What is the Telco Customer Churn Dataset dataset?

IBM Telco customer churn dataset with 5,000 customers, 21 features, and 18.5% churn rate. Industry-standard benchmark for churn prediction models and customer retention analysis.

How can I download the Telco Customer Churn Dataset dataset?

You can download the Telco Customer Churn Dataset dataset in CSV or Parquet format directly from this page. Each table has its own download buttons.

What format is the Telco Customer Churn Dataset dataset available in?

This dataset is available in CSV and Parquet formats. CSV is great for spreadsheet applications, while Parquet is optimized for data analysis tools like Pandas and DuckDB.

What is the license for the Telco Customer Churn Dataset dataset?

This dataset is available under the Apache 2.0 license. See the full license at https://www.apache.org/licenses/LICENSE-2.0

How many tables are in the Telco Customer Churn Dataset dataset?

The Telco Customer Churn Dataset dataset contains 1 table: customers.

Telco Customer Churn Dataset

Last updated 2 months ago•December 27, 2025

Source:IBM Sample Data Sets via Kaggle

License:Apache 2.0

Version:1.0

Time:Point-in-time snapshot (synthetic data)

Location:Not applicable (synthetic data)

Created by Dataset Agent

Overview

The Telco Customer Churn Dataset is an industry-standard benchmark dataset created by IBM for customer analytics demonstrations. It contains telecommunications customer data designed for analyzing and predicting customer attrition—when customers stop doing business with a company. This dataset has become one of the most widely-used resources for learning churn prediction, with thousands of notebooks and tutorials built around it.

Why Churn Matters: Acquiring new customers costs 5-25x more than retaining existing ones. With industry-average telco churn rates of 15-25% annually, even small improvements in retention can significantly impact profitability.

The dataset contains 5,000 customer records with 21 features covering demographics, services, and billing information.

View Source

SQL

SELECT COUNT(*) AS total_customers FROM customers.csv

Data

Total Customers
5,000
1 row

The overall churn rate is 18.5%, with 925 customers having left the service—representing a moderately imbalanced classification problem.

View Source

SQL

SELECT COUNT(*) AS total_customers, SUM( CASE WHEN Churn = TRUE THEN 1 ELSE 0 END ) AS churned_customers, ROUND( SUM( CASE WHEN Churn = TRUE THEN 1 ELSE 0 END ) * 100.0 / COUNT(*), 2 ) AS churn_rate FROM customers.csv

Data

Total Customers	Churned Customers	Churn Rate
5,000	925	18.5
1 row

Feature Categories

The 21 features are organized into four logical categories that capture different aspects of the customer relationship:

Demographics: Gender, SeniorCitizen, Partner, Dependents — customer personal characteristics
Services: PhoneService, MultipleLines, InternetService, OnlineSecurity, OnlineBackup, DeviceProtection, TechSupport, StreamingTV, StreamingMovies — subscription details
Account: CustomerID, Tenure, Contract, PaperlessBilling, PaymentMethod, MonthlyCharges, TotalCharges — billing and account information
Target Variable: Churn — whether the customer left within the last month (Yes/No)

Key Insights and Churn Drivers

Analysis reveals several critical factors influencing customer churn. Contract type emerges as the single strongest predictor of customer retention.

View Source

SQL

SELECT Contract, COUNT(*) AS customer_count, SUM( CASE WHEN Churn = TRUE THEN 1 ELSE 0 END ) AS churned, ROUND( SUM( CASE WHEN Churn = TRUE THEN 1 ELSE 0 END ) * 100.0 / COUNT(*), 2 ) AS churn_rate FROM customers.csv GROUP BY Contract ORDER BY churn_rate DESC

Data

Contract Type	Customers	Churned	Churn Rate (%)
Month-to-month	1,603	569	35.5
One year	1,687	283	16.78
Two year	1,710	73	4.27
3 rows

Critical Finding: Month-to-month contracts show a 35.5% churn rate—more than 8x higher than two-year contracts (4.27%). This is the strongest predictor of churn in the dataset and a key target for retention strategies.

