“I’ve suggested using it for a couple of things here recently but got persuaded otherwise,” he says. “But I still think I’m right, and I’m really persistent.”
What Scapa recognizes is that machine learning “leverages the fact that there’ s a huge amount of computing available to try to solve problems, predict the future and prescribe solutions.”
Indeed, Big Data is revolutionizing business, much as the Internet began doing a quarter-century ago, and machine learning is emerging as one of the most important tools for utilizing it. Just as with the web before it, CEOs like Scapa realize they must use machine learning to gain a competitive advantage for their companies—or they’ll most certainly be seeing their rivals do it instead.
Machine learning is a branch of artificial intelligence that enables machines to learn on their own, without much human supervision, drinking deeply from the well of Big Data. Computers
essentially write and follow their own programs based on the statistical relationships they discover in unstructured data—and are roiling industries ranging from credit cards to automobiles
in the process.
“It’s speed and the ability to learn from data that gives machine learning the power to provide tremendous insights in ways that humans could never do on their own or with basic business-intelligence tools,” says Mike Tuchen, CEO of Talend, a Los Altos, California-based big-data integration firm.
Or, as Altair’s Scapa put it, machine learning “can use algorithms to mine historical data for outcomes that are different than with traditional simulation.” At financial companies, for instance, machine learning can assess insider-trading activities and identify potential fraudulent activities that could trip a regulatory investigation. At utilities, machine learning can work with digital smart meters to identify sources and patterns of energy consumption in a house so that the company can “generate revenue around that data,” says Josh Sutton, the global artificial-intelligence lead for Sapient consultants.
Consider how machine learning is improving credit-card authorization by leaps and bounds. “The system learns by itself and actually keeps evolving algorithms by itself, as authentication keeps automatically refining on its own,” says Raja Rajamannar, chief marketing officer for New York-based MasterCard. “So the accuracy of what we’re doing increases, while the false positives and false negatives keep declining.”
Or look at what True Fit is doing with machine learning. The Woburn, Massachusetts-based startup amalgamates all sorts of data about an individual consumer, including purchase patterns,
to curate online clothing selections just for them, and machine learning makes TrueFit more helpful as it learns more about the customer.
Automatic Data Processing (ADP) has figured out how to use machine learning to advise clients which employee time cards need to be reviewed and which don’t, based on previous diligence patterns of the employees—sort of like the “pre-check” feature for travelers deemed safe that the federal Transportation Security Agency initiated at airports.
The French company, which has U.S. headquarters in Andover, Massachusetts, also applies machine learning to its building-management systems to make structures more eco-friendly and comfortable for occupants.
Atlanta-based GeoDigital is using machine learning to apply its detailed three-dimensional models of the world to a number of industries. For utilities, GeoDigital can precisely locate assets as mundane as power poles and then use machine learning to recommend maintenance steps and schedules. Geo-Digital’s 3D-mapping programs also are feeding the revolution in the automobile business known as self-driving.
For vehicles to be truly driverless, yet safely complete their routes, they’ll rely on a cloud of data and a flurry of instantaneous communications that will keep them apprised of all the conditions around and within the vehicle—both static, like the contours of roads, and dynamic, such as traffic conditions and whether a pedestrian is darting in front of the car at that moment—and advise them how to respond.
“What used to be acceptable becomes an impediment if you need a vehicle to make decisions on its own about whether to make a maneuver immediately, or prepare for traffic conditions a few miles away,” says Anupam Malhotra, senior manager of the connected vehicle for Audi of America. “Machine learning is how we close the gap to make truly automated driving possible.”
Big Moves in Big Data
Machine learning is just one way to leverage Big Data. Here are examples of how CEOs are using a variety of tools with Big Data to advance their companies:
• At TransUnion, the Chicago-based credit-scoring giant, a new data-powered product called Credit Vision is providing lenders with more granular and nuanced assessments of a consumer’s risk factors, which is helping them unearth more high-quality prospects and make more sound loans.
“This also is helping us increase our footprint with the lenders we serve and boosting the ‘stickiness’ of our other products, such as fraud solutions,” explains TransUnion CEO Jim Peck. “That helps us maintain and grow our positions with our customers.”
• Edmunds.com, the automotive-information web site, has been rolling out a new product called Car Code, which allows consumers to directly text dealers about vehicle pricing, availability, features and questions. Predictive analysis of the resulting millions of conversations turns up key findings that will help make the next texting exchanges more effective.
• Neil Clark Warren returned to his previous CEO position and boosted the importance of data massaging to help another Santa Monica-based company, the dating site eHarmony, recover against key competitors such as Match.com. Actionable insights from data analytics included the fact that dating candidates from outside the U.S. are more comfortable being matched with someone who smokes cigarettes or drinks booze than eHarmony’s American members are, so it broadened the matching algorithms for overseas users.
• Intel has used its Big Data expertise to diversify into healthcare over the last couple of years with wearable monitors, ranging from 24/7 tracking of Parkinson’s patients so they can get improved treatment to the introduction of smart headphones with sensors that can measure heart rates and which carry the lifestyle brand of rapper Curtis “50 Cent” Jackson.
• Philips North America is analyzing data from its medical monitors to predict which users will end up in a hospital in the next 30 days, helping acute-care providers intervene to alleviate chronic conditions at home and pre-empting the need for expensive readmissions to a hospital.
• AMN Healthcare Services is crunching data with predictive analytics to extend scheduling of nurses and its other contracted staffers as long as four months, with 95% reliability that it will work out satisfactorily for each staffer, by using patterns such as the absenteeism and overtime histories of individual nurses. That compares with the historical practice of being able to schedule out only two or three weeks—and on a hit-or-miss basis, at that.
