Features | From Pivot Magazine

When the machines take over

If AI makes audit virtually foolproof, what’s left for auditors? Plenty, it turns out.

A Facebook IconFacebook A Twitter IconTwitter A Linkedin IconLinkedin An Email IconEmail

Solon Angel MindBrdige founder and Eli Fathi, MindBridge CEOSolon Angel, MindBridge founder (left) and Eli Fathi, MindBridge CEO (John Londono)

Solon Angel decided the accounting industry was in trouble in 2015, at an American Accounting Association conference. He was listening to a speaker—“Very smart; I’m not going to share his name,” Angel says—who at one point told the audience that accountants should learn SQL, the (notoriously complicated) data-management computer language.

“I lost it,” says Angel, who has black curly hair and a habit of hammering out his syllables with his fingers as he speaks them. “I literally started screaming. Never. This is stupid. Accountants go to school, spend a lot of time getting their CPA. They should never become data scientists, scripters or SQL experts.”

To be fair to the unnamed source of Angel’s ire, he was trying to address a serious challenge for the accounting industry. The audit landscape is changing. Investors are increasingly relying on unaudited information to assess companies, and some high-profile business failures, such as Carillion in the U.K., have prompted many to wonder how auditors missed the underlying problems. A key issue remains the “expectation gap” between what audits are really meant to do, and what the politicians and public often think they should do—i.e., guarantee that every transaction is examined and every potential problem is caught. “If the profession is not meeting the demands of society, then that is our problem, not theirs, because without the trust and confidence of wider society, we cannot deliver,” Michael Izza, chief executive of the Institute of Chartered Accountants in England and Wales, told this magazine in 2018.

The reality is that auditors assess high-risk areas. They aren’t required to verify every transaction and flush out every oversight or fraud within the ledgers of the companies they investigate. The sheer volume and complexity of data required to run a modern company is structurally overwhelming. That’s why the unnamed speaker told his audience to learn code. Angel thinks he has a better idea: harness artificial intelligence to perform deep, expansive dives into the data-laden innards of companies large and small.

“Not all fraud can be found running data through algorithms”

That idea came to the 36-year-old Ottawa-based entrepreneur out of frustration and a desire to bring order to chaos. Angel was born in Brazil, where he had a chaotic childhood, to put it mildly. His parents had a messy divorce, and he spent part of his childhood in the French West Indies; his family was isolated for six months in Guadeloupe during Hurricane Hugo in 1989. “It was every man for himself,” he says. “There was no rule of law.”

Angel spent his early years in tech in Silicon Valley, and later moved to Ottawa around the time Bernie Madoff went from fêted investor to the face of financial chicanery. Angel remembers marvelling at how, despite the supposedly advanced auditing tools available to the industry, Madoff was able to fool generations of investors and auditors, not to mention the U.S. government, mostly by the force of his own personality.

“The Madoff case really resonated with me because the SEC and [the big firms] don’t hire unqualified people. They hire very good people, and yet they weren’t able to catch that fraud scheme,” Angel says. “Risk management as a whole is completely outdated. In a dynamic, fast-moving world, controls need to evolve dynamically and be adaptive, and large organizations don’t have that capability.”

The end product of Solon’s frustration is MindBridge Ai. The Ottawa-based company produces Ai Auditor, which uses artificial intelligence to do what is often physically impossible for flesh-and-blood auditors: review and analyze 100 per cent of a company’s transactions. The company says Ai Auditor greatly enhances the ability to detect risk, fraud and mistakes—and that it might just revolutionize the industry.

“AI will not replace auditors. Auditors that use AI will replace auditors that don’t”

Others think so, too. MindBridge, which has more than 70 employees, has raised $13 million in total funding from venture capital firms including New York-based Reciprocal Ventures, Toronto’s The Group Investments and San Francisco-based 8VC, along with the National Bank of Canada. It has 200 customers in 10 countries.

MindBridge is far from the only player. Just a short drive away in Ottawa, another company called AuditMap.ai applies machine learning to internal audit, and the Big Four are all investing in their own tools. Such automation will likely cause casualties in the industry—but perhaps not the kind most imagine. “AI will not replace auditors,” says John Colthart, MindBridge’s general manager of audit and assurance. “Auditors that use AI will replace auditors that don’t.”

Modern auditing was born during the Industrial Revolution. From the outset, audits were meant to increase transparency and boost public confidence in the manufacturing behemoths dominating the economy, as well as the banks servicing them. Of course, auditing is far more difficult today than it was 175 years ago. “Companies increasingly operate across borders; rules and regulations continue to proliferate; and capital markets, products and transactions evolve at a much faster pace than accounting and audit standards,” reads a 2014 Forbes Insights report. 

