2024 · 3 months·Lead Data Scientist
Early-warning system for customer churn
Built a signal that flagged at-risk accounts 6 weeks before churn, giving the retention team time to act.
-14%
Churn reduction
6 wks
Lead time
120k
Accounts covered
Context
Retention was reactive — the team only learned an account was leaving after it had already lapsed. Leadership needed an early signal they could trust and a clear story about why accounts leave.
Approach
- 01Framed the problem with stakeholders: defined churn precisely and aligned on what a usable lead time looked like.
- 02Engineered behavioral features (usage decay, support friction, billing events) from event-level logs.
- 03Validated drivers with causal checks, not just correlation, so the narrative held up under scrutiny.
- 04Packaged the findings into a one-page narrative + dashboard the retention team reads weekly.
Impact
The story changed how the org thought about churn — from a billing problem to an engagement problem. The model now drives a weekly outreach list and cut churn 14% in the first two quarters.
Causal inferenceForecastingStorytelling