, moderates the effectiveness of AI systems; and (e) governance mechanisms—such as auditability, explainability, and independent oversight—ameliorate concerns about bias and errors. The study contributes to knowledge by providing empirical evidence on the performance, acceptance, and equity implications of AI-driven tax compliance within reformist economies, advancing theories of technology adoption in public administration, and expanding the empirical base for using realist evaluation in taxation contexts. It offers a practical framework for policymakers and practitioners to balance efficiency gains with fairness and legitimacy. The main conclusion is that AI-enabled tax compliance can enhance revenue and efficiency when embedded in an integrated governance model that emphasizes data quality, transparent decision rules, stakeholder engagement, and robust human oversight. Recommendations include (1) establishing transparent AI governance with explainable models and audit trails; (2) investing in data quality, interoperability, and capacity-building for tax officials; (3) designing targeted communication strategies to build taxpayer trust and improve perceived fairness; and (4) implementing continuous monitoring and independent evaluation to detect and remedy emergent biases and system inefficiencies.