Customer Tenure and Value Analysis

Customers have an average tenure of 36.9 months (ranging from 1 to 72 months), with average monthly charges of $69.85 and customer lifetime value averaging $2,586.79.

View Source

SQL

SELECT ROUND(AVG(tenure), 1) AS avg_tenure, MIN(tenure) AS min_tenure, MAX(tenure) AS max_tenure, ROUND(AVG(MonthlyCharges), 2) AS avg_monthly_charges, ROUND(AVG(TotalCharges), 2) AS avg_total_charges FROM customers.csv

Data

Avg Tenure	Min Tenure	Max Tenure	Avg Monthly Charges	Avg Total Charges
36.9	1	72	69.85	2,586.79
1 row

Tenure shows a strong inverse relationship with churn—newer customers are significantly more likely to leave:

View Source

SQL

SELECT CASE WHEN tenure <= 12 THEN '0-12 months' WHEN tenure <= 24 THEN '13-24 months' WHEN tenure <= 48 THEN '25-48 months' ELSE '49-72 months' END AS tenure_bucket, COUNT(*) AS customers, SUM( CASE WHEN Churn = TRUE THEN 1 ELSE 0 END ) AS churned, ROUND( SUM( CASE WHEN Churn = TRUE THEN 1 ELSE 0 END ) * 100.0 / COUNT(*), 2 ) AS churn_rate FROM customers.csv GROUP BY tenure_bucket ORDER BY MIN(tenure)

Data

Tenure Range	Customers	Churned	Churn Rate (%)
0-12 months	1,525	412	27.02
13-24 months	823	178	21.63
25-48 months	1,102	189	17.15
49-72 months	1,550	146	9.42
4 rows

Demographic Patterns

Customer demographics reveal nuanced patterns in churn behavior. Senior citizens and customers without family connections show elevated churn rates.

View Source

SQL

SELECT CASE WHEN SeniorCitizen = 1 THEN 'Senior Citizen' ELSE 'Non-Senior' END AS segment, COUNT(*) AS customer_count, SUM( CASE WHEN Churn = TRUE THEN 1 ELSE 0 END ) AS churned, ROUND( SUM( CASE WHEN Churn = TRUE THEN 1 ELSE 0 END ) * 100.0 / COUNT(*), 2 ) AS churn_rate FROM customers.csv GROUP BY SeniorCitizen

Data

Segment	Customers	Churned	Churn Rate (%)
Non-Senior	4,016	724	18.03
Senior Citizen	984	201	20.43
2 rows

View Source

SQL

SELECT Partner, Dependents, COUNT(*) AS customer_count, SUM( CASE WHEN Churn = TRUE THEN 1 ELSE 0 END ) AS churned, ROUND( SUM( CASE WHEN Churn = TRUE THEN 1 ELSE 0 END ) * 100.0 / COUNT(*), 2 ) AS churn_rate FROM customers.csv GROUP BY Partner, Dependents ORDER BY churn_rate DESC

Data

Partner	Dependents	Customers	Churn Rate (%)
No	No	1,242	19.65
Yes	No	1,277	18.79
No	Yes	1,249	17.93
Yes	Yes	1,232	17.61
4 rows

View Source

SQL

SELECT gender, COUNT(*) AS customer_count FROM customers.csv GROUP BY gender

Data

Gender	Count
Female	2,555
Male	2,445
2 rows

Service and Billing Analysis

The dataset includes detailed service subscription information. Internet service type and payment methods show distinct patterns:

View Source

SQL

SELECT InternetService, COUNT(*) AS customer_count, SUM( CASE WHEN Churn = TRUE THEN 1 ELSE 0 END ) AS churned, ROUND( SUM( CASE WHEN Churn = TRUE THEN 1 ELSE 0 END ) * 100.0 / COUNT(*), 2 ) AS churn_rate FROM customers.csv GROUP BY InternetService ORDER BY churn_rate DESC

