Artificial intelligence-driven continuous auditing: Machine learning frameworks for real-time fraud detection in digital budgets and enterprise systems
There is an increasing amount of fraud in digital budgets and enterprise systems that traditional audit methods are not able to keep up with. Periodic audits that are sample based are often ineffective in detecting fraud occurring between audits. Today, with organizations transitioning to enterprise resource planning (ERP) and digital budgeting tools, transactions have become so large as they cannot be managed through manual checks. This situation provides a resourcing opportunity, as audit practice has lagged behind digital transformation. This paper is aimed at filling this gap with the help of a systematic literature review with the PRISMA design. Machine learning and deep learning have been used as the main approach for fraud detection, including the random forest model, gradient boosting machine, and neural network. Real-time monitoring is done with the assistance of anomaly detection and predictive analytics, while explainable AI is essential for gaining auditor trust and regulatory acceptance. In order to boost continuing auditing pipelines, more frequently, the three technologies, robotic process automation, blockchain and machine learning are combined. Compared to private sector financial statement fraud, public sector budgeting systems and enterprise risk management (ERM) remain underexplored. Not only that, the review will also highlight incompleteness in the metrics used to evaluate, less test on real-world behaviors and yet uncorrected data privacy issues in the studies. Such holes hamper the potential for repeatability and adoption. It synthesizes its argument by stating that continuous auditing is evolving from a compliance tool to a governance component.