Automatic complaint identification is crucial to organizations as they provide insights to meet the customers’ requirements, including handling and addressing the complaints. However, compared to manually tagged complaints in financial institutions, automatic identification of complaints and associated severity level could direct them to specific teams, save crucial time addressing the complaints, and help understand the complainers’ needs. In this work, we curate a new corpus of Financial Complaints (FINCORP) for aiding complaint identification and complaint severity classification research. We present a framework for joint learning of (a) binary complaint classification, (b) complaint severity level classification, (c) emotion recognition, and (d) sentiment analysis. The proposed approach outperforms the state-of-the-art approaches and other baselines based on the extensive evaluation.