Telkomsel is currently working to enhance the basis of the digital business by pursuing digital transformation. Given the intense competitiveness in the digital telecommunications industry, this is a crucial task. MyTelkomsel Apps, a self-service digital channel, was introduced by Telkomsel on March 25, 2016, as evidence of the company's digital transformation. MyTelkomsel Apps streamlines the purchasing of credit, data packages, kartuHalo bill payments, and exchange point self-service for Telkomsel customers. However, there is still a low conversion rate in the MyTelkomsel Apps between active users (users of the app) and package users (purchasers of the product). There are active customers who previously actively purchased products but do not currently do so on MyTelkomsel Apps, or what is known as MyTelkomsel Lapser, which is one of the causes of the poor conversion rate. Therefore, the issue at hand today is how to reactivate MyTelkomsel Lapser so they can renew their product to MyTelkomsel Apps. In order for Telkomsel to offer tailored treatment to recover lapsed transactions in MyTelkomsel Apps, the goal of this research is to give Telkomsel knowledge into the precise profile of MyTelkomsel Apps lapsers. 2.206.636 clients are lapsers as of March 2023. The decision tree algorithm is used to predict lapsers who are in the high prospect group to become non lapsers and The K-Means Cluster algorithm is used to create lapser segments from 2.206.636 lapsers based on relevant variables that have been determined, and in one data set with a period of March 2023. There are 70 variables made up of 1 geographic variable, 8 psychographic variables, and 61 behavioral variables. 1.914.232 lapsers are collected from the data cleansing procedure, which may then be processed to separate them into 20% testing data and 80% training data. With a model accuracy of 97.25%, it is predicted that 45,116 (2.35%) consumers will make purchases at MyTelkomsel and 1.869.116 (97.64%) customers will stay lapser after the variables are discovered and the lapser data is processed using the decision tree method. The top 5 significant variables from the model are the number of data package purchases made through the MKIOS channel, the UMB channel, acquisition-type purchases, physical voucher purchases, and transaction-type purchases of data packages. Furthermore, using the K-Means Cluster technique, 45.116 consumers are divided into clusters based on the top 5 important variables. The silhouette index for these 45.116 high-prospect clients is divided into 3 clusters, with n_cluster = 3 having the biggest value (0.78), compared to the other n_cluster. Low Data User (94.89%), Medium Data through UMB Channel (4.14%), and Physical Voucher User with Medium ARPU (0.97%) are the 3 clusters that were formed.

Telkomsel can perform behavioral targeting for the three clusters based on the prediction model & clustering results to give tailored product gimmicks.