Artificial Intelligence-Driven Financial Reporting: Transforming Customized Financial Statements, Decision Intelligence, and Assurance Practices

Financial statements are prepared from trial balances that have passed internal-control checks and, for external reporting, have been audited before stakeholders rely on them. Converting this verified ledger into the views needed by investors, creditors, regulators, and management remains largely manual, causing rework, delays, and duplicated reconciliation. This study designs and evaluates an artificial intelligence (AI)-coordinated architecture that generates stakeholder-specific yet reconciled reports from a single governed trial balance. Following a design science approach, the architecture combines centrally configured accounting policies and KPI definitions, lineage metadata, account classification, anomaly detection, sequence forecasting, financial natural language processing (NLP), role-aware customization, and human-in-the-loop explainable AI (XAI). The prototype was evaluated on 15,000 trial balances for 500 entities (12,000 synthetic and 3,000 derived from public SEC filings) against rule-based, robotic process automation (RPA), and commercial generative-AI baselines. It achieved a line-item disclosure fidelity (F1) of 0.965, an AUC of 0.97 for misclassification detection, and a mean absolute percentage error (MAPE) of 4.8% for one- to four-quarter forecasts; in 250 simulated multi-entity scenarios, preparer effort fell from 38.0 to 4.2 person-hours per reporting cycle. Simulated stakeholder and auditor responses illustrate the expected gains in perceived usefulness and trust. Limitations include synthetic and statement-derived data, simulated rather than observed perception measures, and an unevaluated narrative-generation module.