Audit Applications
Auditors play a critical role in our society by providing assurance services related to financial statements and internal controls. Auditors render an opinion after accumulating and evaluating evidence (i.e., data). Given the ability of Al systems to analyze vast amounts of structured and unstructured data, Al can be employed in all phases of a financial statement audit - from planning, data gathering in fieldwork, completing the audit, and issuing the final audit opinion.
Baldwin et al. (2006) documented several applications of Al for auditing tasks. For example, neural networks have been used for performing analytical review procedures and risk assessments. Al algorithms can be used to assist with classification tasks for transactions (e.g., collectible debt vs. bad debt, legitimate transaction vs.
Applications of Al in Accounting 27 unauthorized). ES has been used for materiality assessments, internal control evaluations, and going concern judgments.
Although many examples of the applications of Al in audit mentioned in the literature are initiatives from large firms, small firms can also take advantage of Al for audits. For instance, Garbelman Winslow CPAs, a small firm based in Marlboro, Maryland, uses an Al platform called Ai Auditor developed by MindBridge Analytics to identify high-risk transactions during the audit planning process. Specifically, the firm uses machine learning algorithms to analyze the entire general ledger. The Al platform then compares the general ledger data against benchmarks to group the transactions into three risk categories: low risk, medium risk, and high risk. As Samantha Bowling (2019), partner at Garbelman Winslow CPAs, stated in her article: "Al is absolutely more comprehensive; it alerts us when things don’t look right and tells us where to start and where our risk is going to be. It shows the risk at the transaction level” (para. 3).
Bowling also stated that small public accounting firms could implement the technology for less than $10,000, but the price depends on the size of the firm. In her article, Bowling describes that because the system is cloudbased, she can connect her QuickBooks clients directly to the MindBridge platform and upload the entire general ledger. The software then categorizes the transactions into the different risk buckets at the transaction level. Bowling notes that using AI gives their firm a competitive advantage over firms that use traditional sampling. She also uses MindBridge when determining whether to accept or reject a client and in pricing their services. If a client has several risky transactions, she can either reject or raise the price appropriately to account for the increased risk.
Given the capabilities of AI for the audit function, it is no surprise that the world’s largest accounting firms, including the Big 4, have invested hundreds of millions of dollars in various Al-related technologies (Rapoport, 2016). K.PMG, for example, uses Clara, an audit platform that integrates AI and automation to read and extract data from both structured data (e.g., general ledger) and unstructured source documents (e.g., paper-based invoices, emails, instant messages, social media). The data is then reconciled to company records, and any discrepancies are flagged for review by humans. A task that originally took several hours is reduced to several minutes, freeing up the audit team to focus on other areas of higher risk (K.PMG, 2019).
Deloitte’s Argus tool uses machine learning and natural language processing to extract key accounting information from just about any type of electronic document (e.g., sales agreements, leasing and derivative contracts, invoices, meetings minutes, and legal letters toimprove the quality and efficiency of an audit) (Deloitte, 2015). Argus can review a population of documents, such as leases, and then uses ML to isolate and visualize any modifications to a contract that deviated from a standard contract, and then export the information into a workpaper for additional analysis by the auditor (Raphael, 2017).
EY uses natural language processing to analyze lengthy text documents such as legal contracts and leases to determine compliance with accounting standards (Nickerson, 2019). EY developed a tool called EY Document Intelligence, which uses natural language processing to review thousands of contracts as part of an audit engagement. Traditionally, auditors might manually review a single contract that contained hundreds of pages to identify key terms. The EY Document Intelligence tool analyzes a much larger volume of contracts in less time with greater accuracy. For example, in analyzing a lease agreement, the EY tool extracts the commencement date, lease amounts, and any related clauses. The auditor then selects the most relevant values using their professional judgment (Duffy, 2019). The tool continues to learn each time that it interacts with an EY auditor, becoming more effective over time. EY notes that deep learning will enable machines to analyze large amounts of unstructured data such as emails, social media posts, and conference call audio files.
PwC* partnered with H2O.ai, an Al firm based in the Silicon Valley, to build GL.ai which is a bot that analyzes “billions of data points in milliseconds, seeing what humans [cannot], and applying judgment to detect anomalies in the general ledger” (PwC, n.d., para. 1). The algorithm is designed to replicate the decision making process of an experienced auditor by analyzing all transactions in the general ledger. The more tasks the Al tool analyzes, the smarter it becomes. PwC notes that using GL.ai has increased efficiency and effectiveness, completing the analysis in less time than it would take a human auditor, while also providing more significant insights. This allows auditors to focus on areas with the most significant risk.
