Uh-oh, Reuters has a robot writer who can churn out earnings reports

Extra-extra, the robots are coming.

Two dozen Reuters journalists have tested the tool Reuters calls Lynx Insight that can write two-thirds of a story based on company earnings reports. But at least for now, robots aren’t replacing journalists. Lynx alerts reporters through email, messenger service or via reporters’ data terminals when it recognizes trends and anomalies in data. Reporters can also input a company or search term into the tool to surface insights. Market data is the focus for now, but in time, Reuters will roll the tool out to other sectors.

“We’ve recognized that the key thing is not to make a machine write in-depth stories, but to marry the strength of machines with the strengths of humans,” said Reg Chua, executive editor of editorial operations, data and innovation for Reuters. “Machines are tireless at sifting through data and detecting patterns; humans exercise judgment, provide context and gather quotes.”

According to Chua, Reuters has another dozen people on the technology side setting up the platform, analyzing the data and generating the language recognition.

Reuters has 2,500 reporters, and while not all of them will need to use Lynx Insight, saving short periods of time at this scale will have an impact. The tool’s purpose is to uncover insights that would otherwise take a long time to surface, while improving accuracy and efficiency.

Elsewhere, The Associated Press estimates that it’s freed up 20 percent of reporters’ time spent covering corporate earnings through automation. According to Reuters’ annual survey of digital leaders, which included nearly 200 publishing executives, 91 percent of respondents cited production efficiency as a very or quite important priority this year. The Washington Post, meanwhile, found new audiences through its homegrown artificial intelligence technology.

But measuring the impact of automation in newsrooms is difficult. In the future, Reuters will develop a feedback loop so that the tool knows which of its insights the reporter used and improves their value.

“We’re not looking at the volume of stories as a success metric,” Chua said. “Telling you what information you can dismiss quickly is already valuable. We’re in a speed race to make sure we can monitor company results and economic indicators.”

A financial benefit will be using Lynx Insight to unlock potential products to sell to other news sources so that a standard weekly football report for a Manchester, England, news publisher, for instance, always leads with stories about Manchester United, he said.

“Reuters is known for being early on stories,” said Dr. Janet Bastiman, chief science officer at AI platform StoryStream. “In the rush to get there first, it will only become more important to uncover stories faster. Being able to personalize it for paying clients will be absolutely critical.”

“Machines are only as good as you train them. Incorporating a feedback loop will improve it,” she added, “but ensuring there are no biases built in takes a lot of careful testing.”

Another tool, Reuters News Tracer, which surfaces and assigns a value to content on Twitter based on how likely it is to be true, gives reporters a head start on breaking news. It has helped break 50 major news stories and given journalists anywhere between an eight- and 60-minute head start, according to the company.

To encourage time-poor reporters to use Lynx Insight, it needs to be as easy as possible to use. “I don’t want to have to train people to use the telephone,” Chua said. “It should be able to understand that you are already writing about a company like Microsoft to avoid culture clash in the long term.”



More in Media

News publishers hesitate to commit to investing more into Threads next year despite growing engagement

News publishers are cautious to pour more resources into Threads, as limited available data makes it difficult to determine whether investing more into the platform is worth it.

privacy sandbox

WTF is Google’s Protected Audience?

FLEDGE stands for ‘First Locally-Executed Decision over Groups Experiment’ and makes ad auction decisions in the browser, rather than at ad server level.

Digiday’s History of Ad Tech: In the beginning …

A look at the genesis of ad tech, from the first online display ad in 1994 to the dotcom crash.