Telecom Customer Churn Analysis

Research-based churn analytics across Airtel, BSNL, Vodafone, and Reliance Jio. A Python EDA pipeline that surfaces retention strategy by customer segment, plan type, and usage behaviour.

Year 2024
Role Data Analyst
Type EDA Research
Timeline 4 weeks
Dataset Multi-carrier Indian telecom
Carriers Airtel, BSNL, Vodafone, Jio
Status Shipped
Tools Python, Pandas, Matplotlib, Seaborn

The Problem

Customer churn in the Indian telecom sector runs between 25% and 35% annually across major carriers, with new entrants (Jio post-2016) forcing legacy operators to compete on price at the expense of margins. The research question: which customer segments churn most, and what signals precede churn in the data?

The goal was not to predict individual churn (that requires a production ML pipeline), but to produce an EDA report that gives a retention team actionable segment-level insight: who to target, with what offer, and why.

Data Sources

The dataset combined public telecom research data with carrier-published aggregate statistics across four networks. Key variables:

  • Plan type (prepaid, postpaid, bundled data-voice)
  • Tenure band (0–6 months, 6–12 months, 1–2 years, 2+ years)
  • Monthly spend quintile
  • Average monthly data consumption
  • Number of customer service contacts in past 90 days
  • Carrier (Airtel, BSNL, Vodafone, Jio)
  • Churn flag (binary, self-reported or derived from account closure)

Stack

Python 3.11
Pandas
Matplotlib
Seaborn
Jupyter Notebook
NumPy

Key Findings

Three patterns drove the majority of churn across all four carriers:

  • Tenure cliff at 6 months. Churn rate is highest in the first 6 months (38% in the first quarter, falling to 18% by month 12). Customers who survive past one year have a churn rate of under 8%.
  • High data use, low spend. Customers in the top data-usage quintile but the bottom spend quintile churn at 2.4x the rate of balanced-usage customers. These are value-hunters, likely to leave for any better data-per-rupee offer.
  • Customer service contacts predict churn. Any contact in the past 30 days doubles churn probability; two or more contacts triples it. This is a clean, real-time retention signal that requires no model.

Carrier Comparison

Jio showed the lowest churn rate (14% overall) driven by aggressive bundled pricing. BSNL had the highest (41%), concentrated in postpaid customers aged 45+ on legacy plans with no data bundle. Airtel and Vodafone sat in the 26–31% range, with Vodafone's churn skewing toward urban prepaid and Airtel's toward rural postpaid.

Outcome

Delivered as a 28-page EDA report with visualisations and segment recommendations. The rule-based retention flag was the primary deliverable: something a team can use immediately without deploying a model.

Results

What the Data Showed.

4
Carriers analysed (Airtel, BSNL, Vodafone, Jio)
38%
Churn rate in first quarter, highest risk window
2.4x
Higher churn for high-data, low-spend segment
61%
Of churners captured by the rule-based retention flag