How Digitalhaus Franken tackles churn using AI

Find out how a German publisher uses machine learning to predict the cancellation behaviour of subscribers. The aim is to counteract this at an early stage with a personalised customer approach in order to reduce the cancellation rate in the long term

Image generated using artificial intelligence

By: Moritz Kircher

Subscription cancellations present significant challenges for many digital media companies. For us at Digitalhaus Franken, this is not only about economic stability but also about our mission to provide people in our region with well-founded information, contributing to the democratic opinion-forming process. To significantly reduce churn rate, we are implementing an AI-based prediction system PULSE – Predictive User Loyalty System for Engagement.  PULSE is the name of the project through which Digitalhaus Franken is developing its own AI model that will help us calculate individual cancellation probabilities for each of our subscribers in the future. We are convinced that well-informed individuals can better participate in societal discussions and make well-founded decisions.

What we have achieved so far

The selected model aims to calculate a cancellation probability for each subscriber, enabling the Digitalhaus Franken team to proactively prevent potential cancellations through appropriate marketing measures.

Where do we currently stand? After first assembling an interdisciplinary project team within our organisation, we successfully merged customer data and information from our analytics tool into a data warehouse. We then established hypotheses about which behavioural characteristics of our customers might influence cancellation probability and selected the corresponding metrics. How long has a customer been subscribed? How frequently do they use their subscription? How long do they use the subscription per visit? Which payment method do they use? Through which referrer did they subscribe? All these questions and more play a role in determining how long a customer maintains their subscription, as initial analyses have shown.

Which AI-Model provides the best predictions?

Since we aim not just to uncover the timing of cancellations once, but to continuously predict cancellation probabilities, the data must feed into an automated model. This first requires the appropriate infrastructure – based on the Microsoft Azure environment and Databricks, a cloud platform for data analysis and AI. And of course, everything in compliance with applicable data protection regulations.

Currently, our project work focuses on training different AI models and evaluating the results. The selection includes logistic regression, Random Forest, XGBoost, and Support Vector Machine. Which model will ultimately prevail will be decided shortly after the completion of the testing phase.

Our goal is to identify as many potential subscription cancellations as possible with the model. However, explainability also plays an important role in selecting the “final” model. We want to know not only who has a high cancellation probability but also why the model has classified the customer this way.

If several models are close in the quality of their predictions, their operational resource efficiency will also factor into the decision. This is because we are not simply calculating cancellation probabilities once, but building ongoing operations with continuous training to refine the model. Not least, the marketing measures used to prevent cancellation must be fed back into the model. Ideally, marketing automation should be integrated at the end, enabling the entire system to run fully automatically.

Precisely tailored marketing measures for each individual customer

And that brings us to the next steps. By the end of the JournalismAI programme, we aim to launch a Minimum Viable Product (MVP). To achieve this, our next task is to select the AI model we will use to operate our system. Once selected, the data will be integrated into dashboards for further analysis.

Once the system is in place and delivering reliable data, the real work within our organisation begins. Our marketing team will then need to continuously develop custom-tailored measures to effectively prevent customer cancellations.

And as a reminder: ideally, this should happen before customers even consider cancelling their subscription. Because every subscription we retain helps ensure that people in our region stay informed with well-researched journalism.

—

This article is part of a series providing updates from 35 grantees on the JournalismAI Innovation Challenge, supported by the Google News Initiative. Click here to read other articles from our grantees.