Data

Internet Service	Customers	Churned	Churn Rate (%)
DSL	1,646	327	19.87
Fiber optic	1,713	310	18.1
No Internet	1,641	288	17.55
3 rows

View Source

SQL

SELECT PaymentMethod, COUNT(*) AS customer_count, SUM( CASE WHEN Churn = TRUE THEN 1 ELSE 0 END ) AS churned, ROUND( SUM( CASE WHEN Churn = TRUE THEN 1 ELSE 0 END ) * 100.0 / COUNT(*), 2 ) AS churn_rate FROM customers.csv GROUP BY PaymentMethod ORDER BY churn_rate DESC

Data

Payment Method	Customers	Churned	Churn Rate (%)
Credit card (automatic)	1,241	239	19.26
Mailed check	1,269	238	18.75
Electronic check	1,229	230	18.71
Bank transfer (automatic)	1,261	218	17.29
4 rows

Model Performance Benchmarks

This dataset is well-suited for binary classification. Based on published results and common implementations, here are typical performance benchmarks achievable:

Class Imbalance Note: The 18.5% churn rate creates moderate class imbalance. Consider techniques like SMOTE, class weighting, or threshold adjustment to optimize for your business objective (e.g., prioritizing recall to catch more churners).

Feature Engineering Suggestions

To improve model performance beyond baseline, consider these feature engineering approaches specific to this dataset:

Total Services Count: Sum of all service subscriptions (PhoneService, InternetService add-ons) — more services often correlates with lower churn
Tenure Buckets: Convert continuous tenure to categorical (new: 0-12mo, established: 13-36mo, loyal: 37+ mo)
Average Monthly Spend: TotalCharges / tenure — normalizes for customer lifetime
Contract Risk Score: Combine contract type with tenure (month-to-month + low tenure = highest risk)
Service Bundle Flags: Create binary flags for common service combinations
Charge Increase Proxy: MonthlyCharges vs (TotalCharges/tenure) — may indicate recent price changes

Data Quality Notes

Known Issue: The TotalCharges column may contain blank/whitespace values for new customers with tenure of 0 months. Convert to numeric type and handle these as 0 or null values during preprocessing.

- [x] CustomerID is unique — suitable as index
- [x] No duplicate records in the dataset
- [x] Churn is cleanly encoded as boolean (Yes/No)
- [ ] TotalCharges requires type conversion from string to numeric
- [ ] Some service columns have "No internet service" values that need handling

Sample Data Preview

Sample Customer Records

#	Customer ID	Gender	Senior	Tenure	Contract	Monthly Charges	Churn
1	CUST-00000	Female	Yes	28	Two year	$112.17	No
2	CUST-00001	Female	Yes	39	Month-to-month	$33.65	No
3	CUST-00002	Female	No	32	Two year	$112.13	No
4	CUST-00003	Female	Yes	53	One year	$89.38	No
5	CUST-00004	Male	No	7	One year	$38.16	No
5 rows

View Source

SQL

SELECT customerID, gender, SeniorCitizen, tenure, Contract, MonthlyCharges, Churn FROM customers.csv LIMIT 5

Data

Customer ID	Gender	Senior	Tenure	Contract	Monthly Charges	Churn
CUST-00000	Female	Yes	28	Two year	$112.17	No
CUST-00001	Female	Yes	39	Month-to-month	$33.65	No
CUST-00002	Female	No	32	Two year	$112.13	No
CUST-00003	Female	Yes	53	One year	$89.38	No
CUST-00004	Male	No	7	One year	$38.16	No
5 rows

Dataset Versions

Multiple versions of this IBM dataset exist on Kaggle with different feature sets:

Limitations and Considerations

While excellent for learning and benchmarking, this dataset has important limitations to consider:

Synthetic Data: Created by IBM for educational purposes — patterns may not reflect real-world complexity
No Temporal Dimension: Point-in-time snapshot without time-series showing when customers churned
Limited Features: Real churn analysis often includes call detail records, customer service interactions, network quality metrics, and competitive offers
No Causal Information: Cannot determine why customers churned (price, service quality, competitor offers, life changes)
Single Geography: No regional variation that would exist in real telco data
Clean Data: Unrealistically clean compared to production data — good for learning, less realistic for production prep

This synthetic dataset was created by IBM for educational and demonstration purposes. While it reflects realistic patterns found in telecommunications data, it should not be used to draw conclusions about any specific real-world company or market.

