Subscription propensity prediction for dynamic paywalls

Dynamic paywalls let publishers balance two business KPIs: maximizing paying subscriptions while minimizing pageview loss. Marfeel’s SDK provides a predict method that returns the probability a reader will subscribe if the current page view is dynamically closed.

Propensity to Subscribe needs to be activated before being available for use. You can request its activation to your Account Manager. Two months of accurately tracking pages closed and subscriptions will be needed before the algorithm can be properly trained.

Usage

window.marfeel.cmd.push(['compass', (compass) => {
    compass.predict().then(result => {

        /*
          result will be an object like
          { probability: 0.01234543344 }

          from here on, use it as needed.
          examples:
        */

          myVar = result;

          myPrediction = result.probability;

          console.log(result);
    });
}])

Threshold decision

The method returns a probability between 0 and 1. Publishers and developers decide the threshold from which it makes sense to close pages. This depends on each publisher’s strategy, but research suggests that a threshold of 0.5% (0.005) yields a Pareto result: retaining 80% of subscriptions while losing only 20% of the traffic, compared to simpler dynamic paywall strategies.

Different thresholds can apply depending on section, content type, content age, user profile, or any other variable that aligns with the publisher’s priorities.

Performance and tracking

A generic probability is always returned. For optimal performance, make sure you set a proper instrumentation of the pages closed with paywall and subscriptions. This way the prediction algorithm learns how users behave on your site specifically.

The returned probability is automatically logged as a pageVar and is accessible in raw data. Compass for Subscription Managers surfaces these KPIs for evaluating dynamic paywall performance:

• % of users that subscribe when finding a dynamic paywall
• Comparison of actual percentage vs predicted probability
• Average predicted probabilities by dimension

What probability threshold should I use for a dynamic paywall?

Research suggests a threshold of 0.5% (0.005) yields a Pareto result: retaining 80% of subscriptions while losing only 20% of traffic compared to simpler dynamic paywall strategies. You can also apply different thresholds by section, content type, content age, or user profile.

How do I call the propensity to subscribe prediction in the Marfeel SDK?

Call compass.predict() inside the Marfeel SDK command queue. The method returns a promise that resolves to an object containing a probability value between 0 and 1, representing the likelihood that the reader will subscribe if the current page view is closed.

What setup is needed before propensity to subscribe works?

Propensity to Subscribe must be activated by your Account Manager. The algorithm also requires two months of accurately tracking pages closed and subscriptions before it can be properly trained.