Can we trust Artificial Intelligence (AI) to audit financial statements?
Artificial intelligence advocates speak of a time to come when these systems will be capable of auditing 100% of a company’s financial transactions. These visionaries foresee the day when AI will enable auditing that is a continuous and real-time process, not a prolonged exercise requiring large teams of accountants working overtime after the close of a fiscal year.
But is AI in auditing a good idea? Or do we even have a choice — is it just part of the data-focused technology wave that all companies must embrace?
We’ve approached our development of AI in auditing from the ground up to ensure that human values remain at the core of our audit work and that auditors have the tools they need to continue to improve audit quality. Here’s what we’ve learned:
—The details matter. Data acquisition is at the heart of auditing. Auditors need to obtain raw business data before they can “audit” it – check the accuracy and alignment of data sets like purchase orders, billing, receivables, payments, expenses, and compensation. Further, auditors regularly consider external data sources to understand risks, plan the audit, and confirm company assertions. To incorporate AI into their audit methodology, auditors need to understand systematically how those data sets are structured; how they differ from one industry, client, or source system; and how to transform the data reliably for use in our solutions.
While virtually all business records today are kept in one electronic data format or another, some data is more easily digestible by software programs than other data. For example, many insurance companies keep their policies in PDF-format files, while information on claims made against those policies is stored in text-based document files. Before those data sets can be reconciled, there needs to be an interface to interpret and align relevant data points across that file format divide. Until now, that interface has been human and manual in nature. To train AI to perform these tasks, we need to supply it with not only with the right data, but also the right decision-making capabilities based on that data.
—AI has something on us. Compared to humans, machines excel at performing such repetitive and time-consuming tasks as data acquisition. Machines and AI-enabled technology will streamline data acquisition challenges faced by auditors. AI will minimize the burdens of once time-consuming tasks of seeking out relevant information, pulling it out of documents, and converting it into usable formats. That will leave humans to review, analyze, and audit.
The army of independent auditors needed to audit a typical Fortune 500 company can be streamlined, and the auditor can spend more time on the judgmental aspects of the audit. Machines excel at processing vast amounts of data efficiently. They’re capable of reviewing massive quantities of data at scale, evaluating what needs to be checked in an audit, and then recognizing anomalies in the data. AI-enabled solutions can quickly and easily identify such things as an unusual spike in orders from a particular geography, an exceptional set of expense items recorded by an individual, or unusually favorable terms contained in equipment leases recorded for a specific supplier .
—But we have something on AI. Clients retain auditors for the assurance they provide over the financials. And that can only come through thoughtful examination and the exercise of judgement — human judgment. AI systems can assist the auditors by acquiring, processing, and churning through the mountains of data that a business’s financial reporting systems generate. But while the machines may more quickly and completely identify patterns and anomalies in massive data sets, more value comes from investigating and deducing the reasons behind the pattern or the anomaly. Only human beings, such as the auditor, can tell the true story behind the data.
As we introduce new technologies, we also have a responsibility to ensure that they’re ready for prime time. AI is going to do what we tell it to do — nothing more, nothing less— and we must remain clear-eyed about the risks. By clearly defining the audit requirements and fostering collaboration between data scientists, developers, and auditors to meet them, we can move the technology beyond the slogans and into practice successfully and responsibly.
—Faster does mean better. Faster can indeed mean better when the processing time of data is greatly reduced. As AI in auditing makes it possible to move toward auditing 100% of data, rather than samples of it, auditors will be empowered to study the totality of a business in an efficient manner. AI can help auditors move from traditional audit- sampling frameworks to visualizations and evaluations of the full picture.
—AI can often help. AI can help in most instances where manually intensive activities occur, and that represents a significant transformation in traditional audits. Data extraction, comparison, and validation are great starting points. AI can significantly speed up digitization of data entry and extraction activities being performed manually, reducing the time spent on audit data preparation. Tying internal payment data to third-party support requires a significant amount of audit and client hours. Testing the existence and valuation of payment transactions can be fully automated with AI, as well as the extraction of support for any substantive testing required.
—Clients have the most to gain. At the most basic level, process efficiency means clients will need to devote less time and resources on responding to queries and requests for documentation, giving them back more time during a critical, deadline-driven time period.
More critically, when external auditors have more time to spend on higher level analysis, they can focus on areas that require increased judgment and contain a high level of estimation uncertainty.
Bill Brennan is a managing partner for audit transformation at PwC; Michael Baccala is U.S. assurance innovation leader at the firm; and Mike Flynn is a principal for advanced risk and compliance analytics solutions, also at PwC.