POTENTIAL LEADS SELECTION SYSTEM DESIGN FOR CROSS-SELLING & UPSELLING IN PT X ACADEMY SERIES SERVICES USING DATA MINING MODEL

• Main Author: Sammy Ivan Kurniawan, M
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/68664
• Institution: Institut Teknologi Bandung

PT X is a startup that focuses on education technology, providing tutoring services including mentorship for studying abroad and tutoring for other languages. One of the biggest contributors to PT X’s revenue are cross-selling and upselling activities between the services that compliment each other in PT X. However, these activities decreased by 64.28% in the early period of 2021. The root cause of this issue is because the sales of PT X feel that upselling and cross-selling takes up a lot of time due to the needs of contacting every customer, but do not yield high returns. Not all customers are willing to buy more products. To solve this problem, this study will focus on creating a data mining model to predict the potential of leads that will upsell or cross-sell in one of the most prominent services for cross-selling and upselling, academy series for language learning. After the model is completed, a potential leads selection system is constructed in the form of a simple prototype application that may be used by sales to predict the potential of the customers, hence increasing the effectivity and efficiency of the whole cross-selling and upselling process. This study refers to the CRISP-DM methodology which is one of the most popular data mining workflow. In it, there are a few steps which includes business understanding, data understanding, data preparation, modelling, evaluation, and deployment. 5 alternative models were used which are support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), light gradient boosting machine (LGBM), and voting classifier. Afterwards, alternatives of imbalance learning strategies were implemented which includes SMOTENC, SVMSMOTE, BorderlineSMOTE (BLSMOTE), and RandomOverSampler (ROS). It is concluded that RF with ROS achieved the best performance with an average of 83.83% F1-Score and 80.13% accuracy that may support upselling and cross-selling activities to increase revenue generation by 4 times compared to the conditions recorded on this study.