AI Churn Intervention

The build

The signals that predict churn already exist — they are just scattered. Usage from Airtable, account status from HubSpot and support-ticket sentiment from Gmail merge into one record per customer. A model scores risk and returns a written rationale rather than a bare number, and the tier decides the response: low risk is logged silently to avoid alert fatigue, medium risk gets a drafted check-in held in Slack for approval, high risk escalates to the account owner with the reasoning attached. The design rule is the point — the model reasons and drafts, a person decides, and nothing reaches a customer automatically.