Case Study Churn Predication
Case Study Churn Predication
Case Study Churn Predication
Telecom market is highly competitive and subscribers actively switch from one operator to another. Cost
burden of 5-10 times to acquire a new subscriber than to retain an existing one therefore retention is im-
Introduction portant from profitability aspect along with revenue. We processed Usage CDRs, recharge data, customer
complains & profile data and built predictive model using AI/ML. Churn prediction & scoring on high risk
subscribers was provided with dashboard for a leading telco in Middle East & Africa region.
Data preparation was an iterative process until we get the desired results from the model evaluation
EDA, Feature Engineering & Data Preprocessing were iterated time and again we add more data point
Technical challenges in providing access to data, inconsistencies in the data structure and environment
Challenges setup
SLA with the data owners (upstream systems) for regularly receiving Usage CDRs, Recharge Data & Pro-
file information for model update and automation
Data streamed from various data sources to our Pronto data science sandbox based on Apache Kafka,
HDFS, Hive & Spark.
Exploratory Data Analysis performed using Pronto’s UI based Data Science Studio and important KPIs fro
EDA added to the final dashboard. Model developed, deployed with automation, scheduled reporting and
dashboarding with customer behavioral profile provided.
The below were the highlights of the findings:
Telco to pay attention to the roaming rates and provide dynamic offers to subscribers using services Solution
from roaming zone
ISD rate are too high and offers are required for High Value Subscribers for certain destinations
Customer query and complaint data has correlation with churn with certain time lag therefore timely
resolution is important
Age on network along with recharge value aggregation considered as a metric for customer loyalty
provides pattern for behavioral change hence focus required to improve CLV
Recommendation from Churn Prediction & Scoring resulted in gradual drop in proportion of high value
Results customer churn and was carried forward to medium segments too. A new project to integrate & auto-
mate campaigns with churn analytics also proposed to further improve business results
Gamma is a telecom focused analytics & data management company offering domain spe-
551 Wilmot Road
cific Complex Event Processing (CEP) platform for Business Assurance & AI/ML based Busi- [email protected]
New Rochelle,
ness Analytics for telcos globally. Gamma enables telcos small to large transition to a real- New York 10804,
+1 914 740 4067
time, full-scale architecture with ML/AI easily and quickly while reducing time-to-value in United States www.gammanalytics.com
their data operations