Still Not Using Machine Learning to Grow Subscriptions? It's Easier Than You Think

01 Nov 2018 | By Tobias Arns

Grow subscriptions with Machine Learning

The rise of Artificial Intelligence and more specifically, data science and Machine Learning, has drastically changed the way companies view their data. It’s no longer a byproduct, but a versatile resource that can generate revenue, solve tricky problems, and improve competitiveness.

Becoming data-driven is a game-changer for publishers

Ever since leading news sites like the Wall Street Journal or the New York Times have become serious about using data science and Machine Learning to grow subscription revenues, the approach is getting a lot of attention in the industry. And rightfully so - because there’s new technology available that enables publishers to predict with astounding precision the likelihood of a given reader becoming a paying subscriber. It adds intelligence to previously static and content-driven paywalls, making them dynamic and enabling them to adapt to what individual readers care about and are willing to pay for. A strong case for abandoning one-size-fits-all approaches once and for all.

Know your users' reading and spending habits better than they do

When you scrutinize your audience data you will most likely see that statistically, the people who are paying for your content have a lot in common. They show certain behavioural patterns on your site, they have certain reading habits and preferences that are specific to a publication. A Machine Learning-powered approach called “propensity modelling” takes advantage of this: It systematically analyzes dozens of factors that influence a reader on his or her journey to becoming a subscriber and steers a dynamic paywall in real-time based on this information. The first step is taking accumulated data from readers who have already subscribed and feeding it into a Machine Learning algorithm.

Customer Journey from Anonymous Reader to Subscriber

Customer journey from anonymous visitor to high-value subscriber

Then, based on the behavioral data from users that have already completed the customer journey, each non-subscribed reader is given a propensity score. As you probably guessed, the higher the score the more likely they are to become a subscriber. In a nutshell, propensity modelling allows you to optimize the conversion of potential subscribers based on what you learned by watching similar readers that converted already. These similarities can be monetized by personalizing the experience your prospects are getting in order to drive subscriptions. 

In order to justify being called Machine Learning, there needs to be some kind of learning involved, i.e. the model’s prediction quality has to improve over time. Therefore, the model is  continually trained and tested with fresh data and the score is updated in real-time every time someone revisits the site - the system learns while working with the data. And since each site has a unique audience and customer journey, the application of the model should automatically adapt to fit the needs of different publishers.

How does it work in practice?

In practice, groups of readers with similar scores can be treated just like audience segments. This means they can be targeted with different user experiences, personalized content, and individual messages in order to give each group the highest chance of conversion. Once the model is in place, the paywall can be shown to readers with a very high score right away because there is a high likelihood that they will subscribe soon. Readers with a medium score might get a few free articles that match their interests in order to keep them engaged. Low propensity readers might see mostly ad-financed, non-premium content.

Afraid this sounds too technical and difficult? It doesn't have to be. Make sure to parther with a vendor that has both the data science know-how to get you started and a solution that was developed with the practical needs of newsroom and business users in mind. 

Leading publishers see dynamic paywalls and propensity modelling as the future

Using Machine Learning to complement your newsroom’s experience about what readers want can be a decisive advantage for a publisher in the race to turn more readers into subscribers. The Wall Street Journal for example, with whom Cxense developed Conversion Engine in close cooperation, has been successfully using a propensity model for over a year now. Their model predicts with 90 percent accuracy whether a visitor to will subscribe. Due to this remarkable precision, the score plays an important role in the personalization strategy of and the behaviour of their dynamic paywall.