Intelligent Accounting Systems for Digital Enterprises: An AI‑Driven Framework for Control and Compliance

Abstract

Rapid digital commerce expansion exposed limitations in traditional accounting platforms regarding reconciliation accuracy, fraud detection, and compliance. To address this, we introduce a hybrid AI-driven framework combining supervised classifiers (XGBoost, DNNs) for reconciliation, unsupervised anomaly detectors (Isolation Forests, autoencoders) for fraud mitigation, transformer-based NLP (FinBERT, BERT) for regulatory tracking, and blockchain RPA workflows for immutable auditing. Evaluated across six core metrics using normalized, anonymized transaction and regulatory datasets, the system outperformed legacy platforms, achieving a 23.7% increase in reconciliation accuracy, an 18.3% latency reduction, an FDS of 0.91 (ROC-AUC = 0.94), an RCI of 0.89, and an EQ of 0.78. Integrating privacy preservation and model explainability, this architecture transitions accounting from basic automation to adaptive, regulation-aware financial intelligence.