Machine learning is a subset of artificial intelligence that excels at using data to find patterns and make predictions. George W. Russell takes a look at how this rapidly evolving technology is shaking up business – including the accounting sector – and how it will dramatically change roles within firms.
By George W. Russell
Any car owner who has taken a vehicle in for repairs knows the tedium of waiting for the mechanic to identify the problem. Heads under the bonnet, intense scanning of the manual, the unscrewing of parts and incessant engine revving and test-driving required to whittle down the many possible afflictions.
The clean laboratory environment of the technology arm of Hong Kong trading giant Jardine Matheson, could not be more contrasting to that of an oily garage. Yet inside a machine-learning application combines the knowledge of experienced automotive mechanics with data from an individual car’s service history and the manufacturer’s technical documentation.
“This means more relevant information can be made available to a wide number of employees, allowing them to perform tasks that they couldn’t previously,” says Mark Lunt, Group Managing Director at Jardine OneSolution.
It is one of many machine-learning technologies under development that can augment – or in some cases replace – human data processors. Machine learning refers to the development of algorithms – procedures or formulas for solving a problem – that discover and improve knowledge through an analysis of patterns in information.
“An important aspect of machine learning is the ability to update such intelligence with or without expert human assistance,” writes Venkat Srinivasan, Chief Executive Officer of Rage Frameworks in Boston and a member of the Institute of Chartered Accountants of India, in The Intelligent Enterprise in the Era of Big Data.
Venkat notes that in fields from energy to electronics, and medicine to manufacturing, machine-learning applications are revolutionizing business processes across the world. “Modern machine learning has emerged as an interdisciplinary field from the study of patterns. [It is] widely used in a number of applied sciences, including computer science and psychology.”
Some of the most groundbreaking work is being undertaken in China. Beijing-based Lenovo uses it to automate personal computer settings, while Hong Kong’s China Light & Power applies machine learning to electricity consumption management.
Chi Tsang, Head of Asia-Pacific Internet Research at HSBC, forecasts that Shenzhen-based Internet giant Tencent’s next phase of growth will come from machine learning. “They will interpret and exploit the vast amount of data on 939 million WeChat monthly active users.”
The finance sector has embraced machine learning as well. Some corporate research firms, such as Blueflag, based in Shenzhen, use it to comb through company financial data seeking matches to known traits for fraud. The Hangzhou-headquartered Zhima Credit – part of the technologically savvy Ant Financial Services Group – uses machine learning to generate personal and corporate credit ratings.
Automating the value chain
Globally, machine learning in accounting firms is still in its infancy, but its deployment is growing. “Businesses no longer need someone to come to their office and sort through mounds of bills,” says Nikole Mackenzie, CEO of Momentum Accounting, a San Francisco practice that focuses on cloud computing. “All this can be done online and machine-learning apps can handle much of the data entry.”
More value-added work is on the horizon, such as supply chain optimization, purchasing cycle improvement, cost control and forecasting, tax planning and accrual calculations. “The possibilities are endless,” says Alex Campbell, Managing Director, Asia, at the New Zealand-based accounting software company Xero.
Campbell believes the speed and efficiency offered by cloud computing has opened the doors for widespread adoption of machine-learning techniques in the accounting profession. “We believe accounting is ideal for machine learning, and have already released [products with] invoice coding capabilities,” he says. “Machine learning and artificial intelligence will drive long-term growth.”
Xero is one of several software companies developing machine learning for higher-value processes. “The objective is to give accountants back the time to focus on advisory services that drive better relationships with their clients,” says Campbell. From simple tasks such as invoicing, software developers have been able to innovate with machine learning to advance into simple tax preparation, payroll and banking integration.
Machine learning’s next iterations could enhance the audit process, such as validation of transaction coding and analysis of general ledger input. “Part of the role of the auditor is to test samples of data to identify inconsistencies,” says Campbell. “Machine learning opens up the possibility of testing the full set of data, which means higher integrity results. An auditor can spend more time testing systems, controls and separation of duties rather than substantiating documents.”
Lunt at Jardine OneSolution sees a future in which machine learning is deployed in investment appraisals and return on investment calculations as well as behavioural, customer and market analyses. “They all have potential to leverage machine-learning technology to support improved decision making,” he says.