Eli Fathi uses the analogy of a lake. An electrical engineer by training and serial entrepreneur, Fathi joined MindBridge as CEO at its inception in 2015. Because today’s companies are increasingly large and complex—deep, wide lakes—auditors limit themselves to little pieces of them to determine their financial health, he says. “It’s like going to the corner of a lake and the fish aren’t biting, and from that you say there are no fish in the lake,” says Fathi, a slender and direct man who shares Angel’s obsession for detail.

MindBridge’s solution is to virtually test the whole lake. From its decidedly nondescript offices in the outskirts of Ottawa, the company hoovers up customer-provided data, including general ledger transactions, accounts payable, accounts receivable and income statements. Ai Auditor, which launched in February 2017, analyzes the data using a hybrid of AI techniques to identify anomalous patterns of activity that warrant further investigation. It learns by precedent what constitutes a regular transaction, and then picks up and flags irregularities into “low-,” “medium-” and “high-” risk baskets (irregularities can be as benign as a mislabelled expense, or as serious as diversions of capital). The software thrives on data, becoming better at its job as it gobbles up more of the stuff. MindBridge now has upwards of 600 million data points, which sit in cloud servers somewhere in the ether. There isn’t so much as a server room in the office. 

“AuditMap.ai uses machine learning to automate, improve and speed up internal audit process”

“Just following the standards and using the tools of the moment is no longer good enough,” says Fathi. “The accounting industry is reaching a point of inflection where senior people who know how to find these problems are retiring, the tools and standards are all outdated and it’s time for a forklift replacement. We need to update standards, retool the team and retrain them.”

Ai Auditor has already had success rooting out errors and fraud. Last fall, Gary Krausz, an American CPA who heads the accounting and auditing department at a firm in Los Angeles, was able to uncover evidence of fraud to the tune of US$2.8 million. Krausz’s company was hired to audit a consumer-products manufacturing company that suspected one of its employees. (Krausz wouldn’t divulge the company name, citing privacy concerns.) The company, which had about $160 million in sales, needed to analyze 18 quarters of its general ledger—over 6.2 million transactions. 

Krausz uploaded the mountain of data to Ai Auditor and performed an audit. It found the bad transactions, calculated their value and pinpointed the specific user perpetuating them. The employee in question had posted large amounts in irregular accounts on multiple occasions. The evidence provided was used in the prosecution of the case. 

“We fed the client’s general ledger data into the software without any filter, and the ‘bad transactions’ popped right out in our sample of high-risk transactions. It’s as though the bad transactions were meant to be found,” says Krausz. “Doing this audit on our own would have taken at least five times longer.”

MindBridge isn’t the only business bringing AI to audit. In December, Lemay.ai, another Ottawa start-up, announced a spinoff company called AuditMap.ai, which uses machine learning to automate, improve and speed up the internal audit process. “The internal audit function of many organizations—especially at public companies and governments—is gigantic, mandatory and not very effective,” writes co-founder Daniel Shapiro. “Applying deep learning accelerates audit functions by resolving routine legwork bogging down the auditors’ ability to assess and report with no compromise on quality.”

Meanwhile, the Toronto firm CaseWare, which develops software for accountants and auditors, harnesses artificial intelligence across its audit analysis platform. The AI tool performs a deep dive into transactional and industry data to provide insights that auditors can use.

PwC has spent “hundreds of millions” on AI and automated technology over the last several years.

The Big Four firms are also embracing automation. In 2015, Deloitte won the International Accounting Bulletin’s Audit Innovation of the Year award for technology including Argus, its cognitive audit application, which uses machine learning and natural language processing to extract accounting information from virtually any document. In late 2018, the firm also announced plans to triple the Canadian staff count of its AI practice, Omnia, to about 1,000 over the next two years. EY, too, provides AI consulting services and, on its website, reports, “We are embedding emerging technologies such as Artificial Intelligence technology across our end-to-end audit process.” KPMG, meanwhile, has partnered with the team behind IBM’s Watson supercomputer to develop Clara, a “smart audit platform” that incorporates AI, data analytics and other technologies. 

PwC, for its part, has spent “hundreds of millions” on AI and automated technology over the last several years. In 2017, the London-based accounting behemoth partnered with H20.AI, a Silicon Valley developer of open-source artificial intelligence systems, to build a “bot” capable of analyzing billions of data points in order to detect aberrations within general ledger data. The bot, known as GL.ai, is the initial module of Audit.ai, PwC’s suite of AI-based products. 

