Data-driven Product Development: Creating audience-first best practices Data-driven products for subscriber acquisition

By Scott M. Scher, Columbia University

At Tribune Interactive, stakeholders are more comfortable with analytics than algorithms, and that there still needs to be a human element for a level of trust to remain with clients according to VP for data, Alejandro Canterero. Data scientists are a very precious resources.

For Canterero and his team, the average time for integrating machine learning models is about three months. When looking to integrate a new AI system they must give themselves the right amount of time to do so. This is your standard pull data, clean data, test models, find the one that works, and validate against historical models. These models are designed not to automate but to augment product development jobs. The next stage is to get into higher level automation. All of this requires building trust with the rest of the organization.

The revenue form their website is two fold, digital advertising revenue and paid subscribers. They have been running on a traditional paywall system tied to the calendar month, where a visitor will see five free articles before being asked to subscribe. The data team is looking at how to effectively convert audience to paid subscriptions. They started asking the question, “which content matters?” They realized that not all content is created equally, when trying to build the paywall. Customers who only want to read articles about the Kardashian’s the result is significantly lower then those who are reading editorials or local election results. They have found that these different types of users have different behavior and react different to the five free articles before seeing the paywall.

From their analysis they validated their intuition that level of user engagement on the site is important. An obvious question for tronc was, “is the user local?” Since all of their papers are local papers it is more likely for local users to become paid subscribers. They found that the highest rate of conversion came from users engaging with content related to food, their local news, opinion, politics, and sports. Creatively turned into the acronym FLOPS, because who doesn’t love a good acronym.

Alejandro left the audience with these key lessons: sometimes intuition is wrong and features you think are important are not reflected in the data, build data services and integrate them into self-service portals, empower end-users to develop and mange tests, and allow users to pick and choose from different services and combine results.