“It’s a big dissatisfier if nurses can’t plan in advance, and nurses are about one-quarter of a hospital’s budget,” explains Susan Salka, CEO of San Diego-based AMN. “With nurse attrition at historically high levels right now, anything you can do to keep them happy is a financial benefit to you, because they’re more likely to stay.”
Making Big Numbers Work
Effectively leveraging Big Data and machine learning requires more than a CEO’s determination. Here are some other considerations:
Now is the time
It’s feasible now to do machine learning at scale and in real time, because of how capabilities have improved and costs have come down. Yet CEOs will have hesitations. “They’ll have difficulty maintaining trust and balance with these learning systems,” says J. Lance Reese, COO of Limu, a Lake Mary, Florida-based maker of energy drinks and other beverages. To overcome this problem, he advises CEOs to “have more direct contact with technologists” and others who direct machine-learning systems.
Boards are expecting action
Directors are the primary influencer—above CEOs or CIOs—of Big Data adoption strategies in 53% of the companies surveyed by Accenture and GE.
Where companies are striving to make an impact with Big Data within a year or two, board members are driving much of the urgency. That’s one good reason to “make data and analytics a larger part of every board meeting,” John Kelly, leader of Berkeley Research Group’s predictive analytics practice, advises CEOs.
The best way to begin leveraging Big Data is to “have a hypothesis and act on it and test the data to see if you can harness insights” rather than “put all the data together and see what story it tells you,” says Anshu Prasad of A.T. Kearney. That can be difficult for CEOs to do because “all the function leaders are saying, ‘I have to do something with Big Data and analytics,’ so 1,000 tulips are blooming.”
Start with experimentation
Probe the capabilities of machine learning in a limited way at first. “Get someone to play around with it,” advises Altair CEO Jim Scapa. “Get a smart kid.” That’s why one of Altair’s first forays into machine learning is a project at Scapa’s alma mater, Columbia University.
Cost cutting isn’t enough
By now, using Big Data and analytics to cut costs, boost efficiency and grow existing business streams isn’t enough; they’re table stakes. The CEOs seeing the greatest opportunity to drive new revenue streams from big data are those confronting the highest level of disruption from it, CapGemini reports. Many of these companies are actually finding ways to monetize the data itself to create new lines of business.
Silos can be broken
Companies that best harness Big Data break down barriers to its use within and outside. Internally, that means people coming together from various business functions to work with IT to engage the Big Data agenda. It can also mean collaborating with other companies.
Hershey CEO J.P. Bilbrey, for instance, spearheaded creation of a new consortium for data analytics by partnering with Big Data pioneer Palantir, which is getting a group of CPG rivals to swap data to jump in front of emerging trends that affect the industry as a whole.
Garbage in, garbage out
Machine-learning systems can’t correct wrong data and can only complement, not completely replace, human intelligence. “It’s very good at pattern recognition and finding correlations in massive databases, but not so good at reflecting the causal structure of what’s going on,” says Jim Guczcza, chief data scientist for Deloitte Consulting in the U.S. “That’s where domain knowledge comes in handy.”
Humans still count
CEOs must combine Big Data operations with the domain expertise of actual human beings to get the best results. “You can’t just turn it all over to a machine that will do your thinking for you,” warns Edmunds.com CEO Avi Steinlauf. “Tools can help, but you need smart people thinking about things in curious and intuitive ways.”
Extra bandwidth becomes available
Effective machine learning reduces the need for human labor, but that doesn’t necessarily mean it should eliminate jobs. “You can certainly reduce staff and cut costs, but we see many clients looking to leverage this newly freed capacity to enable workers to provide additional value to the business,” says Chip Wagner, CEO of Alsbridge, a technology-consulting firm.
Democracy energizes data
The greatest potential for data can be unlocked by giving front-line workers essentially unlimited access to information that used to be difficult to obtain or required more senior managers to interpret. Tesla, for instance, finds that this approach improves manufacturing and quality because test engineers and others can see things in the data that their superiors have missed.
Knowing is not doing
Judging by one important criterion, few companies so far demonstrate actual Big Data skills: Even though the demand for data scientists has tripled in three years, only 6% of large companies employ as much as one of them. What’s more, because data scientists with experience remain few and far between, the talent chase will get only worse. This is one reason that only 27% of executives surveyed by CapGemini described their big data initiatives so far as successful.
Privacy issues must be addressed
Companies’ use of Big Data creates an unwelcome companion: distrust of that use because of privacy concerns. The more data that B2C companies have compiled about them, their purchases, their behavior and their preferences, the more consumers can be cautious about sharing such information and sensitive about how the data is used. One way CEOs can counter this obstacle is to openly offer benefits to customers in exchange for their private data.
For a retail chain that wants to use data it is gathering about in-store shopper behavior, for instance, this could involve offering free Wi-Fi in the store or enhanced service from a “personal shopper.”
Make it a personal tool
CEOs, including Carlos Rodriguez, of Roseland, New Jersey-based ADP, now are using machine learning to prepare more easily and effectively for board meetings and analyst updates. “Programs can go out and dig up all the information that the CEO used to have to reach for individually, package it up, apply Big Data analytics and present it to him,” says Stuart Sackman,
ADP’s CIO. “For each subsequent meeting the system gets smarter and smarter in condensing and refining data.”
Patience is required
Machine learning, like human learning, remains a work in progress. “It’s never quite done and it’s never quite right,” says GeoDigital CEO Chris Warrington. “But the longer you live with it, the better it becomes. So as a CEO you need to be patient with it.”