Customer Journey Prediction with Context-based Penalty under Imbalanced and Unlabeled Data

Abstract

Customer Relationship Management (CRM) is a strategic methodology that involves studying and predicting customer data, and integrating these insights into marketing decision-making processes to optimize profitability. The focus of this research is on predicting the customer journey, covering the consumer experience from first knowing about a product to making a purchase. However, the challenge lies in the absence of a definitive label set for accurately characterizing the customer journey. To overcome these limitations, we segmented customers with similar characteristics into labels through clustering and analyzed them with domain experts. Based on the created dataset, our model learns the relationship between customer data and different stages of the journey. This enables us to predict when new customers will enter the system. Our proposed methodology addresses the issue of imbalanced data and considers the situation in real-world applications of the predictive model during its development. The main objective is to create distinct marketing strategies for various customer journey stages. This involves grouping existing customers with undefined needs, understanding the requirements of new customers, and implementing tailored marketing strategies accordingly. Through experiments, we quantitatively and qualitatively validate the effectiveness of our proposed methodology for predicting customer behavior. In a case study focused on the financial industry, we used actual bank data to verify that our approach is applicable to marketing in real-world scenarios. We anticipate that our proposed approach will support data-driven decision-making, leading to enhanced customer experiences.