Media Briefing: The Financial Times’ AI paywall is improving subscriber metrics, but not lifting conversions yet

The Media Briefing this week features an interview with the Financial Times’ Fiona Spooner, managing director of the FT’s Consumer Revenue Group, on how their AI-powered paywall has led to an increase in key subscription metrics like average revenue per user and lifetime value — even though it hasn’t led to more conversions yet.

• How the FT is using AI technology to boost its paywall and improve ARPU and LTV
The Wall Street Journal’s newsroom restructuring, ABC News’ union win protecting jobs against generative AI and more.

FT’s smarter, more automated paywall

The Financial Times’s months-old AI-powered paywall has helped improve key subscription business metrics, such as average revenue per user and lifetime value, according to Fiona Spooner, Managing Director of the FT’s Consumer Revenue Group.

But it hasn’t led to more readers converting into subscribers. In fact, the conversion rate dropped 10% since the paywall has rolled out in the past year, according to Spooner. However, she’s not concerned.

“That sounds like it’s a negative, but actually… it shows that [the paywall is] really effective, and meeting that goal of increased value over volume. So even though we might be choosing to sell less right now [as a result of prioritizing LTV], having that increased value is what will set us up for that long-term success,” Spooner said. The FT presently has 1.4 million paid subscribers.

The AI-powered paywall has shown to improve the FT’s ARPU by 6% year over year, according to Spooner. The new model is better than the old paywall at identifying users who are most likely to purchase higher-value products, such as the FT’s premium tier subscription, which costs $75 a month, according to Spooner.

The paywall went live to 95% of the FT’s global readers that have opted in to sharing their data or registered on its site (5% are part of the “control group,” Spooner said) as of the first week of January, following months of testing with a smaller group. The paywall was developed in collaboration with AI firm Sub(x).

The paywall determines the right number of free articles to show each user and the best subscription offer to surface. Spooner declined to share the number of articles readers can access for free. The model can analyze about 50 user data points, such as time of day, location, industry, job seniority, engagement and recency, frequency and volume (RFV) — all while factoring in the user’s likelihood to subscribe and potential LTV. 

While the top tier premium subscription offering is not discounted, discounts for its standard subscription tier range from 10% to 50% (the latter for an annual subscription). A premium subscription currently costs $75 a month, while a standard subscription costs $540 a year (the latter was one was on sale for $319 when this reporter checked yesterday). The paywall model does not have the ability to change subscription prices.

However, the paywall can help retain a subscriber at risk of churning by offering a lower tier product, Spooner said.

The FT can also now promote products beyond the core offering on the paywall, including free registration, newsletters and FT Edit, a curated selection of the publisher’s coverage. Previously, this was all done with manual rules set by the FT’s team (which one?), with only a subscription trial and standard annual offer.

“As a result, many prospective customers were presented with identical offers,” Spooner said. Not to mention it was a “time-consuming and resource-intensive” process, she added.

The new paywall has also led to a 100% increase in customers saved at the point of cancellation, representing an uplift in retention for online trial cancellations, according to the FT. 

Spooner explained how the AI paywall works and what FT is hoping to do next with the technology.

This conversation has been edited and condensed.

How long have you been working on this?

Longer than I would have wanted us to be. I think we’ve been probably working on it for nearly two years … probably really in earnest, the past year or so. It’s definitely not been as quick as we would have wanted it to be, but that’s because it’s been really important to me that we do it in the right way, that it’s not a rushed thing. … It’s not about, let’s do something that increases conversion tomorrow. Let’s do the right thing that builds the best experience and we can learn as much as possible. That really drives long term growth.

What exactly is the AI part of the paywall now?

[Our partnership with] Sub(x) incorporates a whole range of features, using data techniques and elements of AI and machine learning to optimize our subscription models and the offers that we put in front of people. The AI part is both the when and the what. So looking at when — how many free articles a user can access, for example, and when we put that barrier up and ask people to subscribe or to register, and basing that very much on behavior and engagement patterns. And then the other part is the what — what offer, or what combination of offers, to put in front of people based on that user’s behavior and likely intent. So we look at identifying patterns [such as] clicks through the paywall, conversions, revenue — and over time we can segment those users and look at what we present to people, what offer we put in front of them.

How is that different from a dynamic paywall that uses machine learning to determine different subscription offers?

A regular, dynamic paywall follows set rules, whereas our AI-powered paywall rewrites them in real time. … There’s a lot more data behind it. … [It’s] continuously learning, what is the right offer to put in front of those right people. As we’ve grown our portfolio, [that’s] become more complicated. Using this and the AI aspect of it looks at what products and what offers to put in front of the right people.

What are some of those rules?

We have a bundle offer for print and digital, but we know that we can’t fulfill the print product in some U.S. states. So we don’t want to sell a newspaper or a bundle [in certain states], even though that’s our highest price subscription. So just being able to not have that served to those customers is an important one. [Another rule is] if someone’s a lapsed subscriber, they don’t need a long trial [offer] because they’ve already had a previous subscription. [Those rules make sure we] don’t serve conflicting messages to people, because the important thing is that we show that we know who they are, without getting over-personalized.

Does the AI-powered paywall have the ability to offer different subscription prices to different people?

At the moment, the prices are set. There’s no variable on the set prices for each product. Because to me, it’s really important that we’re transparent so we’re not charging someone a higher price for the same thing as someone else, based on their behavior. 

How is the user data collected?

We make sure that we only personalize for users who have opted in. And then some of the other engagement metrics that we use — like recency, frequency, volume — we know whether or not someone’s a lapsed subscriber or registered because of a previous account with us.

