Termination behavior, churn prediction and rotational churn for monthly subscription contracts

Our customer, a leading European Internet company, wanted to compare customer satisfaction between brands after numerous acquisitions within their multi-brand strategy. Termination behavior was analyzed, based on niologic’s product portfolio.


Our customer had bundled products for different customer groups into different brands through acquisitions and a multi-brand strategy. Termination data was implicitly made available by the respective contract management of the brand or subsidiary, but could only be compared manually. In addition, there still were differences in the definition of key figures between brands.


niologic created a common data structure for the customer and merged the contract data of all brands in a data warehouse (DWH). In addition, the products and each’s brand product hierarchy were combined in a common group-wide product hierarchy or portfolio management. Ultimately, there were several product hierarchies for finance, marketing and sales as well as the respective subsidiary as an internal view.

On the basis of the product hierarchy, as well as contract data and transaction data, additions and removals of contracts were modelled as a sparsely populated matrix. The termination behavior was predicted using Kaplan-Meier estimators and Bayes methods. The data structure was created in a column-based InMemory database to ensure high-performance analyses.

Results and customer value

Introducing a common product hierarchy for portfolio analysis, it was made possible to analyze contract changes at every level of the product portfolio. This way, individual brands could also be compared together on those levels (e.g. product group).

This also allowed analying the termination of old contracts and the simultaneous closing of new contracts (rotational churn) within a product group. The customer’s product management was thus able to draw numerous conclusions about customer behavior within the brands and flexibly compare the brand strategy to the operating result. The effects of Rotational Churn caused by the introduction of new services within the latest contracts, could thus be measured immediately after introducing portfolio management.

Customer Lifetime Value (CLV) could also be shown by individually distributing contract duration per customer or aggregated on the product hierarchy levels (simply put: distribution * sales). Therefore, it was no longer necessary to resort to individual averaging or precalculations. Also, the CLV could be compared between brands.