ABSTRACT
Telkomsel as a digital telecommunication company has released a digital-based virtual assistant (VA) application, which is used to facilitate customer transactions and to serve as an information channel. The numbers of purchasing transactions through VA are very low compared to the total active VA users, so it is necessary to analyze how to increase purchasing via VA. This research seeks to provide insight into customer characteristics by using prediction models based on historical data employing behavioral segmentation (occasion, benefits sought, user status, usage rate, and loyalty status). Data were analyzed using a random forest algorithm that utilizes per-segment data and all-segment data. Theoretical approaches used included digital marketing, behavioral segmentation, and customer behavior. Random forest analysis with a total of 22 input variables and 130,388 records resulted in an all-segment data model that produced the highest accuracy, 96%, with 49,948 predicted customers and the predictor importance variables of loyalty point, recharge, data payload, short message service (SMS), and voice transactions.
