AI’s role in the news media industry continues to evolve, and media companies need to know how to leverage it. 

During Wednesday’s INMA members-only Webinar, How Advance Local used AI and data to accelerate their digital subscription business, Neil Katz, chief customer officer at Advance Local, and Sonali Verma, director of business development for Sophi.io, explained the implementation and measurement of the automated Sophi system at Advance Local.

Verma first gave an overview of the platform, developed by The Globe and Mail in Canada over the past 10 years and used by media companies on five continents. Sophi uses AI to predict, automate, and optimise “pretty much every problem that anyone in the news business could have.”

At the end of 2021, Advance Local started using the Sophi Content Paywall Engine in two of its markets — Cleveland, Ohio, and Michigan. After seeing better-than-expected results, Advance Local rolled out Sophi in its other eight markets.

 Advance Local, a U.S. company with more than 50 million readers across 10 markets, is moving toward a digital model and, of its 500,000 subscribers, about 26% are digital. “We still have more print subscriptions than digital, and we’re working to shift that over time,” Katz explained.

Over the past two years, Advance Local worked on developing a “smart funnel” by utilising data and machine learning at as many points along the funnel as it can. Katz noted two of the familiar challenges the company is trying to solve:

1. As print subscriptions decline and digital subscriptions increase, the company needs to manage the speed at which both those things are occurring.
2. Understanding what he called the “yin and yang of ads and subscriptions.” That requires determining if the page — and the reader — is best suited for advertising or for subscriptions.  

“At the enterprise level of our business, this is kind of where the fights happen,” he said, noting that both the ad department and subscription department are looking for a win. “The right way to sort it out is with data and technology rather than the loudest voice in the room and the most persuasive voice in the room.”

Preparing for the journey

When Advance Local started its journey with Sophi, it first needed to get its “data house” in order. It had installed its customer data platform two years earlier but spent months connecting all the data pipes. Next, it had to clean and analyse the data. “Then you need another set of months to start building your segments and testing your segments.”

Eventually, he said, all the pieces were in place and the company could “start doing smart and interesting things.” It built models for every stage of the funnel and evaluated which stages could be done with traditional actions and which would benefit from machine learning.

The Advance Local model relies on subscriber-exclusive content to drive people to subscribe and began using Sophi as a way to select that content. It quickly discovered some advantages to using robots:

1. Robots find stories that, theoretically, should not work. “Wire (stories), syndicated, Dear Abby … it doesn’t make any sense because you can get Dear Abby from almost any Web site — typically for free. So, it’s not rational … but it still works.” He said Sophi has a good track record for picking stories that are worth paywalling — and that human editors would not have included.
2. It finds stories that humans overlook. “We’re pretty good at picking up the big stuff that we know is going to work, but there’s lots of stuff we don’t think about.” For example, a story about a Michigan marijuana retailer closing four stores drove more than 100 new signups. “I’m not sure what he was smoking, but apparently the robot knows marijuana was a good bet,” Katz quipped.
3. Robots don’t sleep. “We publish 26,000 stories a month, and honestly, human editors weren’t paying the same level of attention to each one in terms of trying to evaluate if it was a great subscription story or advertising story — but robots will work tirelessly to do that.”

What the testing found

In the two test markets Advance Local began with, it found a 20% improvement in one and a 45% improvement in the second market when it came to conversion rates using robots. The platform provides the kind of testing flexibility it previously lacked to learn things like the percentage of stories or pageviews that should be gated, Katz said.

“We have a way of testing that now,” he said. But, he added, there are systems in place to allow for human input — for example, if a market is more subscription-focused or another is more advertising-focused, the algorithm can be adjusted to better suit the market.

“It’s a bit of human engineers and … a live piece of software that’s always learning from itself that we think is such a strong combination,” he said.

Based on the incredible test results, Katz said Advance Local was “unbelievably excited to install the technology across the entire company,” but that rollout happened at the beginning of summer, which meant it suffered from the anticipated summer slump. And rising inflation is driving up churn and driving down new subscription starts.   

“I think if I would have done this differently, I would have A/B tested in every market and proven to every market that this is working really good,” he said. As it stands now, some markets are in favour of Sophi, while others doubt its effectiveness.

Understanding what technology can (and can’t) do

While AI has benefitted Advance Local in boosting subscriptions and reducing churn, Katz cautioned Webinar attendees not to get so enamoured with digital discoveries that they lose sight of good old-fashioned analog signals that indicate customers are about to churn: “Sometimes old school signals are totally fine, completely effective, and actually faster to implement.”

He pointed to indicators such as seasonal subscribers, customers who have turned off the auto-renew feature, readers with low levels of engagement, or those who have frequent technical issues as evidenced by multiple password resets and calls to the call center.

“Password issues are our No. 1 technical problem and people’s reason for quitting,” he said, adding that the leading reason for leaving is price. “We’re now trying to figure out how to get this password reset signal back into the CDP or into a churn model so we can use it more effectively.” 

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .