Keywords: Clustering; Deep Learning; Text Mining; Topic Modeling; Word Embedding
Digital care is a growing and important customer service channel that can reduce costs while shortening customer wait times. These advantages are contingent on connecting the customer to the best agent to solve his/her problem. Thus, automatically understanding each customer’s contact reason is a key building block. In this talk we analyze AT&T’s online care chat data, where the chats in our data were routed to an initial agent using information such as the customer’s chat launch page URL, navigation path on the AT&T webpage, and chat menu guide selections. Our findings suggest that while a customer’s online activity reflects his/her self-initiated search path for answers, the customer’s true, detailed intent is not revealed until the conversation starts. To improve routing, we propose an adaptive procedure based on intent identification. In our approach, we learn each agent group’s expertise from recent chat transcripts using scalable deep learning and topic modeling techniques. We then use these agent profiles to align each customer’s intent with the correct agent. We demonstrate the effectiveness of this approach and discuss future directions for intelligent digital care.