At first he thought big data analytics was a fad. But then Thomas H. Davenport, the Harvard Business School professor and author of Competing on Analytics, said his skepticism evaporated after he did some research.
While traditional analytics mostly supported internal decisions, Davenport said big data analytics is much more likely to involve data from outside a company and be focused on creating benefits for customers. That ongoing use cases range from LinkedIn’s recommendations engine to customer loyalty programs at Caesar’s Entertainment and data-collecting engine sensors at GE—the evidence suggests that the work of analyzing large and varied datasets is informing decisions at both established corporations and innovative start-ups, Davenport said.
There also is evidence that companies employing data-driven decision-making are better off, said Erik Brynjolfsson, director of the MIT Center for Digital Business. A survey of 179 large public companies found that such companies were 5 to 6 percent more productive and also correlate to better return on equity. Brynjolfsson posted a research paper about the findings last year.
Brynjolfsson and Davenport discussed their assessment of the big data trend so far in a Nov. 30 webcast organized by MIT Sloan School of Management and sponsored by the analytics vendor SAS.
Judging by questions from the audience, which were dominated by queries about how small- and medium-sized firms could deploy big data analytics to compete with much larger enterprises, the experts were acting as both teachers and professional counselors, encouraging business people new to analytics to experiment with the tools and datasets, but to keep ROI top-of-mind. Other messages from the session:
Data scientists are tough to find, tough to keep. The term “data scientist” is so popular right now that it is losing its meaning because so many are claiming the label, Davenport said. Still, he said these are people who have an ability to solve business problems. Those rare figures who understand how to apply statistical algorithms to business problems want to work with top executives (not just managers). “The thing you really have to have to be successful is an orientation to the scientific method, seeking out sources of error and a commitment to find the truth,” he said.
If you can’t compete in the market to hire such experts, Davenport said you can hire consultants to work on a project, or even run a competition on Kaggle.com, an online platform for data prediction competitions.
Tie data experiments to business objectives. In the short and medium term, managers who have not touched big data should look for an opportunity to experiment. “You need to be thinking about the datasets you have already, and what external assets might be relevant to your business,” Davenport said. It’s essential to start experimenting now because lessons learned will inform future experiments and eventually could lead to building new capabilities.
And pay attention to what others in your industry are doing. “Be careful if competitors are moving faster than you are,” he said. For example, Davenport said he recently asked an executive at an auto maker what he thought of Google’s ongoing plans to develop self-driving cars—a data-driven engineering project. The manager told him that his company would let Google do the experimenting. “That seemed like a bad idea to me,” Davenport said.
The principle applies to small companies, too. They can access external datasets, such as from government sources to conduct experiments, he said. A lot of the big data firms are start-ups, and it takes a relatively small amount of money to acquire a large amount of data.
Organizational issues are more challenging than technology puzzles. There will be ongoing rapid progress on developments with Hadoop and other emerging technologies, Davenport said. Technology problems will be easier to solve than the skills needed to make the technologies work. “The big constraining factor is the people, who are not open source,” he said.
These managers should watch how emerging technologies influence their own business models, and how they influence the culture of their organizations. Brynjolfsson said that in data-driven organizations, leadership is not about asserting confidence in one’s gut instincts. It’s more about be willing to listen to the data and having “a methodology for getting the data to speak.”
Pay attention to data privacy and security. Both Davenport and Brynjolfsson said that there is no near-term or easy solution to the problem of guaranteeing consumer data privacy, and the potential for problems grows as datasets around the world grows. Brynjolfsson said that business, government and academia all have to pay attention to the issue and develop “new institutions and new ways of thinking” about data privacy and security.