Bloomberg Media has a robot writing story summaries

robot reporter

This week, Bloomberg Media rolled out Bulletin, a newsfeed-style feature in its mobile app that gives users personalized stories every time they open the application. The stories’ one-sentence summaries are written by an artificial intelligence-powered program that scans the stories for meaning.

Bloomberg produces a torrent of news every day. Bulletin doesn’t scan opinions, editorials or special investigative stories, because the technology that reads the stories to summarize them is best at understanding documents that have similar structure, such as stories about company’s quarterly earnings, or the fluctuations in a commodity market. Other publishers such as Reuters and The Washington Post have done robot-written stories that rely on structured data.

For now, AI isn’t smart enough to understand the nuances of an argument, though the ability to understand less structured news stories could soon be upon us. “I think it’s realistic [for AI] to be able to summarize a standalone news story, accurately, imminently, within 12-24 months,” said Cennydd Bowles, a UX designer who’s worked with publishers including the BBC and the author of “Future Ethics.”

Bulletin came out of the BHIVE, a five-person research and development group inside Bloomberg Media that researches how people use the publisher’s products. It conducts two or three audience surveys per month, on subjects ranging from advertising to personalization, and uses them on product development.

“BHIVE is really designed to understand our users — what motivates our users, getting closer to their behavioral patterns, and understanding how our news fits into their day to day lives,” BHIVE head Ambika Nigam said. “[For Bulletin], the specific question we asked was, ‘How do our users digest the news?’”

The answer was, in small bursts, throughout the day, so the BHIVE team decided that the Bloomberg News app should be filled with new stories every time the user opened it. An algorithm picks stories based on recently, popularity, relation to market activity and what users have read in the past.

The goal of personalization isn’t to get rid of reporting staff, Nigam said. “We want this to be additive.”

Bloomberg Media’s head of product, Julia Beizer, identified personalization as a key priority for the publisher as it tries to drive reader loyalty and subscription revenue. Personalization technology has been useful to publishers trying to keep paying customers engaged, though publishers focused on ad revenue have found less to like about the technology.

In a preliminary test that showed Bulletin to a random 20 percent of app users, 26 percent of them refreshed the Bulletin feed at least once. Nigam said that the share of users that engage with Bulletin has been increasing steadily since the rollout, but wouldn’t share specifics.

Bulletin started on the mobile app because that audience is smaller and engages more content than the audiences Bloomberg Media has on other surfaces, such as desktop or mobile web. But over the next few months, the plan is to incorporate it into other Bloomberg News products, including desktop. Balancing the goal of super-serving readers with making sure they see the most important news of the day will be a priority.

More in Media

Can AI analyses about AI content reveal anything about AI and copyright?

Three AI companies share analyses about copyright concerns, websites blocking web crawlers and how much AI text includes protected content.

Research Briefing: Publishers bank on their own first-party data amid Privacy Sandbox concerns

In this week’s Digiday+ Research Briefing, we examine publishers’ reservations about Google’s Privacy Sandbox, how subscriptions aren’t the revenue driver they once were for publishers and how X is once again telling advertisers it’s serious about brand safety, as seen in recent data from Digiday+ Research.

How a revamped Green Media Product hopes to solve ‘problematic placements’

Scope3 unveils GMP+ with Sharethrough as debut partners to extend ‘Green PMPs.’