LEVERAGE UP PREDICTIVE ANALYTICS TO INCREASE NUMBER OF PREPAID CREDIT PURCHASER IN ECOMMERCE APPS
Ronny Arnaz, ST (a*), Dr. Gadang Ratmantoko (b)
a) Graduate Study Program of Management, Faculty of Economics and Business, Telkom University, Bandung
*Ronny_Arnaz[at]telkomsel.co.id
b) Lecturer of Schools of Economics and Business, Telkom University, Bandung
Abstract
Telkomsel is the largest mobile operator in Indonesia with 195 million customers serving its customers spread across Indonesia, including in remote areas, outer islands and border areas of the country. Telkomsel is currently transforming to digital company and one of its programs is to launch MyTelkomsel Apps as a digital channel and cooperate with other Ecommerce Apps to provide convenience to customers in conducting self-service online, such as prepaid credit purchase or Top Up. However, the conversion rate of Telkomsel subscribers who have done Top Up in Ecommerce Apps is still very low so the problem is how to increase the number of prepaid credit buyer in the Ecommerce Apps. The objective of research is to build insight for Telkomsel on predict customers who likelihood to purchase prepaid credit in Ecommerce Apps by clustering and using Logistic Regression technique. Therefore it can be identified customer profile and significant variables of customers to purchase prepaid credit in Ecommerce Apps based on the results of predictive analytics. Logistic regression used to predict customers who likelihood to purchase prepaid credit in Ecommerce Apps using 17 numeric variables input. Analytics based table generated from total population of prepaid credit purchaser in Ecommerce Apps as of Aug 31st 2018 by 297.172 customers and create random sampling of non purchaser in Ecommerce Apps by 127.359 customers out of 15.055.473. Analytics based table divided into 2 data sets, 70% training data set and 30% testing data set represent of both prepaid credit purchaser and non-purchaser. Based on training data set and testing data set by 424.531 customers, 259.747 customers are predicted likelihood to purchase prepaid credit in Ecommerce Apps with model accuracy reached 79%. Top 6 significant variables that affected customers to purchase prepaid credit in Ecommerce Apps are total recharge transaction of non-Ecommerce, total recharge transaction, number transaction of Ecommerce Apps, data packet transaction of Ecommerce Apps, broadband revenue and volume usage of Ecommerce Apps. After building development model then applied into all population of Ecommerce Apps by 15.055.473 costumers. There were identified 3,089,120 customers who highly likelihood to purchase prepaid credit in Ecommerce Apps. To increase efficiency and effectively in marketing programs, it is need to choose the priority of costumer. The segmentation is conducted for non-purchaser with the K means clustering and it can be determined which cluster that give the greatest predictive gain. The result of K-Means clustering show that number cluster k=5 and cluster-5 give the best segmented costumers with gain 99.8%. The prediction results of highly prospect customers (345.965) will be divided in to 2 clusters with Two Step cluster techniques, namely low HVC customer ( 67.5% = 233.365) and medium HVC customer ( 32.5%=112.600). Based on the results of the prediction model and clustering, so behavioral targeting can be done to give the product or gimmick then the Champagne is more targeted.
Keywords: Costumer Segmentation, Behavioural Targeting; Predictive Analytics
Topic: Digital Technology For Business