Machine-Learning-Based Business Analytical System for Insurance Customer Relationship Management and Cross-Selling - Research proposal

Aug 5, 2026

Effective cross-selling practices strengthen customer relationships and streamline insurance business processes. This study proposes a novel three-stage Machine Learning-Based System (MLBS) to identify potential insurance customers and improve customer relationship management through better targeting. It combines under-sampling with an ensemble strategy, selecting the best training samples via artificial neural networks and using stacking for prediction. The approach achieves superior recall, precision, and AUC versus baseline models, enhancing support for cross-selling campaigns.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages17

Reading Time43 min

AccessFree to download

Introduction

Cross-selling as a marketing management process

Prior customer identification and prediction approaches

CRM, BPM, and the insurance cross-selling research gap

Study objective and proposed MLBS overview

A Machine-Learning-Based Business Analytical System for Insurance Customer Relationship Management and Cross-Selling Xiaoguang Tian Purdue University Fort Wayne Jun Todorovic Purdue University Fort Wayne Zelimir Todorovic Purdue University Fort Wayne Effective cross-selling practices are integral to maintaining strong customer relationships and optimizing business processes within the insurance industry. This study presents a novel three-stage Machine Learning-Based System (MLBS) designed to enhance the identification of potential insurance customers and improve customer relationship management. This study proposes combining under sampling strategies and an ensemble approach to improve prediction performance. The proposed MLBS method involves selecting the best training sample using artificial neural networks and employing the stacking ensemble approach. It yields superior prediction results, exhibiting the highest recall, precision, and Area Under the Curve (AUC). These advancements substantially bolster the efficiency of cross-selling strategies. This research pioneers the application of stacking ensemble learning within the cross-selling domain, representing a novel contribution to the business field. The outcomes underscore the superiority of the MLBS system over baseline models across multiple performance metrics, thereby significantly enhancing support for cross-selling campaigns in various businesses. Keywords: digital transformation, insurance cross-selling, customer relationship management, machine learning, artificial neural networks INTRODUCTION The innovative approach to marketing management has been associated with increased organizational performance and profitability. Marketing management is also a critical value-adding process in the business value chain of many organizations (Osarenkhoe & Bennani, 2007) . As a subdomain of marketing management, cross-selling is commonly used by the financial service industry to increase sales and profitability. Although cross-selling strategy focuses on selling peripheral and more expensive services to existing customers (Vyas & Math 2006), it must be noted that not all cross-selling efforts are worth the time and monetary investments made in them (Rosen, 2004) . For this reason, increasing the effectiveness of cross-selling activities while utilizing fewer resources is vital. As a result, many organizations today are attempting to be more proactive and innovative to gain competitive advantages through information and technology. The information available in current corporate databases enhances business opportunities and gains further insights. These insights will help “provide better customer service, make call centers more efficient, cross-sell products more effectively, help sales staff close deals faster, simplify marketing and sales processes, discover new customers, increase customer revenues, etc.”(Osarenkhoe & Bennani, 2007, p. 156) . Some studies explored various methods of identifying potential customers as the target market for cross-selling efforts. For example, Kamakura et al. (1991) developed a probability model to identify prospects based on their acquisition of financial products. Building on their original work, Kamakura et al.(2003) improved their model by adding a “factor analyzer” to enhance their model efficiency. Harrison and Ansell (2002) used the statistical method known as the “survival approach” to identify prospects and determine which product will be bought and when the purchase will be made. Knott et al. (2002) developed a next-product-to-buy (NPTB) model to predict which product a customer would most likely buy next using four different statistical techniques focusing on logistic regression, multinomial logit, discriminant analysis, and neural nets. Knott et al. (2002) found that, although the NPTB model generated incremental crossselling profits, its predictive power was not influenced by the statistical technique used dur

FAQs

How does the MLBS perform compared with baseline models?

The system delivers higher recall, precision, and AUC, demonstrating superiority across multiple performance metrics.

Aug 5, 2026

Upload by Emma Wilson