This month the New York Times launched a new personal-technology column with the clever title “Machine Learning.” But behind the scenes, the newspaper is also doing real machine learning, and for a very serious purpose: it wants to predict who is going to unsubscribe before it happens.
The Times has hired Columbia University applied mathematician Chris Wiggins as its first “chief data scientist.” Wiggins, who has built predictive computer models to determine the origin of viruses, will now lead a small group that tries to use mathematics to help the 162-year-old publisher maintain or expand its subscriber base.
Wiggins says he took the part-time job because he’s a big fan of the publication and wants to help keep its 1,200 editors and reporters in their jobs. He thinks data science can help with “a business model that has been severely disrupted.”
The problem is that advertisers are buying fewer newspaper ads. U.S. newspapers’ print ad sales collapsed from $47 billion in 2005 to $19 billion in 2012, according to Pew’s State of the News Media report. Online advertising hasn’t come close to making up the shortfall.
In response, the Times is depending more on subscriptions. It has raised the cost of the printed paper and is selling digital subscriptions with the help of its digital paywall, launched nearly three years ago with significant success. In reporting otherwise glum yearly results this February, the company said it had seen a 19 percent increase in digital-only subscribers during 2013.
Marc Frons, chief information officer of the New York Times Company, which employs about 500 people in IT, technology, and programming, says that a year ago the company pulled together information that had been in “various silos” and formed a business intelligence group to start exploiting it. It’s part of a wider trend in which companies are using data to guide business decisions more directly (see our report “Data and Decision Making.”)
Wiggins will head a team of three or four people with a mandate to experiment and to determine whether the Times’ business problems can be solved with machine learning, a set of statistical methods that use existing data to make predictions about similar situations.
The Times doesn’t lack for data—its readers make nine million visits a day to its home page. “But we really needed someone to give us insights about why people subscribe and how to retain them,” says Frons. “Before they pick up the phone and say ‘I want to cancel,’ you could predict by the patterns of their behavior, like not logging in as much, that they might do that.”
The Times’ effort is still relatively modest in scope, at least compared with those of the largest Web companies. Google this year spent $400 million to acquire a single machine-learning startup (see “Is Google Cornering the Market on Deep Learning?”), and Amazon is currently advertising positions for 40 machine-learning scientists—in addition to scores it already employs.
What’s notable is how traditional media organizations are catching up. In October, News Corp., publisher of the Wall Street Journal, hired Rachel Schutt, another Columbia statistician specializing in networks, as senior vice president for data science. That’s also a newly created role. Schutt wrote in an e-mail that she’s going to be a “centralizing force” for “a number of machine learning and predictive modeling projects” that were already under way.
Big publishers are also looking to keep pace with media startups, some of which have been hiring Wiggins’s students. Websites such as BuzzFeed, publisher of a mix of silly and serious news, have shown that mathematical modeling can help increase stories’ “virality”—a measure of how widely the content gets shared online. Sites like BuzzFeed and the Huffington Post now get more page views than the New York Times, but with far smaller staffs.
For its part, the Times says it’s not after anything so crass as more page views. “Something on BuzzFeed that goes viral, and gets a lot of clicks, is different than the relationship that the New York Times is hoping for and building its business around,” says Wiggins. “It’s really putting subscriptions, rather than stories going viral, at the center of their model.”
This new data poisoning tool lets artists fight back against generative AI
The tool, called Nightshade, messes up training data in ways that could cause serious damage to image-generating AI models.
Rogue superintelligence and merging with machines: Inside the mind of OpenAI’s chief scientist
An exclusive conversation with Ilya Sutskever on his fears for the future of AI and why they’ve made him change the focus of his life’s work.
The Biggest Questions: What is death?
New neuroscience is challenging our understanding of the dying process—bringing opportunities for the living.
Data analytics reveal real business value
Sophisticated analytics tools mine insights from data, optimizing operational processes across the enterprise.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.