AI-AUGMENTED TAX RISK SCORING FOR SMALL AND MEDIUM ENTERPRISES: A PANEL DATA STUDY

Abstract

This study investigates the impact of AI-augmented tax risk scoring systems on small and medium enterprises (SMEs) over a twelve-year period (2010–2022), employing a comprehensive time series methodology that integrates both classical econometric techniques and advanced machine learning models. By situating the analysis within three distinct phases—pre-AI (2010–2014), transitional e-invoicing adoption (2015–2017), and post-AI hybrid implementation (2018–2022)—the study captures structural and dynamic shifts in compliance indicators, model performance, and sectoral disparities. Data were sourced from national tax authority archives, SME financial statements, transactional data feeds, and macroeconomic indicators, harmonized into a quarterly panel to ensure analytical consistency. Statistical modeling employed ARIMA to isolate baseline trends and seasonality, VAR to examine interdependencies between compliance rates and model accuracy, and LSTM neural networks to capture non-linear, temporal anomalies. Findings reveal two statistically significant regime shifts: first, the 2015–2017 e-invoicing rollout, which reduced high-risk classifications by approximately 14% and increased anomaly detection by 17%; and second, the 2018 AI-hybrid adoption, which improved predictive precision by 12.4%, recall by 15.2%, and reduced compliance rate volatility by nearly 20%. Digitally intensive sectors such as e-commerce and ICT services experienced the largest compliance gains (22–28% risk score reduction), while high-informality sectors achieved modest improvements (6–9%). Spatial analysis demonstrates positive inter-jurisdictional spillovers, with metropolitan adoption driving a 4–6% compliance efficiency increase in neighboring regions. Overall, the results confirm that AI-augmented tax risk scoring, when paired with diverse data integration and robust governance, enhances detection efficiency, stabilizes compliance trends, and progressively reduces regional and sectoral disparities in SME tax enforcement. These outcomes have significant implications for the design of adaptive, transparent, and equitable compliance monitoring systems in tax administrations worldwide.