Case Study: Swisscom improves customer query classification with TensorFlow

How Swisscom’s Custom-Built TensorFlow Model Improved Business Operations by Classifying Text

Swisscom, a telecommunications provider, faced the challenge of automatically classifying a high volume of complex customer queries from emails and chats. Their existing rule-based and machine learning solutions were ineffective, leaving too many requests undetected and unprocessed. They turned to the vendor TensorFlow, using its platform and the tf.keras API, to build a custom solution.

Using TensorFlow, Swisscom's data scientists built and ensembled custom deep learning models with architectures that combined CNNs, RNNs, and fully connected layers. This solution successfully classifies customer intent in short chats and automatically routes more complex emails. The TensorFlow implementation resulted in a significant increase in accuracy, with a 10% improvement for chat classification and a 22% improvement for email triage compared to the standard industry approach.

Swisscom

Athanasios Giannakopoulos

Swisscom