Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Will You Stay or Will You Go? Churn Prediction for an App-Delivered International Calling Service

Pagels Fick, Svante LU (2019) INTM01 20182
Innovation Engineering
Abstract
In markets with fierce competition and low switching costs, the capability to keep customers is just as important as the capability to get new customers. One of the cornerstones of efficiently keeping customers from churning is the ability to identify customers that are more prone to leaving the company than others. This way, churn prevention resources and measures can be efficiently directed towards the customer segments where the impact is the greatest.
This paper investigates how churn can be predicted for an individual customer in an app-delivered international calling service (Rebtel). It also investigates which the underlying behavioral drivers of churn are and how these findings can be used to efficiently prolong the lifetime for... (More)
In markets with fierce competition and low switching costs, the capability to keep customers is just as important as the capability to get new customers. One of the cornerstones of efficiently keeping customers from churning is the ability to identify customers that are more prone to leaving the company than others. This way, churn prevention resources and measures can be efficiently directed towards the customer segments where the impact is the greatest.
This paper investigates how churn can be predicted for an individual customer in an app-delivered international calling service (Rebtel). It also investigates which the underlying behavioral drivers of churn are and how these findings can be used to efficiently prolong the lifetime for customers using Rebtel’s subscription services when calling to India. The method applied to identify churn is the machine learning algorithm Random Forest.
It is found that the Random Forest algorithm is well suited for classifying which customers that are about to churn by identifying complex customer behavior patterns indicating that a certain customer is about to leave the service. When investigating individual customer features to isolate drivers of churn, the method of using the Random Forest algorithm is not the best suited approach and contributes with limited findings.
The recommendations to Rebtel is that a churn prediction model implementing the Random Forest algorithm should be used to segment which users, among users calling to India, that should be incentivized to stay as customers in order to decrease churn. (Less)
Please use this url to cite or link to this publication:
author
Pagels Fick, Svante LU
supervisor
organization
course
INTM01 20182
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Churn prediction, Random Forest, machine learning, Decision Tree
language
English
id
8998380
date added to LUP
2019-12-04 13:24:11
date last changed
2020-08-18 08:20:02
@misc{8998380,
  abstract     = {{In markets with fierce competition and low switching costs, the capability to keep customers is just as important as the capability to get new customers. One of the cornerstones of efficiently keeping customers from churning is the ability to identify customers that are more prone to leaving the company than others. This way, churn prevention resources and measures can be efficiently directed towards the customer segments where the impact is the greatest. 
This paper investigates how churn can be predicted for an individual customer in an app-delivered international calling service (Rebtel). It also investigates which the underlying behavioral drivers of churn are and how these findings can be used to efficiently prolong the lifetime for customers using Rebtel’s subscription services when calling to India. The method applied to identify churn is the machine learning algorithm Random Forest. 
It is found that the Random Forest algorithm is well suited for classifying which customers that are about to churn by identifying complex customer behavior patterns indicating that a certain customer is about to leave the service. When investigating individual customer features to isolate drivers of churn, the method of using the Random Forest algorithm is not the best suited approach and contributes with limited findings. 
The recommendations to Rebtel is that a churn prediction model implementing the Random Forest algorithm should be used to segment which users, among users calling to India, that should be incentivized to stay as customers in order to decrease churn.}},
  author       = {{Pagels Fick, Svante}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Will You Stay or Will You Go? Churn Prediction for an App-Delivered International Calling Service}},
  year         = {{2019}},
}