Bloomberg Media chief operating officer Julia Beizer has explained how the business has used AI to create “synthetic characters” that it can test products and marketing messages on.
Speaking at Press Gazette’s Media Strategy Network event, held at the People Inc building in New York, she also explained how the publisher has grown revenue by 7% this year in the face of falling referral traffic from Google.
Bloomberg Media has added more than 100,000 subscribers over the last year and is nearing 700,000 paying readers and viewers for its consumer media business (as distinct from Bloomberg Terminal).
Beizer told Press Gazette that, in common with other publishers, Bloomberg has seen a fall off in Google search traffic since the rollout of AI summaries.
She said: “I think the scale era for premium journalism has been over for a long time.
“We’ve been saying to ourselves, maybe that traffic we were getting before wasn’t real.
“These weren’t real audience connections. It was a great way to meet people, but it wasn’t a way to really drive a durable business like we’re seeing with subscriptions.
“So we refocused the teams on the kind of metrics that matter, not scale for scale’s sake, but how are we driving engagement?
“We’re going to build a stronger business with people who know our brand, get value out of our content and keep coming back with regularity. But I think it is a seismic shift for the industry, and I think it’s going to mean changes in how we allocate capital.”
She said that Bloomberg is now more focused, for example, on more traditional forms of paid brand marketing than it had been.
Asked about the tactics and strategies which are driving subscriptions growth, she said: “I think it really comes down to fundamentals, taking a step back to make sure we have the right product.
“And that came from investing in content that goes beyond our day to day coverage of the news, like lean back ideas and culture coverage that has really rounded out our offering. And then thinking about the user experience. So we’ve invested in some stickier parts of our product.
“We’ve improved our watch list [a feature which lets subscribers track stocks, funds and other entities], which is something that’s very valuable to subscribers. We’ve improved our app and completely overhauled it, including creating a ‘for you’ tab that really leans into personalisation, which is basically a customer expectation at this point, particularly in our space.
“And we’ve experimented with games, which, as you know, is pretty hot right now.”
Beizer said that by shortening the discounted trial period, readers are staying with Bloomberg longer. Earlier this year it changed from offering readers three months for $2 to one month for $2, before reverting to full price of $400 per year (paid monthly).
“I think what accounts for that is basically audiences said, ‘I’m not going to pay for this product unless I really want it, unless I believe it’s worth that much’. And I think those are the kinds of audiences we want – people who are going to pay and stay.”
Bloomberg has yet to announce a licensing deal with any of the large language models.
But Beizer revealed that the publisher makes extensive use OpenAI’s custom ChatGPT technology in-house.
“We use AI in three ways. Number one is to help our customers, and that’s the thing we find most exciting and interesting.
“We’ve added AI summaries on to our articles. We’ve used AI to improve our search engine, and that has led to 15% click through improvement.
“We also use AI to help our business. So we’ve used AI and LLM technology to build out our contextual targeting [advertising] offering, which has been really important to our video revenue growth.
“Using this technology to understand contextually what’s happening in a video, and provide that alignment for our marketing partners.
“And then third way we use AI is internal workflows.”
On the newsroom side this has involved publishing automated stories based on formulaic company announcements since 2016.
But on the product development side, Bloomberg has created something called “segmentation GPT” to enable it to quickly test and iterate using “synthetic personas”.
“We’ve taken our customer segmentation, so every way we understand our users, and we’ve created custom GPTs for each one.
“So one of our customer segments, for example, is called the Market Mavens. We have released these tools to our editors, to our marketers and to our product people, and what they do with these tools is use these sort of synthetic research characters to help improve their work.
“So we submit, for example, a draft of a marketing email, and sort of interrogate the GPT to understand how that marketing email could be changed to better reach and connect with a Market Maven persona. This has been so fun and transformative for the staff because they’ve had essentially, like, a work buddy to help them figure out how to optimise to fit the needs of our users.
“I want to be clear that we haven’t given up talking to our users. In fact, we still do that all the time and have a monthly check-in when we look at all of our customer data.
“We’re really focused on talking to our users, but this custom GPT has really helped everyone around the organisation keep that subscriber in mind with everything they do. And I think that’s been really, really cool.”
Email pged@pressgazette.co.uk to point out mistakes, provide story tips or send in a letter for publication on our "Letters Page" blog
