The WSJ Dynamic Paywall
Media associated with this campaign
Overview of this campaign
Broadly speaking, we wanted to create a new paywall that met three key objectives…
- The capacity to deal with 15m visitors a week
- An experience that is dynamic and could flex depending on the users likelihood to subscribe – moving away from giving everyone the same experience, e.g. 5 articles free per month
- Increase subscription sales without impacting advertising capacity
Teach a model to learn: To make this ambition a reality we had the objective to first create a propensity model that could predict someone’s likelihood of subscribing. To do this, our data science team started by reviewing data from thousands of customer records with the aim to understand which signals were consistently appearing. For example, the device type the user subscribed on, the number of prior visits to WSJ.com, and data we could infer based on location e.g. average income of the users zip code.
Creating a Dynamic Experience: The next objective was to create a tech solution that in real-time could ingest a propensity score, and in turn serve a specific experience based on the score. In our case determining whether the paywall would be open or closed.
Optimize subscription and advertising revenue: WSJ has a duel revenue stream and we needed to balance the need to maintain capacity onsite to fulfill our advertising needs whilst ensuring we didn’t cannibalize subscriptions by having an overly porous paywall.
Extend the capability to referral channels: Traffic to WSJ.com from Google (hard paywall) and Facebook (first click free) were one dimensional, treating all visitors the same. The new paywall needed to increase sales from these referral channels.
Results for this campaign
Understanding the signals that influence purchase: We now use over 60 variables, both first party and third party data, which are leading indicators as to whether a user will subscribe. Fig 1.
Teach a model to learn: Using machine learning we created a model that is now over 90% accurate at predicting whether a visitor to WSJ will subscribe. See Figs 2 - 3.
Creating a Dynamic Experience: The Dynamic Paywall classifies visitors to WSJ.com into 3 groups based on their likelihood to subscribe (cold, warm, and hot). This classification leads to the paywall experience they receive. In short, we sample content to people less likely to subscribe, while people more likely to subscribe will hit a hard paywall. We also use a 7-day guest pass to help increase the likelihood of someone subscribing. Fig 4.
Optimize subscription and advertising revenue. In November WSJ advertising business was 100% sold through. To sell more we needed more capacity by putting more content in front of the paywall. Previously this would have had a negative impact on subscriptions, but with the use of the Dynamic Paywall, we were able open up the paywall only to people who had a low likelihood to subscribe. We found that opening up the paywall and exposing this group to more WSJ content increased their likelihood of subscribing in future weeks. See fig 5. Sales increased by 2% (looking through a 3 week attribution window). Fig 6
Extend the benefit to referral channels: This Dynamic Paywall logic has now also been applied to traffic arriving to WSJ.com from Facebook and Google.
Since the roll out of the Dynamic Paywall, conversion rates have increased considerably (see Fig 7) leading to double-digit subscriber growth, and helping WSJ to grow to almost 2.5m members, the highest number of subscribers in the history of our publication. Fig 8