Enhancing Financial Close with ML: Oracle Fusion Cloud Financials Case Study

The financial close process has always been one of the most important aspects of corporate finance. Month-end, quarter-end, or year-end closing cycles are directly linked to financial transparency, operational efficiency, and even missing the deadline or incorrect execution and subsequent fallback by the count bookkeepers can cause the results to be non-final or inaccurate, which can also lead to increased financial transparency, operational efficiency, and stakeholder trust risk. As transactions within multinational companies become increasingly complex, manual reconciliations and rule-based anomaly detection mechanisms struggle to keep pace. The current paper presents a case study of Oracle Fusion Cloud Financials, which has been augmented with Machine Learning (ML) algorithms to automate close processes, including anomaly detection in journals and transactions, as well as continuous auditing and reconciliation. In the examined case, Oracle Fusion Cloud introduced ML into its subledger accounting, general ledger, and reconciliation modules and trained them to identify irregularities in journal entries and automate the matching process of account reconciliations. The strategy concentrated on three key capacities: (1) supervised and anomaly free discovery to determine outlier journal designs; (2) predictive reconciliation to propose matches in unreconciled items; and (3) continuous monitoring dashboards to lessen reliance on end- of- the- batchful processes. Efforts during the deployment have shown that it takes 32 percent less time in close cycles and 57 percent better in anomaly detection than the traditional rules-based validations. In addition, the ongoing auditing model facilitated real-time-like compliance testing, which cued the finance teams to intervene at a more opportune time during the accounting period. The challenges of ML adoption issues are also analyzed in the case study, such as data quality problems, as well as the explainability of the models and their complexity of integration with the APIs of the Oracle ERP cloud. A hybrid ML is proposed that consists of unsupervised learning (Isolation Forest, Autoencoders) to be used to discover new anomalies and supervised classifiers (Random Forest, XGBoost) that are trained based on historical audit findings