For chief financial officers and other executives, machine learning might find its most useful role in forecasting. “Companies can benefit significantly from the use of predictive analytics tools to analyse the past data and use algorithms and machine-learning techniques to predict the future,” says Surabhi Kejriwal, Financial Services Industry Team Leader at the Deloitte Centre for Financial Services in Mumbai.
An evolving technology
Machine learning was developed in the 1950s. Arthur Samuel wrote the first computer-learning program: it played draughts, or checkers, on an IBM 701 computer. By the 1960s, computers could discern basic patterns. For the next two decades, artificial intelligence focused on physical capabilities and spatial intelligence, leading to Stanford University’s “Stanford Cart” in 1979, which could navigate obstacles in a room.
In 1984, Science magazine asked: “What is learning? How can a machine modify its knowledge according to experience? How can it be taught to learn from its mistakes?” Technology that mastered these questions meant that by the mid-1990s, machine-learning applications included bid behaviour prediction, insurance premium calculation and manufacturing quality control.
Another early application was forensic accounting: since the 1980s, software could detect mortgage fraud, predict bankruptcy and catalogue access to sensitive information. “The ability to analyse and correlate vast amounts of data have revolutionized fraud detection,” says Matthew Bosher, Partner at Hunton & Williams in Richmond, Virginia, who focuses on financial reporting and corporate governance.
Algorithms are trained to highlight out-of-place transactions for subsequent human review and investigation, says Lunt at Jardine OneSolution. “Another area is approvals processes, such as the vetting of loan and credit applications, where machine learning can quickly screen applicants leading to less human involvement and faster approval times – all with no loss of accuracy or compliance.”
Campbell at Xero sees potential for small- and medium-sized enterprises, which tend to hold their data separately. If their data was aggregated in the cloud, anonymously, machine learning could be applied to identify high-risk points. “With machine learning we can expect to learn exceptions and identify high-risk transactions, and therefore reduce risk, for example in identifying [tax] fraud,” Campbell points out. “SMEs’ data have typically been in silos, which means they can’t be validated against other sources.”
Machine-learning apps can be seamlessly integrated with other types of financial technology. “For instance, combining distributed ledger technology, such as blockchain, for payments processes with fraud detection algorithms could potentially make such systems both more secure and more efficient,” says Lunt.
That opens up many new areas for regulators, notes Bénédicte Nolens, Senior Director, Risk and Strategy at the Hong Kong Securities and Futures Commission. “Machine learning… is increasingly used to supplement or automate trading decisions, as well as to further automate compliance and risk management,” she says. “This journey is only at its inception.”
Effects on employment
What that journey will mean for accounting-sector professionals remains to be seen. According to Martin Ford, the Silicon Valley futurist and author of Rise of the Robots: Technology and the Threat of a Jobless Future, it has become increasingly clear that machine-learning algorithms “are gradually going to consume much of the base of the job skills pyramid.”
Ford cites the disappearance of jobs for lawyers and paralegals who once would have sorted laboriously through cardboard boxes full of paper documents. Some CPAs agree that a similar revolution is headed for accounting. “Accountants who aren’t already or at least considering the transition… will be unable to keep up with the pace at which technology is driving changes in the industry,” says Campbell at Xero.
Mackenzie at Momentum believes it is inevitable that machine learning will take over higher skills. “However, I don’t think this is a bad thing,” she says. “It’s more a matter of using our fundamental skills and understanding of accounting concepts and applying them in the current environment. For example, CPAs are now expected to understand how to use all of this technology, which may seem daunting for some.”
Easier access to new technology could put nimble, proactive practices at an advantage, Mackenzie adds. “I have a feeling we are going to start seeing a bunch of boutique firms pop up who are only doing accounting services and they will be able to offer services at a lower cost,” she says. “It will be interesting to see how this affects the larger firms.”
Campbell says that although technology is second nature to the younger generation of CPAs, they still need to understand the accounting fundamentals that underpin innovations. “An older generation of accountants has the technical know-how, but may lack the understanding of technology to quickly adapt to new processes,” he adds. “In our increasingly digitized age, both will need to leverage each other’s strengths to work effectively in a constantly evolving world.”
Mackenzie sees machine learning as part of a series of evolutions in the profession. “If you look back at the history of accounting before Microsoft Excel and other software, accountants used to use these long spreadsheets and it would take weeks to manually add things up,” she says. “It’s simply a matter of changing our job descriptions a little bit. Heck, maybe the term ‘bean counter’ will finally become obsolete.”
This article was originally published in A Plus magazine.