Z SCORE AND VALIDOAI: AN EXPLAINABLE AI FRAMEWORK FOR PAYROLL ANALYTICS WITH STATISTICAL ANOMALY DETECTION IN SERBIAN SMALL AND MEDIUM ENTERPRISES
Abstract:Labor costs represent one of the largest components of operating expenses in Serbian small and medium-sized enterprises (SMEs), often exceeding 50% of total expenditures. However, existing payroll systems such as Pantheon and Minimax are limited to regulatory compliance and basic reporting, offering little analytical depth. This paper presents the ValidoAI explainable AI framework, which combines automated data processing (ETL) with statistical monitoring of payroll deviations. Twelve months of payroll data from PDF payslips and CSV exports were standardized into six categories: net salary, employee contributions, employer contributions, income tax, Gross 1, and Gross 2. Statistical analysis using the z-score method (|z| ≥ 2) identified significant deviations in October and November, corresponding to bonus payments and seasonal adjustments. The standardization accuracy reached 95.2%, with 98.7% data completeness and 97.1% internal consistency across 1,248 standardized records from six employees over twelve months. Results are presented from two complementary viewpoints: business owners gain insight into total labor costs and liquidity through trend visualizations showing anomalies like October’s 36% increase, while HR managers receive detailed wage structure reports for fairness evaluation and retention planning, including seasonal patterns such as June’s proportional rises in compensation components. The study highlights the potential for payroll systems to evolve into tools for cost monitoring and financial decision-making.
Keywords: Explainable AI, Payroll Analytics, Statistical Anomaly Detection, Managerial Decision-Making, Labor costs