PERANCANGAN MODEL SEGMENTASI PELANGGAN PT X BERDASARKAN NILAI CUSTOMER LIFETIME VALUE MENGGUNAKAN TEKNIK DATA MINING

• Main Author: Inatsan Ajriya Mumtaz, Aulia
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/80793
• Institution: Institut Teknologi Bandung

PT X is a technology company that provides information about job vacancies for blue-collar workers. In an effort to improve user retention and seek funding for the company, the company created a business model, executed by the Digital Services Division. This business model combines affiliate marketing, crowdsourcing, and gamification. The Digital Services Division of PT X acts as an intermediary between other companies and PT X service users, earning commissions to prompt desired actions from service users, while also utilizing crowdsourcing and gamification to encourage service user engagement in affiliate marketing tasks set by other companies. However, observing a declining trend in task completion, it was found that the marketing carried out by the Digital Services Division was done without using targeting or understanding the behaviors of the customers. Therefore, a customer segmentation model using data mining technique will be designed to help the company understand customer behaviors and improve data-driven marketing strategy effectiveness. The customer segmentation model is designed based on Customer Lifetime Value (CLV), data mining techniques, and the CRISP-DM methodology. User data is grouped based on the Recency, Frequency, Monetary (RFM) model along with additional variables such as the number of clicks and success rate. Then, the CLV of each user is calculated based on the weighted value of these variables obtained through the pairwise comparison method, and cluster analysis is performed using the K-Means, agglomerative clustering, GMM, BIRCH, and genetic algorithm. Evaluation of the modeling results shows that the BIRCH algorithm has an advantage in terms of clustering quality scores compared to other algorithms, thus hyperparameter tuning is conducted, resulting in a threshold of 0.8 and a total of 3 clusters with a silhouette score of 0.799. The final outcome of this method is a prototype designed using Streamlit.