By Seb Joseph  •  October 29, 2025  •

Ivy Liu

“We’ve come to the conclusion at the business that the biggest influence on retention is acquisition,” said Graham MacFadyen, consumer marketing director at Financial Times, at the Digiday Publishing Summit Europe, in Lisbon, Portugal.

The experiment that made it possible 

Since the paywall launched in January, the Financial Times has seen conversion rates jump 290% and lifetime value rise between 7% and 10% among the audience segments exposed to the AI-driven system. The publisher isn’t sharing how many readers that actually covers, but internally, the results have been strong enough to turn the experiment into a focal point.

“We feel like the AI is doing a really good job,” said Graham MacFadyen, consumer marketing director at Financial Times, at the Digiday Publishing Summit Europe, in Lisbon, Portugal.

Still, he was careful not to declare victory just yet.

The AI paywall only applies to around 30 to 40% of the publisher’s audience — the readers who’ve explicitly consented to data tracking. That means the model could simply be dealing with readers who were more likely to subscribe anyway.

Or as MacFadyen put it: “We suspect there’s an intent bias in the sample.” 

To find out, he and his team are testing control groups of consented users who don’t see the AI paywall. The goal is to isolate what’s truly incremental, and to sprite the value of the technology from the audience that makes it work. 

How it learns 

The Financial Times AI paywall doesn’t just decide when to show a paywall. It decides what kind of offer to show. Drawing on behavioral signals like visit frequency, time of day and content type, the model predicts a reader’s willingness to pay, then picks from a menu of offers ranging from the £4.99 a month FT Edit app to the paper’s £50-a-month premium tier. 

Unlike earlier versions of the publisher’s dynamic paywall, which operated on simple if/then rules (“show offer after five articles”, for instance), the AI model is probabilistic. It learns as it goes. For readers with sparse data, like say, a first time visitor, it uses lookalikes from previous users to make its best guess. 

Over time, the AI paywall has gradually tightened access, showing fewer free articles and increasing the paywall’s block rate. In other words, a higher share of readers now encounter the paywall and a growing number are choosing to subscribe. 

Rethinking the funnel 

MacFadyen frames the Financial Times’ funnel as a “martini glass” — wide at the top, narrow at the bottom, with most readers walking away before subscribing. The AI paywall is an attempt to reshape that glass by nudging users into the right product earlier, even if that means starting small with a trial, registration or newsletter signup.

Beyond data science, the broader goal is to widen the Financial Times’ appeal. MacFadyen said the publisher is using AI to show the right mix of content and products for different reader types, from students to senior executives, as part of a larger push to make the title feel relevant to a broader audience, not just incense insiders.

What comes next

Inside the Financial Times, the tech isn’t the sticking point. The harder part is deciding how — and when — to merge the AI-driven paywall with the dynamic one, and how much of what’s learned from the consented cohort can responsibly scale to the wider audience.

“There are issues of compliance to think through because one is based on consented users and the other is not,” said MacFadyen. “But yes, I see a convergence of the two over time — or at the very least in the short term knowledge sharing across the two.”

Until then, the team is studying how the model performs across content types and regions, looking at whether readers convert differently depending on what they read — marketers versus politics, for instance — and feeding those insights back into the broader subscription strategy.