The challenge
A key US telecom client had seen customer churn rise over two years. The US telecom market is saturated, so keeping existing customers matters as much as winning new ones.
What we did
- Identified every factor that could plausibly influence churn.
- Ran univariate and bivariate analysis to build churn segments.
- Built logistic regression models to quantify each driver.
- Used CHAID decision trees to find pockets of high and low churn.
The result
The client gained a clear view of what causes customers to churn and began acting on our recommendations to reduce it.
How we would build it today
- Gradient-boosted churn propensity scores refreshed daily in the lakehouse.
- Uplift models that target retention offers at customers who can still be persuaded, not those who would stay anyway.
- A GenAI retention assistant that drafts save offers for care agents, grounded in each customer's history.
This describes our current toolkit for the same problem. It is not part of the original engagement.