ABSTRACT
Customer relationship management is crucial to a company’s ability to sell its goods. It is challenging for small business owners to monitor salespeople’s talks with customers and use that information to increase sales to that customer. Even though a lot of small business owners capture the salesperson’s conversations, very few of them use the sentiments of the discussions in sales predictions. This research aims at using the conversation sentiments as additional features for sales prediction. The conversations were preprocessed, vectorized, and classified to be used as features. Deep learning models including an artificial neural network (ANN) and a gated recurrent unit (GRU) neural network, and ensemble models including voting, bagging, stacking, and boosting were used to classify the conversation sentiments. These features were modeled in cohesion with other features, including day, time, month, quantity, products, and clients for prediction of sales done by utilizing ensemble regression and classification models. The stacking ensemble model with an accuracy of 86.5% and the ANN deep learning model with an accuracy of 90.8% were found to be the most accurate in classifying conversation sentiments. Sentiment of conversations was among the most important features for predicting order-wise and product-wise data.
