Classification of Emails in Customer Support

Challenge

Timely and satisfactory handling of customer inquiries and complaints is always a significant challenge for companies. Our client additionally faced the problem of receiving emails through various inboxes. These messages are often not initially addressed to the correct contact person and therefore frequently need to be manually forwarded to the appropriate department. This manual process can significantly delay response times. As a result, our client decided to implement automatic email routing using AI, allowing customers to continue sending messages to familiar inboxes, even if they are not the correct ones.

Approach

In close collaboration with the client, we analyzed the existing email inboxes. We thoroughly examined the KPIs of each department, evaluating the path of an email within the client's infrastructure—from receipt to response. Automatic email forwarding was achieved using deep learning models in the NLP domain. We utilized existing, freely available models and adapted them to the client's specific terminology, achieving excellent results with moderate resource usage. The correct contact person for the messages was automatically extracted from the email thread, allowing the deep learning model to be trained on a very large dataset with minimal manual effort from the client.

Result

Our client can now automatically determine the correct department for an email with very high accuracy. This prevents manual forwarding, frees up resources for customer service employees, and ultimately leads to faster response times. Additionally, we laid the groundwork for enabling more comprehensive, automated handling of incoming messages across various channels in the future. The developed model is now being transitioned into production in further steps.