Page 82 - Demo Faces

82
DATA SCIENCE AND ROBOTICS FOR AUDIT
Project name:
Transforming internal audit with data science and robotics process automation
Year:
2019
Team:
Do Keun Cho, Wenyi A. Wang, Pierre-Francois Gadpaille,
Adrian C. Edrosa, Christ Ian B. Badana, Candy Chu-Chao
Manual data gathering and review
CHALLENGE: Traditional audit testing
Takes a long time to complete
Limited to small number of transaction samples
Classifying countries as low, medium, high-risk
Manual and subjective
Limited auditors to incorporate more indicators which are not easily assessable
SOLUTION
1. Continuous monitoring system
Checks
all treasury transactions and send exception reports to the users on a daily basis
Incorporates
machine learning using Long Short-Term Memory
Employs
robotics process automation concepts
2. eOps loan covenants checking system
Automatically
downloads project documents, converts each document to text files, checks loan covenants clauses against eOps covenants monitoring pages, and produces a summary of missing or mismatched clauses
Saves
auditors an average of 40-60 hours of manually checking loan documents
NEXT STEPS
3. Country priority assessment model
Delivers
a second opinion on which country OAG has to prioritize for audit
Employs
broader data inputs and clustering (k-Means and DBSCAN) to group countries into di erent risk classification
Extracts
news article titles from S&P for the sentiment analysis
Process automation and data science to supplement our auditors’ judgement
The loan covenants tool
Further testing for accuracy and exception handling
Applying it for all active projects in eOps
The concept may be expanded to cover other tests for completeness
Country priority assessment model
Exploring supervised machine learning to automatically classify countries as low, medium, and high-risk based on some indicators/factors