Artificial intelligence (AI) and machine learning have become industry buzzwords. This article discusses the state of AI and machine learning as they relate to accounting, and the challenges the profession may face moving forward as these technologies start to spread.
AI and Machine Learning
AI, sometimes referred to as cognitive computing, is currently considered a broad category of technologies that can imitate or simulate human behavior. Some AI aim at mimicking human behavior, some aim at surpassing human performance. In a recent article, InfoWorld contributor Kevin Gidney, cofounder of Seal Software, refers to AI as a broad description of any device that mimics human or intellectual functions, such as mechanical movement, reasoning, or problem-solving.1 He defines machine learning as a subset of AI consisting of a statistical and data-driven approach to creating AI, such as when a computer program learns from data to improve its performance. Information technology research firm Gartner defines AI as systems that change behaviors without being explicitly programmed, based on data collected, usage analysis, and other observations.2 Regarding machine learning, Gartner views it as a technical discipline to solve business problems using mathematical models that can extract knowledge from data (as opposed to traditional software engineering, which aims to solve business problems by explicitly defining the software logic). For Gartner, AI employs machine learning, deep neural networks (a variant of machine learning),3 and other technologies to analyze huge amounts of data beyond simple algorithms to achieve new levels of performance and insight. Examples of other technologies employed by AI include natural language processing (ability of a system to understand written or spoken human speech), planning systems (ability to find optimal path), and agents and bots (ability to perform complex, repetitive tasks involving interaction among multiple data sources or systems).
In the context of accounting, AI and machine learning technologies – embedded in existing applications or combined with other technologies such as robotic process automation (RPA)4 – are expected to automate a significant part of the mundane tasks performed by CPAs today, such as document and data collection from clients and third parties; document recognition and classification; data extraction from documents and entry into accounting, auditing, tax, or other systems; approvals (such as invoice or expense approvals), confirmations, and reconciliations; computations and analysis (from trend analysis to predictive analytics); and answering inquiries.
The current fascination around AI and machine learning is related to the explosion of digital information combined with significant improvements in technological areas, such as mobile computing, internet of things, big data analytics, speech recognition, and image recognition. Add to that the significant improvements in availability and affordability of computing power and web-enabled infrastructure and storage. Popular AI and machine learning platforms for software developers include Amazon AI, Machine Learning on AWS, Microsoft Azure Machine Learning Studio, and IBM Watson. Other big names making AI and machine learning a part of their core strategy include Google, Oracle, SAP, and Salesforce, to name only a few.
Consumers are already using AI and machine learning technologies in their day-to-day lives via their smartphone (including apps such as Siri, Cortana, Pandora, Waze, and Uber), virtual personal assistant speakers (including Amazon Echo and Google Home), and social networks (such as LinkedIn and Facebook, which use machine learning to increase the relevance of their feeds or ads). AI and machine learning technologies are rapidly making their way into the business-to-business world too, with early adoption reported in many sectors, including financial services, health care, manufacturing, retail, supply chain, and government.
Adoption of these advanced technologies within business applications offers new levels of process automation as mentioned above (such as capturing and classifying documents, extracting relevant data, etc.). Other examples of usage exist in auditing and tax. For example, Deloitte has partnered with Kira to automate document review, including extracting relevant terms from contracts, leases, employment agreements, invoices, and other legal documents.5
Incorporation of AI and machine learning into the accounting profession is not limited to Big Four firms. At its latest annual business summit in Philadelphia, EisnerAmper LLP discussed its progress in implementing IBM Watson’s capabilities in auditing and tax, primarily to automate tasks related to document review and data analysis.6 The firm also presented a video showing an AI-enabled RPA solution it recently implemented as an illustration of advanced business process automation in action, from the digital capture of an accounts payable invoice to its processing.
Some challenges inherent in the accounting implementation of AI and machine learning include the varying degrees of maturity of these applications, data normalization and quality, a lack of standards, a lack of skills among employees, security and privacy concerns, a lack of transparency (black box systems provide limited transparency on the system’s reasoning, assumptions, and computations), the danger of overreliance on AI systems, and the potential risks related to human bias.
CPA firms are using AI, machine learning, and RPA to increase efficiency and quality. By using these technologies, CPAs are reallocating their time to performing high-value, high-impact tasks. These technologies are also resulting in more complete and accurate audits, which in turn increase stakeholders’ confidence when reviewing audit reports.
