Behavior-Based Retention Engine
Watches product behaviour, predicts who is about to churn, and launches the right save-play before they leave.
−31%
Churn
18 days early
At-risk detected
40+
Saves / month
6-figure
Revenue retained
The problem
Churn was only discovered after the cancellation email. By then, it was over.
What I built
Behavioural signals (usage drop, ignored emails, support tone, billing failures) feed a risk score. Above threshold, the engine picks a save-play: an AI-written check-in, a human call task, a discount offer, or an onboarding re-run — and measures which play actually saved them.
Architecture
4 stages- 1
Event ingestion: product usage, email, support, billing
- 2
Rolling risk score per account (weighted decay)
- 3
Play selector → AI message / human task / offer
- 4
Outcome tracking → the model learns which play works
Stack
Got a process like this eating your team's week? Let's kill it.