In the summer of 2016, Charles Weinstein, CEO of New York City–based accounting firm EisnerAmper, had an epiphany: Machine learning could either destroy his business or remake it.
A 35-year veteran of the industry, Weinstein sensed that the practice of accounting—issuing financial statements three months after the fact—while still necessary, was losing relevance in the real-time, data-driven economy. So he organized a three-day partner meeting to consider how machine-learning capabilities in particular might remake the traditional accounting firm for the digital era, enabling it to help its clients look into the future rather than simply reporting on the past.
Weinstein invited a partner in charge of global innovation at a big-four accounting firm (not a direct competitor) to talk about the moves his firm was making. As the visitor spoke, it became plain to Weinstein that there was little time to waste.
“That meeting was a watershed moment; it created a united mindset for the firm around deciding to lead, not follow, when it came to leveraging technology in the accounting space,” says Weinstein. “It was clear that machine learning would have enormous benefits for our clients and our firm.”
Eighteen months later, the 1,500-person firm is using machine learning to develop smart auditing tools, where the software actually learns how to learn in order to make the process more effective and more efficient. It has rolled out systems in the areas of revenue recognition and process controls, with plans to become fully machine-learning driven within three years.
Machine learning will free up EisnerAmper’s practitioners to spend more time providing clients with high-level advisory services and strategic consulting, while offering traditional auditing services at a lower price than competitors. And that’s just for starters. “In many of our lines of business,” says Weinstein, “the change will be highly transformational.”
That’s the thing about machine learning: It changes everything. While people said the same thing about the internet, the internet turned out to be simply the platform for change. Machine learning is the change, with the potential to alter the very nature of the enterprise.
“What people don’t understand is that machine learning at scale doesn’t just make a company smarter,” says Keith Strier, EY global and Americas advisory leader for artificial intelligence. “It literally redefines the nature of what a company is and how it operates and how it does what it does.”
Indeed, companies that are adopting machine learning are not in search of incremental improvements in the status quo; they see it as a competitive differentiator. Making the Most of Machine Learning: 5 Lessons from Fast Learners, a study by The Economist Intelligence Unit (EIU) and SAP, found that among the companies that are already using machine learning and seeing benefits from it, nearly a third (31%) said machine learning was already yielding business process or business model innovation. The goal with machine learning is to transform an existing business or to create a new one.
The disruptive power of machine learning comes from its ability to learn. A major discipline of artificial intelligence, machine learning uses sophisticated algorithms to enable computers to learn from large amounts of data without being explicitly programmed. The more data the algorithms can access, the smarter they get.
Through constant learning, machine learning improves the execution of tasks and processes autonomously and continuously. That differentiates it from traditional software such as enterprise resource planning (ERP) or customer relationship management (CRM) solutions, where you get a one-time benefit from automating a process and then make incremental improvements over time.
Because it is always evolving, machine learning creates new business opportunities as it grows smarter and more effective—without creating additional overhead. “It’s probably the most significant business implication of machine learning: the enablement of nonlinear growth,” says Cliff Justice, principal in KPMG’s Innovation & Enterprise Solutions team. “You are able to offer more products and more services and your cost doesn’t necessarily have to accelerate with your growth. That’s huge.”
But machine learning doesn’t just affect business models; the changes cascade throughout the organization. “Machine learning will change our service-delivery model, our staffing paradigm, our learning and development programs—ultimately leading to an enhanced value proposition for clients,” says Weinstein.
And while machine learning represents a big leap in technology power, it doesn’t require much messing with the guts of existing technology. Indeed, it is manna for frustrated business leaders who have been sitting on top of mountains of untapped data in their organizations for years now.
That has led to extraordinarily fast uptake. As little as two years ago, few people even knew what machine learning was. Now it has blasted past bleeding-edge adoption to become a top strategic priority at companies across industries. The EIU/SAP survey found that 68% of respondents are already experimenting with it.
The survey also found that machine learning goes right to the top and bottom lines. Forty-eight percent of companies that are already seeing benefits from machine learning cited increased profitability as its most predominant gift, while around half (48%) said they expect revenue growth of more than six percent through 2019.
Machine Learning Moment
Around the same time EisnerAmper was having its partner meeting in New York City, global law firm Pinsent Masons was having its own come-to-machine-learning moment across the Atlantic in London. An early adopter of advanced robotic process automation, the firm’s technology leaders were eyeing machine learning’s potential to completely restructure the firm’s customer relationships and create new lines of revenue. “The bigger opportunity is to become providers of knowledge-based systems to clients, which means moving from a services model to a product model,” says David Halliwell, the company’s director of knowledge and innovation delivery. “It’s about licensing your knowledge rather than just providing services by the hour.”
