How AI can help predict subscription cancellation and keep readers engaged

Credit: Jonathan Francisca on Unsplash

As a growing number of publishers are looking at subscriptions as their main source of revenue in 2020, the ability to convert a fly-by reader into a loyal subscriber becomes increasingly important.

A lot has changed since the days of blanket paywalls and many news organisation now opt for what became known as 'dynamic paywall'.

The main difference is that a dynamic paywall shows the reader a subscription offer when certain conditions are met, for example, when their engagement reaches a certain level and when they have come in from a specific device. However, an even deeper layer of understanding of reader behaviour can help your newsroom not only convert the reader into subscriber, but also keep them engaged and motivated to renew their support.

Most paywall providers now offer publishers the possibility to set up bespoke access. But to understand and leverage reader engagement, tracking and analysing real-time data can be a game-changer.

This, at least, is what Jarosław Góra, co-founder of Deep BI, tries to help publishers with. The basic thought is quite simple: the AI tools look at reader engagement, monitor any changes and automate paywalls in real-time to convert readers to subscribers. These automated decisions are based on understanding what kind of content attracts the reader and predicting when they are likely to cancel their subscription in the future.

"We are using a machine learning-based propensity score that looks at recency, frequency and volume, as well as up to 100 other indicators," said Góra, explaining that volume means both the number of articles read and the time spent engaging with the content. 

This data then helps build reader profiles and shows the paywall or newsletter sign-up forms at precisely the crucial moment to give the highest probability of conversion.

It essentially uses the principle of impulse purchase: that moment when you need or want something and it suddenly becomes available. You are more likely to buy it there and then, as opposed to considering a range of options.

Deep BI tracks and evaluates reader data over one week, one month and six months. It can spot the moment where reader preferences change, which informs the propensity to register/subscribe/churn score.

"We give publishers a tool that helps them plan editorial in a more efficient way," said Góra.

"We identify what made the user purchase the product and score it to capture the major events: registration, newsletter signup, subscription, renewal."

Looking at patterns rules out anomalies, known as accidental events. For example, it would exclude a reader who spends less than 15 seconds on an article.

This way, the algorithm can divide information into two groups: what helps build a relationship with a subscriber and accidental events.

Deep BI has also built another score which attributes subscriptions purchased to different types of content, allowing publishers to attribute even free content which was on a user’s path to subscription.

From there, the content is assigned into one of the three categories: acquisition content (what engaged the reader with your brand), conversion content (what made them subscribe) and maintenance content (what makes them stay, engage with your articles and renew their subscription).

This information can then be used to examine topics, sections and authors, which helps inform editors about audience engagement.

"Sometimes, some authors are brilliant in one section and fall flat in another," said Góra.

While around 70 per cent of news content is predictable or dictated by the news agenda, the rest is accidental and can be explored creatively. When reader engagement information is available to editors, they can then make better decisions when assigning stories to reporters or picking them for the section they convert best on.

Another advantage of using this information is that publishers can exclude content that is not likely to convert users from the dynamic paywall and keep those articles in a section open to all users regardless of engagement level, for example, content from newswires.

Timing is also important as sometimes editors need to decide when is the right time to publish a specific piece of content in a particular section for a maximum effect.

Deep BI then merges the scores and compares them to real-time analytics, providing the editors with a good idea about content performance.

But can all this algorithmic wisdom sway us in the wrong, computer-dictated direction?

"We give the information to the publisher at the beginning of the subscription and inform them when a subscriber is likely to churn," said Góra.

"It is up to the publishers to make the final decision about the way they manage their content."

So while the AI is good for processing a large volume of data, for example, telling you whether your reader is a weekend or weekday user, the ultimate success of the editorial strategy still lies with humans. For now.

Do you want to explore how AI can benefit your newsroom? Join us at Newsrewired on 4 June 2020 and learn from Charlie Beckett, founding director of Polis at the London School of Economics. Click here for more information.

Free daily newsletter

If you like our news and feature articles, you can sign up to receive our free daily (Mon-Fri) email newsletter (mobile friendly).