The New York Times is using machine learning to inform all aspects of its business, from aiding newsroom decision-making to understanding why people stop subscribing.

Chris Wiggins, the news organization’s chief data scientist, addressed this topic at a recent New York conference, where he enumerated the various ways in which such technological innovation is helping the Gray Lady achieve its objectives.

At one end of the spectrum, he explained, initiatives occur when “somebody hands you a bunch of data and says, ‘Give me a simpler description of this’.” (For more details, read WARC’s report: How The New York Times is tapping the power of machine learning.)

A case in point: ‘Readerscope’, an advertising product introduced in 2018, utilizes anonymized data to break out readership insights and trends by geography and audience segment. This tool is intended to guide the development of campaigns and branded content, and to better target digital ads on

Wiggins described it as an “internal dashboard”, but one that is frequently shared with marketers interested in advertising on The New York Times’ web properties.

“So, for example, you might be interested in knowing what topics that C-suite or executives or business decision-makers engage with,” he suggested. “We can say: ‘Here are the topics that overindex in Los Angeles, and here are the topics that overindex in Houston, and here are the types of reader that are overindexing in that place or that country’.”

Machine learning is also helping journalists in novel ways, taking datasets from them and highlighting areas of potential interest. This approach – “a prediction problem where you’re trying to help somebody be more efficient in the newsroom” – led in one instance to a series of stories that resulted in a product recall.

From a business point of view, the Times is keen to understand why online subscriber customers may not renew and has deployed machine learning to identify those attributes of people that seem to correlate with becoming a subscriber and those that tend to correlate with them leaving the company.

While such indicators are not based on definite causation, Wiggins acknowledged, they can still offer valuable hints concerning how the business can move ahead.

“We work hard to find ways of representing user behaviors in ways that may be suggestive to our friends in marketing or product [teams] about things they might want to do to change the product,” he said.

Sourced from WARC