Churn Prediction Model for a Regional Telecom Provider

A regional telecom provider in South India was facing a steadily rising churn rate and lacked a data-driven way to predict and prevent customer drop-off. While they had rich customer data (usage, complaints, payments), it was underutilized.
Vortunix was brought in to design and deploy a machine learning-based churn prediction system to empower their retention team.

Here’s how we delivered impact:

1. 1.Data Consolidation & Feature Engineering Combined customer data from 6 sources: usage logs, support tickets, billing history, SMS logs, and app interactions. Created 35+ meaningful features (e.g., average recharge delay, complaint frequency).
2. 2.ML Model Development Built and trained models using XGBoost and Logistic Regression. Selected final model based on F1-score and business interpretability.
3. 3.Model Deployment Deployed via REST API (Flask) integrated into the client's CRM system. Customer service reps received churn risk scores along with actionable recommendations.
4. 4.Dashboarding & A/B Testing Built a Power BI dashboard to monitor model performance and ran A/B tests comparing targeted retention campaigns vs non-targeted.

1. 82% Model Accuracy (F1-Score: 0.79)
2. 18% Reduction in Churn over 3.5 months
3. ROI of 3.5x the project cost within the first quarter
4. CRM-Embedded Risk Scores improved proactive engagement by 60%
5. Data-Driven Retention Strategy built using model output & BI insights

1. 1.Deployed Azure Purview: Centralized scanning and data mapping
2. 2.Set Data Access Policies: Role-based access integrated with Azure AD
3. 3.Established Lineage & Classification: Business glossary + data sensitivity tagging
4. 4. Built Approval Workflows: Automated governance request pipelines
5. 5. Created a Trust Layer: Certified datasets with full audit traceability