Page 83 - Demo Faces

Traditional audit testing is time-consuming, entails manual
data gathering and review, and is limited to a small number of transaction samples. As a result, classifying countries as low, medium, or high risk is too subjective and prevents incorporation of more indicators that could improve accuracy.
By automating the process and using data science to supplement auditors’ judgment, the system can be improved. The project comprises three elements. First, a continuous monitoring system checks all treasury transactions and sends exception reports daily, uses machine learning, and employs concepts from robotics process automation.
Second, the eOps loan covenant checking system automatically downloads project documents, converts them into text
les, checks the loan covenant clauses, and summarizes any mismatches or missing clauses. This saves auditors an average of 40 to 60 hours of manual checking.
Third, the country priority assessment model delivers a second opinion on which country the O ce of the Auditor General has to prioritize for audit, classi es countries by risk, and even extracts news articles from Standard and Poor’s news service for sentiment analysis.
“For the past few years, the O ce of the Auditor General has been on this journey to innovate some of our processes for us to achieve e ciency and obtain greater assurance in our audits. We’ve also considered innovative ways to help strengthen and improve the internal control processes of our clients.”
Candy Chao
Senior Audit O cer, O ce of the Auditor General
83
Implementation and Internal Processes