Winter 2021 Today’s topics • Auditing complex IT environments – Testing electronic accounting systems and automated internal controls
• Use of artificial intelligence in auditing
– Introduction to machine learning and AI – MindBridge Ai auditor, group assignment 5 Auditing complex IT environments • Auditing through the computer—auditing by testing automated internal controls and account balances electronically
• Three approaches to test the effectiveness of
automated controls when auditing through the computer: – Test data – Parallel simulation – Embedded audit module approach Test data • Uses the auditor’s test data to determine whether the client’s computer program correctly processes valid and invalid transactions
• Three main considerations when using this approach:
1. Test data should include all relevant conditions that the auditor wants tested. 2. Application programs tested by auditors’ test data must be the same as those the client used throughout the year. 3. Test data must be eliminated from the client’s after testing Parallel simulation • Involves the auditor’s use of audit software to replicate some part of a client’s application system
• Auditors commonly do parallel simulation testing
using generalized audit software (GAS) – Computer programs used by auditors that provide data retrieval, data manipulation, and reporting capabilities specifically oriented to the needs of auditors. Embedded audit module approach • A method of auditing transactions processed by IT whereby the auditor embeds a module in the client’s application software to identify transactions with characteristics that are of interest to the auditor; – The auditor is then able to analyze these transactions on a real-time, continuous basis as client transactions are processed.
• In recent years, artificial intelligence has started to
become used in audits to identify high risk transactions, and transactions patterns associated with errors/fraud Machine learning • A form of artificial intelligence that uses pattern recognition and inference to improve a program/algorithm without explicit instructions Unsupervised learning Supervised learning Machine looks for patterns or Machine is trained to find groupings in big data. Machine specific patterns or groupings in is not taught what to look for. big data. Machine receives training data and knows what it is looking for. Applying AI to financial audits • AI audit programs are trained to detect errors and fraud using: – Analysis of transaction/account relationships – Machine learning – Statistical methods – Input of auditing experts
• AI programs recognize and highlight irregular or
high risk transactions (or groups of transactions) Augmenting financial anomaly detection with machine learning and AI
Domain expertise Statistical Machine Cross
business rules Modelling Learning correlation
MindBridge‘s Standard tests Benford’s law, Expert Score, Correcting Controls
Ai Auditor Advanced tests Statistical Outlier detection combining methods Completeness checks Variance Reinforced Learning Traditional audit activities that AI can aid in the completion of • Find journal entries that do not balance • Find gaps in journal entry number sequence • Find high-dollar journal entries • Find round-dollar journal entries • Sample journal entries • Find all entries containing specific account(s) • Find entries posted on weekends Learning to audit • Machine learning also allows for the AI auditor to find: – Journal entries with accounts that are not typically seen together – Journal entries that appear in a pattern that is not typical of regular business operations – Incomplete pairs of journal entries such as sales and COGS Example of a CFO fraud scheme • Jack (CFO and a former auditor), knew most of the audit procedures a company performs at year-end and more importantly how to evade them: – Recorded fictitious sales in small amounts to avoid creating an unusually large transaction. – Was also careful to record the sales at least five days before year-end to avoid any cutoff tests.
• How can this fraud be more reliably detected?
AI’s assistance in auditing • This type of fraud would be much more quickly detected by AI than by standard audit procedures – Fictitious sales invoices – Sales with no corresponding cost of goods sold • Exaggerated profit margin for otherwise legitimate sale – Otherwise legitimate revenue pulled into earlier accounting periods – Pushed cost of goods sold into later accounting periods – Double-booked sales transactions Case Study A: Details • Builder Inc. is an innovative construction company located in Western Canada. It employs staff of 150 and is relatively successful. Even though the company experiences steady growth in recent years, competition is fierce and margins are shrinking.
• A significant portion of the CEO’s pay (Steve Smith) is
profit-related. During audit planning this risk is identified and appropriate audit procedures are to be planned and executed to address this risk. Case Study A: General Journal • The company provided you with a partial General Journal (Case_A data set.xlsx). • The total balance of the Journal is around $4.4 million. • The FY start date is January 1. – For all cases, you should make the fiscal year studied 2014. • Note that the materiality for this audit engagement is $10,000 – Reduce the weighting for the material value control point to 5% in the MindBridge tool. • It is also important to note that bookkeeping is regularly done on weekends. Case Study A: Task • Analyze the general journal using the MindBridge tool and addressing the above fraud risk. • Identify no more than 20 transactions using the various investigative features in MindBridge’s Ai Auditor. • Create ‘tasks’ for these transactions and export them from the Audit Plan section.