Aulia Pinkasari, Meiyin Monica Amilia Putri, Gibral Abdurahman, Ahmad Fadhil Rizqi, Ken Ditha Tania, Allsela Meiriza, Ahmad Rifai

Penerapan Model Recency, Frequency, Monetary untuk Segmentasi Pola Perilaku Pelanggan Indibiz

Introduction

Penerapan model recency, frequency, monetary untuk segmentasi pola perilaku pelanggan indibiz. Segmentasi pelanggan Indibiz menggunakan model RFM dan K-Means clustering untuk mengidentifikasi pola pembayaran. Optimalisasi strategi penagihan berbasis data transaksi di Telkom Sumbagsel.

Abstract

PT Telkom Indonesia Witel Sumbagsel menghadapi kondisi Data Rich, Information Poor (DRIP), yaitu melimpahnya data transaksi yang belum dimanfaatkan secara optimal sebagai dasar pengambilan keputusan pada unit Payment Collection. Penelitian ini menerapkan metodologi Knowledge Discovery in Databases (KDD) menggunakan model RFM (Recency, Frequency, Monetary) dan algoritma K-Means Clustering untuk mengidentifikasi pola perilaku pembayaran pelanggan. Dataset terdiri dari 46.355 transaksi periode Januari–Desember 2025. Jumlah cluster optimal ditentukan menggunakan metode Elbow dan menghasilkan empat segmen pelanggan (k=4). Evaluasi menggunakan Silhouette Coefficient memperoleh nilai 0,3463 yang menunjukkan kualitas klaster yang dapat diterima. Hasil segmentasi mengelompokkan pelanggan ke dalam kategori loyal, potential, standard, dan churn-risk, sehingga mendukung penyusunan strategi penagihan yang lebih tepat sasaran dan berbasis data. Penelitian ini menunjukkan bahwa integrasi RFM dan K-Means efektif dalam mengubah data transaksi menjadi wawasan praktis bagi manajemen penagihan telekomunikasi.

Review

This paper addresses a highly pertinent challenge in data-rich environments: transforming vast transactional data into actionable intelligence. Focusing on PT Telkom Indonesia Witel Sumbagsel, the research effectively tackles the "Data Rich, Information Poor" (DRIP) condition within their Payment Collection unit. By employing the well-established Knowledge Discovery in Databases (KDD) methodology, integrated with the RFM (Recency, Frequency, Monetary) model and K-Means Clustering, the authors propose a systematic approach to segment customer payment behavior. This endeavor directly supports the development of more targeted and data-driven collection strategies, offering significant practical value to the telecommunications giant. Methodologically, the study demonstrates a robust approach. The choice of RFM, a classic and powerful model for customer value analysis, combined with K-Means, a widely accepted unsupervised learning algorithm, is highly appropriate for identifying distinct customer segments from transaction data. The substantial dataset of 46,355 transactions, albeit for a future period (Januari–Desember 2025), provides a solid foundation for the analysis. The authors' diligence in determining the optimal number of clusters using the Elbow method (k=4) and evaluating cluster quality with the Silhouette Coefficient (0.3463, indicating acceptable separation) further underpins the credibility of their findings. The resulting segmentation into 'loyal,' 'potential,' 'standard,' and 'churn-risk' categories offers clear, interpretable, and actionable insights for management. While the research provides a valuable framework and practical insights, future work could explore enhancing the predictive power of the segmentation. Although the Silhouette Coefficient is acceptable, there might be scope to achieve higher cluster separation by incorporating additional features beyond RFM, such as service usage patterns, historical complaint data, or demographic information if available. Furthermore, the paper could briefly discuss the implementation challenges and successes of deploying such a model in a live operational environment. Nevertheless, this study effectively demonstrates the utility of combining established data mining techniques to derive strategic value from raw transactional data, offering a commendable contribution to the field of customer analytics in the telecommunications sector.

Full Text

Comments

You need to be logged in to post a comment.