Who is leaving, and why?
A retention team's dashboard on the IBM Telco sample: where churn concentrates, how long customers stay, what goes with leaving, and what an offer might change. Filter or click to dig in.
Customers
7,043
of 7,043
Left last month
1,869
churned
Churn rate
26.5%
overall 26.5%
Monthly revenue lost
$139.1k
31% of this group's bills
Tenure, left vs stayed
18 vs 38
months, average
Bill, left vs stayed
$74 vs $61
per month, average
How long customers stay · by contract
Kaplan–Meier survival: the share still a customer after each month, from tenure and who left
What goes with leaving
Churn rate for each value of each feature, features with the biggest gaps first; dot size = customers. Click contract, internet or payment to filter.
Where churn concentrates
Churn rate by contract and tenure; customers in each cell. Click a cell to filter.
Monthly bill
Share of stayers (grey) and leavers (red) at each bill level
Revenue walking out
Monthly bills of customers who left, by contract and internet
Top cell: 72% of all lost monthly revenue.
Highest-risk segments
Combinations of contract, internet, payment and tenure with 40+ customers, by churn rate. Groups, never individuals.
| Contract | Internet | Payment | Tenure | Customers | Churn | Lost / month |
|---|---|---|---|---|---|---|
| Month-to-month | Fiber optic | Bank transfer | 0–12 months | 87 | 82% | $6.0k |
| Month-to-month | Fiber optic | Electronic check | 0–12 months | 631 | 71% | $37.2k |
| Month-to-month | Fiber optic | Mailed check | 0–12 months | 123 | 63% | $6.2k |
| Month-to-month | Fiber optic | Credit card | 0–12 months | 75 | 61% | $3.9k |
| Month-to-month | Fiber optic | Electronic check | 13–24 months | 255 | 59% | $13.4k |
| Month-to-month | DSL | Electronic check | 0–12 months | 282 | 50% | $6.3k |
| Month-to-month | Fiber optic | Electronic check | 25–48 months | 286 | 49% | $12.9k |
| Month-to-month | Fiber optic | Credit card | 13–24 months | 62 | 45% | $2.5k |
What if we act?
The Telco project's logistic regression (ROC AUC 0.8347). Age is held fixed and never asked.
Chance this customer leaves
74%
Offer a one-year contract
74% → 57% (-17 pts)
Move them to automatic card payment
74% → 66% (-8 pts)
Add tech support
74% → 65% (-9 pts)
Add online security
74% → 63% (-11 pts)
A model of association, not a promise: offers can change who stays for other reasons too.
What the churn says
Written from the same numbers as the tiles, for the current filters
- 26.5% of these 7,043 customers left last month, taking $139.1k of monthly bills with them (31%).
- The biggest gap is contract: Month-to-month churns at 43%, Two year at 3%.
- After a year, 70% of month-to-month customers are still here, against 100% on two-year contracts.
- Leavers had been customers for 18 months on average (stayers 38) and paid more: $74 vs $61 a month.
- Riskiest group: month-to-month, fiber optic, paying by bank transfer, 0–12 months: 82% of 87 left.
Behind the dashboard
The data
IBM Telco Customer Churn (sample data, fictional customers): 7,043 customers of a telecom company, with their services, contract, billing, tenure and whether they left within the last month.
What's left out
Customer IDs, gender, age (senior citizen) and household fields (partner, dependents) are dropped before the data reaches this page. Everything shown is something the company can act on.
Churn rate
Customers who left ÷ customers in the group. Revenue lost = their monthly bills.
Survival curves
Kaplan–Meier: each month, the share of customers still at risk who did not leave, multiplied up. Customers who stayed count until their current tenure.
What-if
The logistic regression from the Telco churn ML project, run in the browser. It shows association, not cause.
How the model was chosen and tested: Telco churn project.