Ensuring effective and timely communication in the fund administration industry has become an essential element of client relationship management (CRM). A CRM team periodically evaluates asset managers perception of service quality using surveys, where it can be difficult to evaluate service quality objectively. Within the asset management industry, despite ongoing technological advances, the main channel of communication remains email. The sheer volume of email, and the industry’s reliance on it, can lead to practical problems that impact CRM. In this work we draw insights from the email communications between an operations team and two-sample clients to understand client relationships in a way not previously possible. The results are presented and we discuss how these can quantitatively support and improve service quality evaluations in CRM. For this application, we exploit the social relations in emails via a graph-based approach. A deep learning framework is described that allows a graph-based inspection of the email communications between asset managers and their fund administrators operations teams. The presented framework integrates a natural language processing model to transform email subject lines to embedding representations, a knowledge graph to transform the email communication links into a graph representation, and a graph neural network to process the embedding representations and classify the email communications. The classification of critical conversations via email is a demonstrative example of a scalable graph-based approach that allows the use of machine learning to process, learn, and explore the relations existing between the emails.