Table Overview

customers

Contains 5,000 rows and 21 columns. Column types: 4 numeric, 5 text, 12 boolean.

5,000 rows21 columns

customers

5,000

rows

columns

Data Preview

Scroll to see more

customerID	gender	SeniorCitizen	Partner	Dependents	tenure	PhoneService	MultipleLines	InternetService	OnlineSecurity	OnlineBackup	DeviceProtection	TechSupport	StreamingTV	StreamingMovies	Contract	PaperlessBilling	PaymentMethod	MonthlyCharges	TotalCharges	Churn
CUST-00000	Female	1	false	false	28	true	true	DSL	false	true	false	false	false	true	Two year	true	Bank transfer (automatic)	112.17	3,140.76	false
CUST-00001	Female	1	false	true	39	false	true	Fiber optic	true	true	true	false	false	true	Month-to-month	true	Mailed check	33.65	1,312.35	false
CUST-00002	Female	0	true	true	32	false	true	No	true	true	false	true	false	false	Two year	false	Electronic check	112.13	3,588.16	false
CUST-00003	Female	1	false	true	53	true	false	No	false	true	false	true	true	true	One year	false	Bank transfer (automatic)	89.38	4,737.14	false
CUST-00004	Male	0	false	true	7	true	true	No	true	false	false	false	false	true	One year	true	Credit card (automatic)	38.16	267.12	false

Row 1

customerIDCUST-00000

genderFemale

SeniorCitizen1

Partnerfalse

Dependentsfalse

+16 more columns

Row 2

customerIDCUST-00001

genderFemale

SeniorCitizen1

Partnerfalse

Dependentstrue

+16 more columns

Row 3

customerIDCUST-00002

genderFemale

SeniorCitizen0

Partnertrue

Dependentstrue

+16 more columns

Showing 5 of 5,000 rows

Data Profile

5,000

rows

columns

100%

complete

5.0 MB

estimated size

Column Types

4 Numeric5 Text12 Boolean

High-Cardinality Columns

Columns with many unique values (suitable for identifiers or categorical features)

customerID(5,000 unique values)
TotalCharges(4,938 unique values)
MonthlyCharges(3,934 unique values)

Data Dictionary

customers

Column	Type	Description	Example
`customerID`	string	Unique identifier	"CUST-00000", "CUST-00001"
`gender`	string	Gender	"Female", "Female"
`SeniorCitizen`	numeric	Numeric value (range: 0 - 1)	1, 1
`Partner`	boolean	Boolean flag	false, false
`Dependents`	boolean	Boolean flag	false, true
`tenure`	numeric	Numeric value (range: 7 - 53)	28, 39
`PhoneService`	boolean	Boolean flag	true, false
`MultipleLines`	boolean	Boolean flag	true, true
`InternetService`	string	Internet Service	"DSL", "Fiber optic"
`OnlineSecurity`	boolean	Boolean flag	false, true
`OnlineBackup`	boolean	Boolean flag	true, true
`DeviceProtection`	boolean	Boolean flag	false, true
`TechSupport`	boolean	Boolean flag	false, false
`StreamingTV`	boolean	Boolean flag	false, false
`StreamingMovies`	boolean	Boolean flag	true, true
`Contract`	string	Contract	"Two year", "Month-to-month"
`PaperlessBilling`	boolean	Boolean flag	true, true
`PaymentMethod`	string	Payment Method	"Bank transfer (autom...", "Mailed check"
`MonthlyCharges`	numeric	Numeric value (range: 33.65 - 112.17)	112.17, 33.65
`TotalCharges`	numeric	Numeric value (range: 267.12 - 4,737.14)	3140.76, 1312.35
`Churn`	boolean	Boolean flag	false, false