Yet much of PwC’s investment in machine learning has actually been spent on its flesh-and-blood employees. “A lot of the cost is actually in people,” says Michael Paterson, national assurance leader for PwC Canada. “It’s the broad upscaling of all our people in the next few years around automation and being able to work with data. For AI to work, you need data and people who can work with and understand its results.”

In other words, PwC isn’t hiring fewer people because of AI. Rather, it still hires the same number of accounting students—if not more—and the company is employing people with different skill sets than it did even 10 years ago. This reflects a broader truth when it comes to AI: by performing the often-rote tasks like data acquisition, formatting and inputting data, automation allows humans to instead give “value added” services to clients. Paterson says this will make auditors’ work more interesting. “The need for insights and analysis and trust around information is actually increasing,” he says. “Assurance jobs are going to continue to exist, but they’re going to be more important. Being a CPA, working in assurance, understanding a client’s business, talking to people and adding insights—that’s the stuff that’s challenging and valuable.”

Angel likens Ai Auditor’s progress to that of self-driving cars.

Currently, artificial intelligence and machine learning tend to shine in the “front end” of an audit, the labour-intensive bit of the practice in which incoming data from various enterprise resource-planning software platforms are harvested, shaped and made uniform. The “back end”—interpretation and analysis of the results—remains the purview of humans. 

Angel likens Ai Auditor’s progress to that of self-driving cars. Their autonomy is measured in levels, from one (driver braking assistance, lane departure) to five (full automation). “MindBridge is at level four, where the driver can still override the machinery but the car is driving itself,” he says. “Similarly, the auditor is always in control of the auditing process. They can always override the process.”

It begs the question: what about level five, where humans aren’t necessary at all? In 2016, World Economic Forum founder Klaus Schwab published “The Fourth Industrial Revolution,” a 192-page future-casting compendium of data-based predictions of how technology will shape the world in the coming decades. Among its predictions: a high likelihood that 30 per cent of all audits will be performed by AI by 2025—a number that will only increase as artificial intelligence is adopted in the industry. “An environment can be envisioned in the future where AI replaces a range of functions performed today by people,” he writes.

Like Schwab, the MindBridge executives are already envisioning a time when the company can issue audited statements on behalf of its clients—a prospect that would arguably take auditors out of the equation altogether. Others, however, doubt a fully automated audit will ever come to fruition.

“One of the challenges with AI as a regular is that you no longer have the same visibility to how the technology works”

CPA Carol Paradine, CEO of the Canadian Public Accountability Board (CPAB), says that while AI has incredible potential to improve audits, not everything can be automated. Human auditors can apply professional judgment and examine biases in ways AI can’t, she adds, pointing to a fraud she uncovered early in her own career. She noticed something was fishy when she visited a company she was auditing and found significantly fewer employees than she expected. “I found it because I was physically out at the site,” she says. “Artificial intelligence will be able to verify whether certain calculations are correct, find outliers and look for unusual contracts. But it probably won’t be able to verify that an employee actually works for that organization—or even exists. Not all frauds or unusual transactions can be identified by running data through algorithms.”

Artificial intelligence poses regulatory hurdles, too. “One of the challenges with AI as a regulator is that you no longer have the same visibility to how the technology works,” she says. “You’re inherently putting more reliance on what I call the ‘black box’ ”—that is, a self-taught AI platform that coders, regulators and even auditors themselves may not fully understand. “As firms implement artificial intelligence, they will likely have a number of questions around how to demonstrate that enough work has been performed,” she says. “They may look for guidance from the standard-setters.” If and when those standards are being developed, she says, CPAB will provide input.

“In some cases, AI will help auditors do what they need to do faster—such as reading lease contracts, identifying lease terms, analyzing large data sets to identify anomalies or uncover subtle relationships that a human may have not been able to identify themselves,” says Taryn Abate, director of audit and assurance at CPA Canada, which recently released a foundational resource called “A CPA’s Introduction to AI: From algorithms to deep learning, what you need to know,” in collaboration with the American Institute of Certified Public Accountants. CPA Canada will be following up with papers on big data and AI, and on audit and AI. “It is my belief,” says Abate, “that AI will augment the role of auditors, rather than replace them.”

MindBridge’s Colthart agrees. “My first thought is: if auditors do their job right and adopt technology, they will not go away,” he says. “My other thought is: if they don’t, they will be removed altogether.”