Telecom Customer Churn Analytics

An end-to-end analytics project using Python, MySQL and Power BI to examine customer attrition, revenue, contracts, services and recorded churn drivers across 7,043 telecom customers.

Verified portfolio metrics

The committed reports are generated directly from data/raw/telco.csv:

Metric	Value
Customers	7,043
Churned customers	1,869
Churn rate	26.54%
Total revenue	$21,371,131.69
Average monthly charge	$64.76
Average tenure	32.39 months

Key descriptive findings from the dataset:

• Month-to-month contracts have a 45.84% churn rate, compared with 10.71% for one-year and 2.55% for two-year contracts.
• Fibre-optic customers have the highest recorded churn rate among internet types at 40.72%.
• The most frequently recorded churn category is Competitor, representing 841 of the 1,869 churned customers.

These are descriptive associations in this dataset, not causal conclusions.

Business questions

1. What is the overall churn rate and revenue exposure?
2. Which contract types and services have the highest churn rates?
3. Which recorded reasons contribute most to churn?
4. How do payment method, tenure and satisfaction differ across customer outcomes?
5. Which segments should be prioritised for deeper retention analysis?

Project structure

.
├── assets/                         # Reproducible charts used in this README
├── data/
│   ├── raw/telco.csv               # Source dataset (7,043 rows, 50 columns)
│   └── processed/                  # Generated clean CSV for MySQL
├── notebooks/01_connect_mysql.py   # Secure connectivity/query example
├── powerBI/README.md               # Dashboard design and PBIX guidance
├── reports/                        # Verified JSON/CSV outputs
├── scripts/
│   ├── build_reports.py            # Rebuilds reports and charts
│   └── prepare_mysql_data.py       # Creates snake_case MySQL input
├── sql/                            # Schema, load script and analytics views
├── src/telecom_churn/              # Reusable Python analysis/database code
└── tests/

Reproduce the analysis

python -m venv .venv

Windows:

.venv\Scripts\activate

macOS/Linux:

source .venv/bin/activate

Install dependencies and the package:

pip install -r requirements.txt
pip install -e .

Rebuild all reports and charts:

python scripts/build_reports.py

Run tests:

pip install -r requirements-dev.txt
pytest

MySQL workflow

1. Copy .env.example to .env and enter a local MySQL username/password.
2. Generate the clean snake_case CSV:

python scripts/prepare_mysql_data.py

3. Run sql/01_schema.sql.
4. Update the absolute CSV path in sql/02_load_data.sql, then run it with LOCAL INFILE enabled.
5. Run sql/03_analytics_views.sql.
6. Test the connection:

python notebooks/01_connect_mysql.py

Credentials are read from environment variables and are not stored in source code.

Power BI

The original repository already contains powerBI/telecom_churn.pbix. Preserve that binary file while applying this package. See powerBI/README.md for recommended pages and validation checks.

Data quality notes

• Churn Category and Churn Reason are populated only for churned customers, so their missing values are structurally expected for other customer statuses.
• Internet Type is blank for customers without an applicable internet type.
• Customer ID is validated as unique before analysis.
• Reported rates are calculated as churned customers divided by all customers within the displayed segment.

Security correction

The previous repository version included a plaintext MySQL password in notebooks/01_connect_mysql.py. This version removes the credential and uses .env variables. Any password previously committed should be changed because deleting it from the latest file does not remove it from Git history.

Limitations

• The project is descriptive and does not claim that a segment attribute causes churn.
• No predictive model or out-of-sample evaluation is included.
• The dataset represents a fixed snapshot and may not generalise to another telecom provider or time period.

Author

Govardhan Reddy — MSc Big Data Analytics candidate

Licence

Code is released under MIT. Confirm the source dataset's licence and attribution requirements before redistributing it elsewhere.