In addition to the examples above, there are several other examples of how Big Four firms are implementing AI and machine learning. According to a recent Forbes article, Deloitte is using natural language processing to review hundreds of thousands of legal documents to identify change control provisions as part of a client’s sale of a business unit. The article states that this process used to keep dozens of employees occupied for half a year, but with the implementation of the natural language processing system, the number of employees on the task was trimmed down to eight and the time the company spends on the task was down to less than a month. The same article describes how EY is using natural language processing to review leases to ensure that they comply with the new lease accounting standards. Prior to the implementation of its system, EY had to manually review each lease – a process prone to error and inefficiency. The natural language processing system is said to be three times more consistent and twice as efficient as the manual process. The Forbes article also relates how EY is using machine learning to detect anomalies and fraudulent invoices. It indicates that the technology is 97 percent accurate at identifying faulty invoices, and has enabled the organization to minimize its risk exposure when it comes to violating sanctions, anti-bribery regulations, and other aspects of the Foreign Corrupt Practices Act.7
Assurance departments are looking to AI to significantly increase the efficiency and quality of audits by enabling them to focus on high-risk areas and reduce manual tasks. Instead of performing sample-based testing on a random selection of transactions, AI can be used in the planning stages to analyze the entire general ledger and identify transactions considered high risk or “out of the ordinary.” By doing this up front, auditors can focus their attention on the transactions that represent the most risk and plan the audit accordingly. In this situation, AI is allowing assurance professionals to perform better audits, which in turn provides more value to their clients and more confidence to stakeholders and investors who are relying on audit reports.
Surfing on the wave of popularity around AI, RPA solutions are gaining popularity too, especially when it comes to internal control testing. Internal audit departments are looking to RPA to assist with manual testing. For instance, when performing control testing, internal auditors typically have to log in to several systems to obtain the appropriate evidence for testing. Obtaining the evidence alone for each sample selected is cumbersome, and could take as long as an hour for each sample being tested. RPA solutions could obtain the evidence for the auditor and, depending on the type of test, may even be able to complete the testing. By using RPA, internal auditors will be able to focus their time on high-value tasks, such as analyzing the results of testing to determine if significant control deficiencies exist.
On the corporate side, accounting departments are looking to RPA to assist with manual and repeatable processes. For example, accounts payable invoice processing is typically a manual process that requires input from multiple departments. Thus, payments on invoices are often delayed and, as a result, business can be disrupted and opportunities for discounts on payments are missed, resulting in higher costs for the organization. In addition, business-critical vendor relationships could be impaired due to inefficient payment processes. RPA can be used by accounts payable departments to automate a significant portion of the invoice processing workflow. An RPA solution can be used to automatically read invoices, import specific data from invoices into the enterprise resource planning (ERP) system, perform a three-way match, route the invoice to the appropriate personnel for approval, and submit the payment to the vendor. In addition, an RPA solution can identify exceptions and notify the appropriate personnel for manual review.
Even though AI and machine learning can provide significant value, CPAs must be aware of the risks these technologies bring to the profession.
One important risk that CPAs should understand is algorithmic bias.8 If the data used to train an AI system or the algorithm used in AI is inherently biased due to poor input of data or poor algorithm design, then the AI technology will potentially output bad results, causing wrong decisions to be made.
Algorithmic bias could have significant implications on a financial audit. For instance, if auditors are relying on AI technology to identify high-risk transactions and patterns of fraud as part of the audit, but the data used to train the algorithm did not include appropriate historical data to enable the technology to appropriately identify high-risk transactions and patterns of fraud, then there is a significant risk that the auditor may rely on poor results from the system. If the auditor is relying on this data to plan testing procedures, then there is a risk that the auditor will design inadequate procedures based on the results of the AI technology.
In addition, CPAs (in practice or industry) should consider security and change management risks associated with AI technology. AI systems may have access to vast amounts of sensitive data, so CPAs must ensure appropriate controls are in place to mitigate the risks of unauthorized access and unauthorized disclosure of sensitive data accessed by the technology. Furthermore, CPAs should ensure that organizations employ change management controls that mitigate the risk of unauthorized, incorrect, and inadvertent changes being made to the technology. If these types of changes are occurring, then auditors face the risk of making wrong decisions based on poor results.
As always, CPAs should consider the relevant regulations. For example, CPAs should be aware of potential risks of violating privacy regulations, such as the European Union’s General Data Protection Regulation, by storing new types of data. As the use of AI becomes mainstream, CPAs will have to pay close attention as rules and guidelines emerge. The AICPA, the Securities and Exchange Commission, and the Public Company Accounting Oversight Board will likely develop rules and guidelines around the use of AI when performing attestation engagements.
One of the hallmarks of the CPA profession is the requirement for CPAs to adhere to a high level of ethical standards. As adoption of AI and machine learning grows, it is important to contemplate the ethical implications of this technology for the accounting profession. AI and machine learning systems are not inherently ethical or unethical; however, unscrupulous programmers could design AI and machine learning models that lead to negative outcomes. For example, imagine an AI algorithm that is biased against individuals or groups that belong to a certain racial, religious, or socioeconomic category. This could lead to unintentional discrimination by underwriters responsible for approving loans or in determining insurance premiums. As a result, firms using this technology and their auditors could be exposed to potential lawsuits.
