Imagine sitting on your couch with your friends and playing a round of soccer on Playstation. You choose your player based on the strength stats. Christiano Ronaldo with his high shooting skills is perfect for scoring. Your friend chooses defender Sergio Ramos to prevent you from scoring. Each player has unique strengths.

Based on our data analysis, we know that written articles have unique strengths as well. Like becoming a world-class soccer player, it is hard to write good and well-rounded articles. Writing those pieces requires a lot of editorial experience.

To help our editors, we are building a data-driven coach to develop the unique strengths of each article.

Each article has certain strengths. Our attackers are articles creating reach and subscriptions. Our defenders are bringing subscribers back and the midfielders are engaging them. Like a soccer team, we need the right mix of articles for our business success. A sole focus on subscription-driving articles can alienate loyal customers. Focusing on reach articles drives ad revenues but the subscription conversions might suffer.

Creating a winning strategy

Thus, we decided to focus on a 1+ strategy to make every average article a bit better and grow as a whole. To achieve this, we created a tool for the editors and site managers.

The CREAM Tool coaches the editor based on the following questions:

• Conversion: Does my article drive subscriptions?
• Reach: Does my article drive traffic for ad monetisation?
• Engagement: Does my article keep readers on the site?
• Anti-churn: Is my article popular among subscribers?
• Monetary value: How much revenue does it generate?

The tool shows each article’s strengths based on tracking metrics. The editor can compare his article to similar articles and continuously improve it. Data science models run in the background to find the advice with the highest impact. We are currently exploring which article features impact a certain strength and how to extend the article’s lifecycle.

For strategic content planning, we created the topic analyser. IAB classification is often indistinct for the local news market. So, we developed a tailor-made machine learning model. We trained the model on our existing content. It labels every new article with one of 14 custom topics that each have several subtopics.

A database stores each article’s content profile. The content profile contains metadata like page type, content state, and word count. The database links content profiles to traffic metrics.

This allows us to identify:

1. Which content is missing in our editorial portfolio?
2. How can we improve the paid-to-free content ratio?
3. How can we loyalise our readers and subscribers?

We found that the topic of sicherheit (safety) has a high subscriber retention and stickiness. The subtopics showed us, that within sicherheit, short police and accident reports are of high interest. Thus, we are discussing how to grow the topic cluster with similar content.

To identify the right amount of articles for each topic, we look for stagnation in the traffic metrics. We also found that Facebook referrers are driving subscriber retention within certain topics. Those insights allow us to improve the targeting of social media campaigns.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .