Customer Churn Prediction Using Machine Learning Models | SQream

Business Value Case

Customer Churn Prediction Using Machine Learning Models

Up to 15 %
churn reduction
47% customer retention
from the targeted churn group
10 months of aggregated
multi-channel data

Tier 1 in Latin America has adopted SQream as a scalable, cost-effective, ultra-fast platform to utilize never-used data about their customers to predict their churn rate, two-months before the churn, for a period of 10-months. The data resources that were used were never analyzed before due to the large
volume. 47% of the target population that were identified as a churn candidate accepted an enhanced offer (next-best-offer) and remain the telecom’s customers. Most of them signed a renewal during the first week. Minimizing churn rate is also part of achieving a high Customer Lifetime Value (CLV).

Industry Vertical: Telecom
Economic Buyer: Head of Customer Experience
Enabler: BI Teams; Lead Data Scientist

 

Why Do Anything?

The previous platform failed to process all the required resources for churn analysis. Increasing the retention period of customers was one of the main strategies to generate more revenue. The business was looking for an enterprise-ready technology which was capable of processing massive amounts of data in an efficient way.

Why Now?

Since Covid-19, strict travel restrictions caused roaming incomes to fall sharply. The business was looking for creative ways to increase its revenue.

Why SQream?

  • Ability to rapidly join many tables with large datasets.
  • Ability to recognize only delta of changes.
  • Ability to rapidly ingest massive amounts of data.
  • Ability to query large data sets on petabyte scale.
  • Ability to work Pub/Sub with Kafka, and trigger transformation based on business events.

Fastest time to insight on any size data

Business Challenge

Acquiring a new customer costs 600% more than the cost of retaining the customer likely to churn. Predicting customer churn is critical for telecommunication companies to be able to effectively retain customers. Telecoms apply Machine Learning models to predict churn on an individual customer basis and take counter measures to retain their customers. This telecom needed churn data-driven analysis that would assist in identifying key stages in the customer journey, where people were falling off, allowing the pinpointing of specific strategies to improve their interactions with the brand and to improve their loyalty.

Situation/Pain Business Impact

Data mining techniques were applied on top of the current data warehouse system, but the model failed to provide good results using this data. Data sources which were massive in size were ignored due to the complexity in dealing with them. The data warehouse was not able to acquire, store, and process the huge amount of data at the same time. In addition, the data sources were from different types, and gathering them in the data warehouse was a very difficult process. This meant that adding new features for data mining algorithms required a large investment of time, high processing power, and more storage capacity.

The SQream platform deployed with multi connectors and gathered 10-months of data including: CDR; IPDR; network inputs (call drops; data usage); social media inputs; agent sentiments and more. The marketing experts decided to predict the churn two-months before the actual churn action, in order to revisit the offer to these targeted customers.

SQream Solution Components
  • SQream Kafka Connect
  • SQream Avro Connector
  • SQream SQL Joins
  • SQream Ad-Hoc Queries
  • SQream Cross-Joins
  • SQream for External Table
  • SQream Uploader
  • SQream for Clouds
Trusted by:

Architecture Considerations