It connects to Stripe and pulls 13 behavioral features: payment pattern shifts, invoice failure trends, spending changes compared to a customer's first few months, and whether a single user is the only one active on the account. That last one, "champion dependency," is one of the strongest signals we've found. When an account has one active user and that person goes quiet for 18+ days, cancellation usually follows.
The model also does MRR decomposition to separate new-customer revenue from existing-customer revenue. This catches hidden churn, the revenue loss that gets buried under acquisition growth. You can be up 8% this quarter while 12% of your existing base is quietly leaving.
Once it scores every account, it acts on it. Dunning sequences when payment behavior degrades, outreach when risk spikes, and save offers tailored to the specific risk driver (a payment-issue customer gets a different response than a disengaged one).
5-minute Stripe connection, first scores come back immediately. Optional JS snippet or Segment integration if you want product usage signals on top of billing data.
https://churnburner.com
Curious what behavioral signals others have found predictive of churn.