The technology leverages privacy-preserving analytics architecture to enable enterprises to mutually share insights of overlapping customers without moving any of their data. The solution utilizes a federated learning approach where each enterprise has its own model that is trained locally with its own data of overlapping customers. In the training phase, encrypted customer ids are used by models to identify common customers and only intermediate representations are shared between models through a federated learning execution engine. Once the training is done, in the prediction phase, only the insights on the usage behavior of common customers are shared with the participating enterprises.
A Telco and its OTT partner leveraged Flytxt’s privacy-preserving analytics to determine the potential churn risk of their common customers. Download and read the case study to know more about how this solution predicted churn with 86% accuracy without moving any data between the enterprises.
The technology can be leveraged to analyze data from different B2C enterprises to derive insights from multi-party data to support use cases such as identifying credit risk scores, sentiment scores, product affinity, loyalty scores, etc. of common customers between the different enterprises.