What have the results been so far?

There’s been about an 80% increase in the number of paywalls shown. So that’s showing that we’re engaging more users in their journey. We have seen a 17% increase in the number of users progressing through the paywall. And a 6% improvement in lifetime value, which is what we’ve been prioritizing. 

At the moment, there’s no detrimental impact to user engagement levels or advertising [revenue], which is also really important. And despite that drop in conversion, because all of the other metrics are up, it gives us the confidence to carry on and that we’re doing the right thing.

Any impact to other metrics, like churn or bounce rate?

No, at the moment our churn rate is holding really strong. We are continually improving that. 

So this hasn’t resulted in more new subscribers?

No, I think because we’ve been prioritizing value experience. We’ve not been optimizing it for volume… We’re focused on delivering our revenue targets. I don’t think the FT has ever been a volume play, but that doesn’t mean that we don’t want to increase our audience. We absolutely want our audience to be as big as our brand. But it’s about getting the right people and building a sustainable subscriptions business. So I think if we’re providing the value for our customers, the volume will follow. 

You mentioned previously showing personalized offers when testing this paywall. Can you give an example of what that might look like?

An easy example is a student. We offer them a four-week trial, then we’ll proactively send them a student offer to roll onto. If we know that they are C-suite subscribers, then we will talk to them about a premium subscription. 

Is that something that you weren’t doing before with the paywall?

Not in the same way. We would do it manually, through segmented email campaigns, for example, but that would take us a long time.

What are you working on next?

We are looking at continuing to build it out across retention — being more proactive than reactive and adding in more customized offers. At the moment, we have our core suite of products in the standard premium registration trial. We’re looking at building out different products and newsletters, corporate subscriptions and our events business offers.

We offer monthly and annual subscriptions and trials. At the moment, we’re looking at whether or not we include different trial durations.

What we’ve heard

“This space is evolving rapidly, and the distinctions between models for our use cases are becoming increasingly subtle.”

— Nate Landau, chief product and technology officer at TheSkimm, on evaluating which large language models to use for different applications.

Numbers to know

£200 million: LADBible Group’s global revenue target, from £90 million today, according to CEO Solly Solomou.

20%: The percentage of subscribers The Washington Post won back from those it had lost after choosing not to endorse a presidential candidate.

210: The number of staff laid off at CNN, representing 6% of its total staff, as the company focuses on its digital products.

40: The number of employees fired at NBC News.

What we’ve covered

How Perplexity’s one-person team is working to build relationships with publishers

• While Jessica Chan has been in the role for less than six months, she’s well-equipped to lure publishers into Perplexity’s ad revenue share program, given her experience and the hard and soft skills she’s developed during her career, according to conversations with her boss, former colleagues and publishing executives.
• Publishing execs that have worked with her seem impressed by her personality and intelligence so far — a good place to start, in an ecosystem where relationships between tech companies and publishers can often turn cold.

Read more about how Chan is tackling a challenging role here.

What Forbes, Dotdash Meredith, BuzzFeed and other publishers are saying about AI in 2025

• Executives at Business Insider, BuzzFeed, Dotdash Meredith, Dow Jones, Forbes shared how they’re currently using AI at their respective companies, and where they will invest in the technology in the future.
• Those publishers are using AI technology to for on-site search functions, advertising targeting, article summarizations and image generation.

Read more about what those publishers are doing with AI this year in the Digiday+ Research report here.

Some creators are looking for alternatives to TikTok’s editing app CapCut

• After TikTok and its sister app CapCut were shutdown, some creators are looking for alternatives to the video editing app.
• However, other creators said the app’s strengths make it hard to give up.

Read more about what other apps creators are testing for video editing here.

What we’re reading

The Wall Street Journal undergoes newsroom restructuring

The Wall Street Journal’s editor-in-chief Emma Tucker announced a newsroom restructuring to streamline verticals, and that some bureau chiefs who were part of the Life & Work coverage area would be let go, according to a memo shared by New York Times reporter Katie Robertson. Life & Work will no longer be a standalone beat and will now be part of a “Corporate” coverage group that also includes manufacturing, autos, aviation, retail, food and agriculture. The Journal is also creating a new tech & media reporting team.

ABC News newsroom union wins new protections against generative AI 

ABC News‘ unionized writers have voted to ratify a three-year contract that offers some new employment protections if the newsroom adopts the use of generative AI, according to The Hollywood Reporter. In part, it protects their jobs from being eliminated due to the technology being implemented in their workplace.

Quartz is publishing AI-generated articles

Quartz has begun publishing short articles generated by AI technology by aggregating coverage from other outlets, TechCrunch reported. The 400-word stories are published under the byline “Quartz Intelligence Newsroom.” A Quartz spokesperson said the goal is to give Quartz’s editorial staff the time to work on longer and more deeply-reported stories. 

White House reserves two seats in press briefing room for “new media”

White House press secretary Karoline Leavitt said she’s keeping two seats open in the press briefing room to “new media voices,” such as “independent journalists, podcasters, social media influencers and content creators,” Axios reported. The first two occupants were Mike Allen of Axios and Matt Boyle of Breitbart.

AI journalist Karen Hao on what DeepSeek may mean for the future of journalism
Chinese company DeepSeek’s AI Assistant is on par with — and even surpassed — the performance of competitors like OpenAI’s ChatGPT in some cases, and at a fraction of the cost, Hao said in an interview with the Reuters Institute. The fact that the model is open-source means it may be safer to use, but there are still questions that need to be answered around data privacy and copyright infringement in the industry, Hao said.