Multi-labeled call transcription classification is essential for public and private sector organizations, as they spend a lot of time and workforce manually classifying phone call queries. Implementing a machine learning-based auto-classifier can efectively assist in this task, especially by reducing the time and resources required. The paper proposes an efficient call transcription classifier that not only reduces manpower but also saves time significantly. The first step in transcript cleaning involves several essential processes, such as converting the transcript to lowercase, applying word embedding techniques, and removing numbers, punctuation, and stopwords. The second step involves designing the model to incorporate four separate classifiers, each trained independently. Each classifier consists of a bi-directional LSTM layer, an embedding layer, and three subsequent dense layers. These dense layers use the ReLU as an activation function, and softmax as a final layer. The experimental results demonstrate that all four classifiers have achieved precision, recall, and F1-score greater than 80%. In conclusion, we conduct a comparative analysis of the results against existing studies, demonstrating that our model has exhibited superior performance.