Attrition models typically consider only structured attributes and past behavior of customers to segment them based on their propensity to defect. However, this methodology has its shortcomings. Specifically, it does not take into consideration the unstructured data, which is generated in customer conversations. For instance, a customer, while interacting with a bank's customer service agent might show his displeasure with the bank's services and bring it to their notice through chats or messages in Facebook or Twitter. Frequent incidents of this nature are a good indicator that the customer is ready to defect. Models based on text mining of unstructured data have clearly indicated a strong correlation between certain customer queries and intent to defect.