Credex Finance catches churning accounts 45 days before they cancel.
A churn scoring model trained on 18 months of cancellation history. At-risk accounts flagged automatically, retention sequences triggered without a rep lifting a finger.
Churn was visible in hindsight. Never in advance.
Credex Finance's customer success team was reactive by design, accounts that cancelled were reviewed after the fact, but the signals were always there weeks earlier. Login frequency dropping, support tickets spiking, payment delays creeping in. By the time a rep reached out, the customer had already decided to leave. There was no system for catching it early.
A model that reads the warning signs before the customer makes a decision.
Behavioural signals turned into daily scores. Retention workflows triggered automatically. The CS team focuses on high-value saves, not on manually scanning account lists.
Churn signals identified
Eighteen months of cancellation history analysed to surface the behavioural signals, login frequency, support tickets, payment delays, that preceded churn.
Feature engineering
Rolling usage metrics, recency and frequency scores, and support sentiment features engineered from raw event logs. Domain input from the customer success team shaped every predictor.
Scikit-learn classifier trained
Gradient boosting classifier trained on labelled historical data, validated on a 90-day holdout. Precision and recall thresholds agreed with the CS team before deployment.
Scores written to CRM
Model scores refreshed nightly and written directly into the account record in Postgres. CS reps see a churn probability score and contributing factors without leaving their dashboard.
Retention sequences triggered
Accounts crossing the risk threshold automatically enter an n8n retention workflow, personalised outreach, usage tip sequence, or escalation to account manager based on score band.
Outcome tracking closed the loop
Retention outcomes fed back into the training set quarterly. Model improves with every cohort rather than drifting against changing behaviour.
Retention is proactive now. Not a post-mortem.
Churn rate reduction
Of flagged at-risk accounts retained after entering the automated retention sequence in the first 6 months.
Early warning window
Average lead time between model flag and the point the customer would have cancelled without intervention.
Manual list reviews
CS reps no longer scan account lists. Every at-risk account surfaces automatically in their CRM view.
Days to production
From data handoff to a live scoring model writing daily risk scores into the Credex CRM.
The smallest stack that solved the problem.
Replace gut feel
with a model.
30-minute call. We'll audit your data and tell you exactly what is and isn't ready for a custom ML model, free, no deck.