← All projects

Subscriber Churn Agent

What it does

A nightly job that scores active subscribers for cancellation risk, explains why, and proposes one retention action from a fixed menu. Every proposal passes through a deterministic policy layer before anything can be sent. A review console lets people approve actions, handle adverse-event reports and edit policy without code.

It runs on synthetic data in dry-run mode, with failure drills documented.

The nightly loop
  1. Subscriberssynthetic data
  2. Risk modelcalibrated score + reasons
  3. Proposalone action from a fixed menu
  4. Policy layerdeterministic, decides what may send
  5. Review consolepeople approve or reject

Tools

Python, uv, a calibrated risk model, Claude, Render