Thousands of complex rules may be associated with one AI model. Therefore, transparency of the programming becomes paramount to help avoid algorithmic bias. As opposed to a “black box” approach, the rules engine should be made visible to authorized users of the systems by providing the ability to explain the reasoning behind the results or recommendations. AI and machine learning software firms, particularly those that participate in the accounting realm, should work closely with CPAs (who are bound by a code of conduct) to ensure that systems behave ethically in the performance of services. This undoubtedly would help mitigate the risk of unethical or biased coding. Significant opportunities will emerge for CPAs to review AI and machine learning systems and the internal controls around these systems to help ensure applications comply with ethical standards, that the coding is unbiased, and that the rules engines are transparent to authorized users. Achieving adequate internal controls and transparency will require a coordinated effort among software developers, accounting firms, and professional organizations such as the AICPA, the Institute of Management Accountants, the Institute of Internal Auditors, and ISACA.
Where existing laws and regulations do not adequately address the risks and ethical challenges associated with AI and machine learning, regulators will likely respond by either creating new rules or modifying existing ones. They will have to respond quickly, but also be cautious about tailoring legislation or regulations too narrowly. As Channa Wijesinghe, CEO of the Accounting Professional & Ethical Standards Board in Australia, noted, “It is going to be a challenging task for regulators and lawmakers to stay on top of AI due to its rapid pace of development. When legislation is written with specific AI systems in mind, it may become obsolete by the time the legislation is approved.”9 Developing legislation and regulations that address specific AI systems today must be flexible enough to encompass the inevitable advances that will take place.
How will the profession ensure a high ethical standard as AI and machine learning technologies become more prevalent? Many experts call for the development of a framework that specifically addresses the potential ethical challenges of AI and machine learning adoption. Professional accounting organizations, such as the AICPA, should work cooperatively with federal and state regulators to develop guidance to ensure that AI and machine learning tools are used appropriately and fairly in accounting and auditing.
AI and machine learning systems have the potential to dramatically improve decision-making and the efficiency of business operations. The ability of these systems to process large volumes of data and produce results with increased accuracy and consistency when compared with humans will have a significant impact on the accounting profession going forward.
Although adoption of AI and machine learning systems is still in the early stages, it is important for CPAs to understand their potential and limitations, including the ethical challenges. AI and machine learning will enable CPAs to spend less time on data preparation and analysis and more time on interpreting results and developing insights. CPAs should develop expertise in AI and machine learning tools as a strategy to add value to their organizations and clients.
1 Kevin Gidney, “Demystifying Machine Learning,” InfoWorld (April 11, 2018).
2 “Machine Learning: FAQ from Clients,” Gartner research note (July 31, 2017).
3 Sometimes referred to as deep learning, a type of machine learning able to identify complex patterns out of large numbers of hidden data layers.
4 RPA is an advanced form of business process automation software.
5 Julia Kokina and Thomas H. Davenport, “The Emergence of Artificial Intelligence: How Automation Is Changing Auditing,” Journal of Emerging Technologies in Accounting, vol. 14, no. 1 (spring 2017).
6 8th Annual EisnerAmper Philadelphia Business Summit, Philadelphia (Nov. 8, 2018).
7 Adelyn Zhou, “EY, Deloitte and PwC Embrace Artificial Intelligence for Tax and Accounting,” Forbes (Nov. 14, 2017).
8 As defined by research firm Gartner, an algorithmic bias occurs when an algorithm reflects the implicit bias of the individuals who wrote it or the data that trained it. Source: “Top Strategic IoT Trends and Technologies Through 2023,” Gartner research note (Sept. 21, 2018).
9 Machines Can Learn, but What Will We Teach Them? Ethical Considerations around Artificial Intelligence and Machine Learning, Chartered Accountants Australia and New Zealand.
J.L. “John” Alarcon, CPA, CGMA, CITP, is chief financial officer for LoanLogics in Trevose and a member of the
Pennsylvania CPA Journal Editorial Board. He can be reached at email@example.com.
Troy Fine, CPA, CITP, CISA, is manager, risk advisory services, for Schneider Downs & Co. Inc. in Pittsburgh and a member of the Pennsylvania CPA Journal Editorial Board. He can be reached at firstname.lastname@example.org.
Cory Ng, CPA, DBA, CGMA, is an assistant professor of instruction in accounting at the Fox School of Business at Temple University in Philadelphia and a member of the Pennsylvania CPA Journal Editorial Board. He can be reached at email@example.com.