ALERT: INMA Global Media Awards recipients unveiled

Globe and Mail uses AI to drive revenue strategies

By Shelley Seale


Austin, Texas, United States


What really happens when news media publishers use Artificial Intelligence (AI) to drive their revenue strategies?

At The Globe and Mail in Canada, a team has been using AI to drive revenue for more than a decade. The Globe launched start-up in 2019 to bring some of those AI tools to market to provide solutions for other publishers for paywalls, automated content, and automated newspaper design.

The Globe and Mail embarked on its journey to create a next generation paywall five years ago.
The Globe and Mail embarked on its journey to create a next generation paywall five years ago.

“That really opened the door to a lot more conversations about the way publishers are thinking about AI,” said Gordon Edall, co-founder and chief revenue officer.

Paywall early adopters

The Globe and Mail was an early implementer of the paywall model, establishing a freemium paywall back in 1998. At that time, there wasn’t a lot of great technology to use for this and it affected ad revenue, so the company took the paywall down. In 2012, it put a new paywall strategy back in place, which was a combination of a hard and metered paywall.

 At that point, no other media house in Canada had a paywall. The Globe knew it had to continue investing in first-rate journalism and user experience, as well as protecting ad revenue.

The Globe and Mail's Gordon Edall explained what the company did right and what it did wrong whilst implementing its paywall.
The Globe and Mail's Gordon Edall explained what the company did right and what it did wrong whilst implementing its paywall.

 Edall shared what the Globe did right and wrong at that time.

 Good moves:

  • The paywall was opaque to the reader.

  • There was space for experimentation and optimisation.

 Bad moves:

  • Metre indexes.

  • Communicating metre counts to the reader.

  • Only listening to the highest-paid person.

“We were trying to pin down a business model — which metre mix is actually the best for us?” he said. “We found that the initial set of opaque rules, combined with experimentation, left us with a giant optimisation space that was very lucrative if we played with it.”

The next-generation paywall

Edall and his team realised they needed to go much further in creating the best paywall model. They started looking at what was happening on the cutting-edge of machine learning, such as in video games, and applying some of that learning to solving the Globe’s paywall problems.

To build the next-generation paywall, the team knew it had to build something different, with certain features.

 It needed a fully dynamic, real-time, personalised paywall with these specific features:

  • A deep-learning model that would learn from mistakes and know when to give up.

  • There should be no metre count at all, and no metre calendar resets.

  • The paywall should work even on a user’s first encounter.

This model works so well, Edall said, that it will often ask a user for a credit card on their first visit — and get it.

“What we’re essentially doing is playing a video game of sorts. It’s spread across hundreds of thousands of users and millions of pageviews, and the game is really quite simple,” he explained. “Every day, we try to set a new high score for the publication. How you score points in the game is by looking at each and every user, and each and every user interaction, and figuring out every time someone tries to read something.”

The decision then is whether the publisher can score more points by letting the user read the content they are trying to access, by asking them to register their e-mail, or even subscribe and pay.

Each of those outcomes is worth something different, but different users are likely to do different things. The key is figuring out propensity and likeliness for certain actions, to determine the “best win” in each case for the publisher.

“When the model makes a mistake and fails, the beauty of this approach is it actually learns from that,” Edall said. “That helps it make better decisions the next time for the same reader, and it lets the system make better bets at scale, too.”

He said the results from this model are impressive.

  • 220% increase in registrations.

  • 51% increase in subscription conversion.

  • 22% increase in registered visitor engagement.

  • 53% increase in loyalty.

The results from using Sophi have been impressive.
The results from using Sophi have been impressive.

Edall said the increase in anonymous user engagement and return rates was a surprising result, and he attributed it to the fact that the model knows when it’s time to give up.

“It knows when not to bug someone because it’s not going to get value directly from them, anyway. And that’s a better user experience that leads to return visits, and higher engagement, and more ad revenue — and eventually, more subscription revenue we think, as well.”

About Shelley Seale

By continuing to browse or by clicking “ACCEPT,” you agree to the storing of cookies on your device to enhance your site experience. To learn more about how we use cookies, please see our privacy policy.