I robot, CPA

In this age of light-speed tech change, how far are we from the day when accounting software or an automated system will perform all the work accountants do today?

March 19, 2016: in a spare room in Seoul, Lee Sedol, world champion of Go, an abstract board game invented in China 4,000 years ago, squares off against AlphaGo, a Go-playing machine designed by Google and powered by an artificial intelligence software, distributed across hundreds of servers, called DeepMind.

An Atlantic Monthly article that covered the event states that Sedol is to Go what Tiger Woods and Michael Jordan are to golf and basketball respectively: a rare virtuoso who defines an era, who sets the pace for everyone else. In a series of five games against DeepMind, Sedol lost four.

This is a much more significant landmark than the famous defeat that chess grandmaster Garry Kasparov suffered in 1997 at the hands of IBM’s Deep Blue. Go is a much more complex game. “Unlike with chess,” explains a Technology Review article, “there aren’t straightforward guidelines for playing the game or measuring progress, which is one reason why Go has historically been so difficult for computers to play.”

But DeepMind exhibits two key features. First, it belongs to the discipline of neural networks, a branch of computer science launched in the 1950s that tries to mimic — through complex organizations of transistors — the way the human brain works. Secondly, and above all else, DeepMind belongs to the artificial intelligence area of deep learning. As neural networks are wont to do, it is constantly learning from its own experience and improving, which can lead it into territory usually reserved for humans: creativity. Following a surprising move by DeepMind in one of the games that Sedol lost, European Go champion Fan Hui commented: “I’ve never seen a human play this move. So beautiful.”


Software programs as capable as DeepMind have not yet reached accounting. Far from it. “Audit firms are stuck in a pre-Moneyball era,” says Ramy Elitzur, CPA, CMA, professor in financial analysis at the University of Toronto’s Rotman School of Management, who teaches partners of accounting firms in his executive courses.

The term “Moneyball” refers to professional US baseball teams that, as the movie Moneyball illustrated, use combinations of big data and data analytics to improve game management and player selection. It is still not DeepMind-type wizardry, but it is certainly more sophisticated than the standard Excel routines many accountants still confine themselves to.

But Elitzur’s pronouncement is too harsh. Big accounting firms have progressed toward the leading edge of accounting automation. For example, a Montreal-based team at PricewaterhouseCoopers (PwC) has developed and installed a software system at some client sites that combines data analytics and machine learning to optimize the procurement function of a company.

The system scours millions of transactions, groups them into an industry-specific taxonomy and automatically identifies spend optimization opportunities, says Ramy Sedra, data and analytics consulting leader at PwC Canada. It calculates how much the company could have saved by taking full advantage of the volume discounts and even what advantages could be gained by renegotiating the agreements.

The software is not just a glorified calculating machine. Its algorithms study matches between purchases and their categories and, as those matches are corrected and improved, machine-learning routines prompt the system to self-correct and self-optimize.

Deep learning can do very little by itself. It needs to feed on vast stores of data and information to show off its abilities. And that’s where big data and data analytics step in as prerequisites.


Up to now, accountants “have typically analyzed only information that they have themselves created,” says Robert Parker, an IT veteran who is a member of the board of directors of the University of Waterloo’s Centre for Information, Integrity and Systems Assurance.

That is changing, and it is probably as significant a change as the advent of deep learning. Now, information is flooding in from countless sources, ushering in the age of big data with its characteristic five v’s — volume, variety, velocity, veracity and value — says Theo Stratopoulos, associate professor of information systems at the University of Waterloo’s School of Accounting and Finance.

The first three v’s — volume, variety and velocity — comprise online transactions, Facebook comments, webzine articles and the eventual flood of data from the Internet of Things.

When accountants essentially manipulated data they had themselves created, they performed analyses and audits along the two continua of guided and structured enquiry. Now, we are entering the realm of two new continua: unguided queries and unstructured data. An accountant who expresses an ongoing concern opinion on the basis of financial statements, past performance and comparisons with competitors’ statements is performing a classic guided analysis using structured data, says Stratopoulos.

But the accountant of tomorrow will be called upon to perform text and sentiment (opinion) analysis as well, mining the information in magazine articles and Facebook feedback about competitors and their products. They will lead a guided analysis using unstructured data. And if they carry out an audit by simply letting a deep learning routine come up with cluster analyses of procurement and payment accounts, website analysis and Facebook pages that show potential fraudulent patterns, they will be leading an unguided analysis using unstructured data.


These new frontiers of data pose many challenges. The first one hinges on the accounting mind-set that has been trained within the confines of guided and structured enquiries. “Any decision that the accountant takes will need to be supported by data,” says Stratopoulos. “We need to teach this kind of mind-set to future cohorts of accountants. It will take a generation.”

Two other challenges are linked to the final two v’s of big data: veracity and value. How can we guarantee that the data collected is trustworthy and how do we determine its value at the outset? An answer to these questions will probably come from another leading-edge technology: blockchain, which lies at the base of the notorious bitcoin.

Blockchain is a vast computer ledger established in peer-to-peer mode between countless computers. Its key feature lies in its extremely robust encryption protocol: any attempt to tamper with the data is immediately visible to participants of the network. Blockchain, says an Economist article, can be viewed as “a machine for creating trust.”

Financial institutions have only recently started experimenting with private blockchain networks and, if successful, these networks will multiply in the billions, says Michael Mainelli, emeritus professor of commerce at Gresham College in London, England, a pioneer who built a shared distributed ledger system 20 years ago. Such blockchain ledgers will find their way into accounting and auditing, ensuring high levels of stability and security to any type of data accountants will be working with. At some point, this will lead to “the total automation of bookkeeping,” predicts Stratopoulos.


Large corporations have made big strides toward the automation of bookkeeping thanks to the contributions of a plethora of technologies such as enterprise resource planning (ERP) and electronic data interchange (EDI) systems. “From the 1980s to the 2000s, corporations had huge accounting departments where armies of people carried out data entry, matching of invoices and countless clerical tasks,” says Sébastien Doyon, partner, leader for management and technology consulting for PwC in Quebec. “ERP system automation decimated these ranks.”

That has certainly not happened yet in small and medium-sized businesses, where the manual entry of bills, invoices and payments still prevails.

For example, the area of payment matching and posting in electronic portals is an extremely complex world that still requires SMEs to perform countless manual procedures, says Gilles Létourneau, president and CEO of Acceo Solutions in Montreal, which helps SMEs fully automate the accounting of payments. “That’s what we work at simplifying [thanks to a cloud-computing portal],” says Létourneau, “and that will allow SMEs to reduce their bookkeeping activities.”


Do the leading-edge developments in IT and accounting inexorably steer the profession toward a world where computers will hold the title of CPA and accountants will fill the unemployment lines?

It is too early to tell. For now, accountants are still in high demand, in spite of the job losses caused by the implementation of ERP and EDI systems through the 1980s and 1990s. “More than 90% of our graduates find a job even before they get their graduate diploma,” says Antonello Callimaci, CPA, CA, associate dean of studies at the École des Sciences de la Gestion of the Université du Québec à Montréal.

One thing is certain: the profession is moving away from the basic bookkeeping chores toward the more sophisticated analytical tasks. Those accountants who remain stuck in the Excel era and who feel intimidated by the new technologies are certainly at risk, says Monique Morden, chief revenue officer at Lendified in Vancouver, a company that provides working capital loans to SMEs using advanced analytics of financial performance and cloud computing. “They’re going to be in trouble.”

But even the analytic dimension of accounting seems to be threatened by all the new technology. Will even PhDs be forced to relinquish their CPA title to a robot?

History has shown time after time that technology creates its own new jobs. We’ve seen that happen, for example, in the automobile, aeronautics and pharmaceutical industries, which have spawned plant workers, mechanics, pilots, flight attendants, airport personnel, researchers and tourist agents. Yes, cars prompted job losses in the world of buggy manufacturers, but have created millions in the new industry.

Stratopoulos suggests that the resilience of a job to automation depends on how much judgment and critical thinking it involves. “So we don’t know. In theory, deep learning has the potential to affect many professions. But I’m optimistic. Humans are creative.”