Prediksi Customer Lifetime Value (CLV) Berbasis RFM (Recency, Frequency, Monetary) dan Machine Learning dalam Bisnis Network Marketing

Customer Lifetime Value (CLV) Prediction Based on RFM (Recency, Frequency, Monetary) and Machine Learning in Network Marketing Business

Customer Lifetime Value (CLV) merupakan indikator strategis untuk mengukur nilai ekonomi jangka panjang pelanggan dan menjadi dasar penting dalam pengambilan keputusan pemasaran, khususnya pada bisnis network marketing yang bersifat non-kontraktual. Penelitian ini bertujuan membangun model prediksi CLV yang lebih adaptif dengan mengintegrasikan model RFM (Recency, Frequency, Monetary) dan variabel beyond RFM yang merepresentasikan karakteristik struktural pelanggan. Penelitian menggunakan 17.185 observasi data transaksi pelanggan yang disusun dalam format time series rolling window tiga bulan, merepresentasikan perilaku pembelian pelanggan unik secara longitudinal. Variabel beyond RFM meliputi point value, jumlah item pembelian, level agen, wilayah, dan status keaktifan pelanggan. Pemodelan dilakukan menggunakan beberapa algoritma machine learning regresi dengan pendekatan time-series cross-validation untuk menjaga konsistensi temporal dan mencegah data leakage. Evaluasi kinerja model menggunakan MAE, RMSE, dan koefisien determinasi (R²). Hasil penelitian menunjukkan bahwa integrasi variabel beyond RFM secara konsisten meningkatkan akurasi prediksi CLV dibandingkan model RFM baseline. Model Random Forest pada skema RFM + beyond RFM memberikan kinerja terbaik dengan nilai R² sebesar 0,827. Analisis interpretabilitas menggunakan SHAP mengonfirmasi bahwa variabel khas network marketing, khususnya point value dan level agen, memiliki kontribusi signifikan terhadap estimasi CLV. Penelitian ini menghasilkan kerangka prediksi CLV yang lebih interpretatif, dan relevan untuk mendukung pengambilan keputusan strategis berbasis nilai pelanggan.

Kata kunci: Customer Lifetime Value; RFM; Beyond RFM; Network Marketing; Machine Learning

Customer Lifetime Value (CLV) is a strategic indicator used to estimate the long-term economic value of customers and serves as a critical foundation for marketing decision-making, particularly in non-contractual network marketing businesses. This study aims to develop a more adaptive CLV prediction model by integrating the traditional RFM (Recency, Frequency, Monetary) framework with beyond RFM variables that capture customers’ structural and behavioral characteristics. The study utilizes 17,185 transactional observations organized in a three-month rolling window time-series format, representing longitudinal purchasing behavior of unique customers. The beyond RFM variables include point value, number of purchased items, agent level, geographic region, and customer activity status. Several regression-based machine learning algorithms were employed using a time-series cross-validation approach to preserve temporal consistency and prevent data leakage. Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R²). The results demonstrate that incorporating beyond RFM variables consistently improves CLV prediction accuracy compared to the RFM baseline model. The Random Forest model with RFM + beyond RFM features achieved the best performance, with an R² value of 0.827. Furthermore, SHAP-based interpretability analysis confirms that network marketing-specific variables, particularly point value and agent level, contribute significantly to CLV estimation. This study provides a more accurate, interpretable, and context-aware CLV prediction framework to support value-based strategic decision-making.

Keywords: Customer Lifetime Value; RFM; Beyond RFM; Network Marketing; Machine Learning