The firm is already using machine learning to help its financial services clients perform due diligence before making bids to acquire financial asset portfolios in order to understand the risks hidden in them. Machine learning streamlines significant elements of labor-intensive, large-scale contract review work.
For example, human reviewers can only afford to look at a representative sample of the financial assets for multibillion-dollar transactions to determine risk; otherwise the costs to the firm and to clients would skyrocket, with uncertain returns. So in the summer of 2017, Pinsent Masons began using its own machine-learning platform to extract, review, and analyze key contract risks in the context of a number of multibillion-dollar transactions.
With machine learning, the system can ingest and review all the available data and provide not only a quicker answer but a better one—and at a lower cost. That wins the firm more business because “we are doing this earlier—and better—than other law firms,” says Halliwell. “And it helps us build a reputation for innovation in the marketplace, which is really powerful for attracting our clients but also for attracting employees.”
Not Just a Services Thing
Intel’s commercial interests in machine learning are clear. It unveiled a new family of chips designed especially for artificial intelligence in late 2017; it’s also been adopting machine-learning capabilities within the enterprise. For several years, the company has been using decision-support systems, powered by machine learning, to analyze some of the 5 billion data points produced each day in each one of its factories to increase uptime, accelerate output, and decrease or prevent faults.
In the past year, the company has begun to employ what it’s learned on the industrial side for customer-facing processes. Intel’s chief data officer and vice president of enterprise data and platforms, Aziz Safa, says there is an unbreakable rule for all machine-learning initiatives in the company: They must be measurable and reportable so that company leaders can be assured they are making the right investments.
A programmer would not be able to understand the value of the raw data, and an accountant is completely unable to figure out what to do with the technology. But working together in small, nimble cross-functional teams, they come up with new ideas.
-Charles Weinstein, CEO, EisnerAmper
Intel is not looking to simply improve what it is already doing—cutting costs here, increasing productivity there. The goal is to uncover entirely new areas of revenue and growth. One of its first revenue-generating machine-learning applications is a sales-enablement system that can identify the resellers that offer the highest probability for sales.
With more than 100,000 reseller-customers, Intel’s sales force traditionally could only afford to focus its efforts on the largest ones. Thanks to machine-learning algorithms, the company can now understand more intimately the broader universe of resellers and their likely needs. The system has already delivered more than US$100 million in additional revenue.
Those topline results have received “a warm reception and enable us to move faster on our trajectory toward more heavily implementing machine learning across the company,” Safa says. Indeed, machine learning is “at the heart” of Intel’s digital transformation, he adds. “In 10 years, it will be embedded in everything we do.”
However, companies must make some fundamental changes to reap machine learning’s differentiating benefits. “Machine learning is a strategic business model underpinned by very complex tools and methods,” says EY’s Strier. “You don’t implement it, you apply it. It requires ongoing organizational commitment to keep feeding, training, refining, and expanding it in ways you may not have even thought about when you started.”
EisnerAmper’s Weinstein, for example, doesn’t yet know all the ways machine learning will evolve within the firm but he is committed to figuring that out over the next decade.
“If an enterprise wants to be a machine-learning enterprise, it must evolve to a state of continuous learning to improve the performance of its business processes, its customer experiences, the way it interacts with suppliers and regulators—whatever the strategic goals are—over time,” Strier says.
C-Level Sea Change
Given the extraordinary amount of change that machine learning will bring to organizations, not only must the C-suite be engaged, it must understand and direct the implementation strategy. At Pinsent Masons, “the first step was creating an understanding within the firm’s leadership that this was something that could be transformational,” says Halliwell. “We’ve worked really hard on understanding what the possibilities are and what machine learning cannot do. Without that understanding and that backing, we would have been struggling.”
For Weinstein, who has led his firm’s machine-learning effort from the corner office, machine learning is an opportunity for him as well as the firm. “It’s really exciting for me to participate in this transformation with our firm and with the entire profession,” he says. “That’s why I’m so engaged.”
What’s more, the investment required to be a leader in the machine-learning trend has been significant, so Weinstein’s buy-in is important. “With the amount of capital required to do this,” he says, “I have to make sure we’re moving in a good direction.”
Given its potential for remaking business models, machine learning can send shock waves of change throughout an organization—through its processes, its staffing models, even its culture. The first machine-learning enterprises are just beginning to grapple with these shifts, some of which become clear early on and others of which will only emerge later.
The change begins with staffing. The machine-learning enterprise can’t be built by traditional developers or engineers alone; it needs business process experts working together with the machines and the techies.
