Match the intervention to how at-risk a user actually is

A 0 to 100 score from how sessions are trending over 30, 14 and 7 days, features falling out of use, the direction of support sentiment, billing signals, what similar accounts did, and account level usage for B2B. Banded low 0 to 29, medium 30 to 59, high 60 to 79 and critical 80 to 100, and recomputed on any significant event so a sudden drop is caught quickly.

All paying users scoring 30 or more, held in bands by that score. The first 14 days are excluded because early signals are unreliable, and so is anyone already in a save flow.

Medium gets helpful content about the features that keep people like them, plus a link back to where they left off. High gets a direct offer of help from their CSM and a 15 minute walkthrough. Critical gets a personal letter from the founder with a calendar link. Sent from success, from the named CSM, and from the founder respectively.

High risk users see a banner pointing at the path most people in their position succeed with. Critical users get a full screen note from their CSM or the founder offering a call. Only when the signal is strong, never to everyone.

For high and critical users, the features their cohort uses successfully and they have never tried, ranked by how strongly each one correlates with staying. It shows on the dashboard, inside emails and in page slots.

Bands are read at entry and again weekly. Low gets one light email on day 7. Medium gets a tip on day 0, in app help at the next login, and recommendations on day 7. High gets a CSM email on day 0, an urgent banner at the next session, and a same day CSM task. Critical gets all of it inside four hours: the founder's email, an agent offering a call, a CSM task and an executive sponsor task. People who improve drop out, people who worsen move up a band. They leave once the score stays under 30 for 14 days, on cancellation, or on unsubscribe.

How users spread across the bands each week, the save rate in each one, how fast a human reaches critical users against a four hour target, the ARR saved per band, which way accounts are migrating, and predicted against actual churn at 30, 60 and 90 days.