Churn Prediction & Retention
Know who's about to leave, why, and what will make them stay.
A churn and retention platform that predicts who is at risk and when, explains the drivers, and identifies which customers will actually respond to an intervention.
01 · Overview
The decision was made weeks ago
By the time a customer closes an account or cancels a contract, the decision was made weeks earlier. Retention teams either react too late or spread discounts across everyone, including customers who were never going to leave.
Built with
- Survival models
- Reason codes
- Uplift modelling
Deployment
On-premise, private cloud or in-region.
02 · How it works
Predict, explain, target
- 1
Read the early signals
Usage decline, payment delays, support contacts and sentiment, falling product breadth, and competitive signals.
- 2
Predict if and when
Gradient-boosted models predict churn probability; survival models estimate how long you have to act.
- 3
Explain the driver
Per-customer reason codes show whether it is price, service, product fit or a life event, which tells you which intervention to use.
- 4
Target the persuadable
Uplift modelling finds customers whose behaviour an offer will actually change, so budget isn’t wasted.
- Usage, billing, supportPlus transactions
- Behaviour features200+ early signals
- Churn & survival modelsIf and when
- Reasons + upliftWho will respond
- CRM & campaignsOutcome feedback ↺
03 · Capabilities
Where it’s used
- Bank account and card attrition
- Subscription and contract renewal risk
- Retail customer win-back
- Relationship-manager prioritisation lists
04 · Outcomes
What we measure
- Lower churn
- Churn rate in treated vs. control group
- Better retention ROI
- Retained revenue per unit of incentive
- Earlier intervention
- Lead time between alert and churn date
- Less wasted discount
- Incentives given to customers who would have stayed
Example scenarios
Illustrative- Retail bankProactive relationship-manager outreach before account closure.
- Mobile operatorTargeted offers for at-risk post-paid customers.
- SaaS companyHealth scores that trigger customer-success playbooks.
- InsurerRenewal-risk scoring ahead of policy expiry.
05 · Governance
Fair, measurable retention
Controls built in
- Holdout groups to measure true lift
- Reason codes on every prediction
- Fair-treatment rules on offers
- Model monitoring for drift after pricing or product changes
Technology
- Python
- XGBoost
- LightGBM
- lifelines
- Uplift models
- Airflow
- PostgreSQL
- Looker
06 · FAQ
Common questions
Do we need a data science team to use it?
No. Scores, reasons and recommended actions flow into your CRM or campaign tools.
How far ahead can it predict churn?
Typically 30–90 days, depending on how early the signals appear in your data. Survival models tell you how long you have to act.
How do we know the retention offers actually worked?
Every campaign keeps a holdout group, so you see the incremental effect rather than customers who would have stayed anyway.
