With the application of RPA, software automatically performs the pre-designed RPA-based audit tasks, giving auditors more time to perform higher-level tasks which require professional judgment, such as evaluating contradictory evidence and designing and applying follow-up actions. Additionally, the scale of audit procedures performed by RPA software is no longer limited by human processing power. Currently, because of limited audit resources, sampling techniques (statistical and non-statistical) are commonly used in processes such as tests of controls and substantive tests of details (Christensen et al., 2014). With the adoption of RPA, auditors can, in certain instances, expand the scale of some procedures from sampling to testing the entire population, avoiding the risks and deficiencies related to sampling and collecting comprehensive audit evidence.

Another benefit of RPA is to minimize or avoid human errors such as mistakes when confirming amounts, errors in workpapers, and ignoring red flags. Once the RPA software is programmed, the tasks and analyses are performed in line with audit standards and pre-defined rules. A recent survey of major accounting firms shows that, compared with the 90% accuracy rate of humans, RPA accuracy achieves 99.9% (Cooper et al., 2019). For instance, one common human error for the confirmation process is that a request can mistakenly not be submitted, and a week or more may pass before auditors realize it never went through. With the RPA-based audits, the program never misses a request in the prepared form.

Finally, RPA may be a way to improve audit quality by reducing outsourcing and offshoring, which take up 10–20% of audit hours (Daugherty et al., 2012). Both regulators and researchers have expressed concern about the risk associated with outsourcing. The PCAOB (2012) has raised the discussion of whether certain offshoring arrangements or structures increase audit risk. Prior literature finds evidence that offshored and outsourced audit work comes with higher risk and lower quality. With the adoption of automation, practitioners suggest that the offshore and outsourced work could be largely replaced by RPA programs, which are better controlled and monitored by US firms (KPMG, 2016).

During the evaluation of the pilot project, two exceptions were encountered. First, one bank was no longer included in the network, so the RPA program was not able to send an electronic confirmation request. The second exception was that the information about the authorized signer was incorrect, which is not uncommon for the confirmation process because clients sometimes forget to update the information to auditors. Broadly, RPA can start leaving exceptions off and progressively incorporate the most common ones. Then, two solutions to deal with errors or exception are discussed.

The first solution is to automate the normal and routine process and leave all errors and exceptions to human auditors. For example, during the confirmation process, it is possible to receive feedback that the request cannot be completed because the bank no longer exists in the network. In this case, the RPA software only identifies the uncompleted request and leaves the follow-up for auditors. Specifically, the type of error (such as request denied, more information needed) and the comments (such as invalid date or invalid contactor) are sent to auditors, and all the follow-up is performed manually.

However, if the errors or exceptions happen regularly, leaving them for the manual process may not be efficient. The solution is to first classify errors/exceptions into two groups, common and uncommon errors, and include follow-up procedures for the common errors in the RPA automation. Once this type of error is detected, the follow-up process will be triggered automatically, and auditors need to deal only with uncommon errors, enhancing the efficiency of the RPA application.

Although RPA can potentially make significant improvements to auditing practice, a key limitation currently is that the software is able to perform only routine tasks and to make decisions based on explicit rules. Therefore, current RPA software is not adaptable to audit procedures requiring professional judgment that cannot be transformed into structured instructions.

Recently, there has been progress in the evolution of technology for applying artificial intelligence (AI) in the industry, with large accounting firms launching numerous projects (Kokina and Davenport, 2017). For instance, KPMG is working with IBM Watson to apply cognitive computing technology to its professional services offerings (IBM, 2016). Deloitte is collaborating with Kira Systems, a contract analysis system, to create cognitive models that examine large numbers of complex documents, extract textual information for better analysis, and assist auditors with the difficult task of document reviewing (Deloitte, 2016). To take advantage  of AI developments and address the limitations of current RPA, practitioners have proposed Intelligent Process Automation (IPA), which refers to combining AI, cognitive automation, deep learning, and machine learning with RPA (e.g., Berruti et al., 2017; UiPath, 2018).

Instead of only mimicking the way people perform routine business processes, IPA leverages the advantages of AI to learn how people make decisions and may be able to perform complex tasks faster and better. To further improve audit quality, accounting firms may eventually consider applying IPA to help auditors perform complex and unstructured audit procedures and make professional judgments.