Part of the reason CEO and board involvement is critical is machine learning’s “impact on resourcing models in terms of new skills that are needed to exploit the technology,” says Halliwell. “One mistake others have made is to think of the investment purely in terms of technology budgets, not people budgets.”
EisnerAmper has been hiring data scientists, business analysts, and application developers into its enterprise technology group. The new hires partner with accounting, auditing, and tax leaders to develop new capabilities.
“A programmer would not be able to understand the value of the raw data, and an accountant is completely unable to figure out what to do with the technology,” says Weinstein. “But working together in small, nimble, cross-functional teams, they come up with new ideas.”
Over time, the machine-learning enterprise starts to look and function differently. “With machine learning, the changes required are large, not just in terms of capital investment but in terms of significant culture change and learning within the organization—breaking through the legacy ways of working,” says Arjun Sethi, partner with A.T. Kearney and leader of its digital transformation practice. While humans still run the machine-learning enterprise, “the nature of that organization will change because of the speed and scale with which machines can learn,” Strier says.
At EisnerAmper, machine learning is accelerating the firm’s shift from traditional accounting services provider to holistic business advisor. Over time, basic accounting services will be heavily automated, and there will be significant opportunities to provide new financial and business insight to clients.
“We’re going to have access to huge amounts of data for those clients,” says Weinstein. “That’s where our skill comes in. We’ve been processing accounting data forever, but we’ve never had the ability to analyze such huge volumes of data. That will enable us to come to market with different products for our clients and help them retool their operations.”
The only thing that’s clear about the changes required to get there is that “they will be huge—just huge,” says Weinstein. “They won’t be coming in the next two years, but it won’t take 10 years either. That gives us some time to work on our staffing models and our learning and development models. It’s early on, and we’re learning.”
Indeed, it’s shortsighted to see machine learning as purely an operational effectiveness engine automating human activities, says Halliwell of Pinsent Masons. “It is creating new products for us—categories of work and outcomes that we just couldn’t do before,” Halliwell explains. “That is partly a product of scale; tasks that were too big to contemplate with only manual resources are now achievable with machine learning. But because we are moving our lawyers from reporting in words to reporting in numbers—from statements of fact to statements of data—our model is now looking at playing the results back in a way that gives our clients access to data and the ability to analyze that, rather than a weighty legal report.”
That will change the way the firm charges for what it does, from billable hours and reimbursable costs to software licensing models or value-based pricing. It will also have a long-term impact on staffing and headcount, but the specifics remain to be seen. “We don’t think anyone has worked out what that will look like,” says Halliwell. “More lawyers? Fewer lawyers? A different mix?”
Right now, the focus is on getting more attorneys to adopt machine-learning capabilities. “Not all lawyers get the way the technology can be used and the way they need to change what they do,” says Halliwell. “The key for us is to identify those who are pioneering its use in different service lines and sectors, getting them up to speed, and then working with them to expand out in their sector.”
The Machine Learning Enterprise
As the amount of data available to companies continues to grow, the quantity and complexity of the data are exceeding human analysts’ processing capabilities. Machine learning will quickly evolve from a competitive advantage to a competitive necessity. It will become increasingly difficult to make sense of the data flowing into the organization without the integration of machines that are able not only to analyze the data but also to learn from it.
Companies seeking the greatest value from machine learning must be willing and able to “adapt the organization faster than your competitors and transform the way your employees work and the way you interact with customers,” says KPMG’s Justice.
Those that recognized this in recent years have a head start in their machine-learning evolutions. “I think we’re quite a long way from really knowing what the future will look like,” says Halliwell, “but the attitude we take is that it’s better to be at the front of the wave, trying to explore these possibilities, having some choices to make and the luxury of time to make those choices, rather than being forced into a decision because we were late to the party.”
Intel recently began creating an enterprise strategy for machine learning, covering everything from the supply chain to sales and marketing. “We started with smaller targeted investments in proofs of concept,” says Safa. “Now that those have been established, it’s easier to move from one area to the other with those as a reference point for how to apply machine learning within the company.”
EisnerAmper is likewise expanding its efforts over time. But Weinstein expects machine learning to have an exponential impact. “We are implementing our machine-learning program incrementally and methodically, service by service,” he says. “But we’re thinking about the end game in big terms. Those firms that aren’t taking steps right now will be left behind.”
Dirk Jendroska is head of Strategy and Operations Machine Learning at SAP.
Dan Wellers is the global lead of Digital Futures at SAP.
Stephanie Overby is a Boston-based journalist.