Turn Amplitude's churn-risk cohort into a weekly CS action list

Every Monday at 9am, turn our Amplitude churn-risk cohort into a ranked customer success action list. The cohort name should be a user-configurable input on the prompt page (default: "Churn Risk") so teams can point it at whatever behavioral cohort they already maintain in Amplitude.

Step 1 — Find the cohort. Use Amplitude's List Cohorts to locate the behavioral cohort matching the configured name (case-insensitive substring match so "Churn Risk", "churn-risk", or "Churn Risk – Enterprise" all resolve). If nothing matches, post a one-line Slack notice explaining that the cohort could not be found and stop.

Step 2 — Export the cohort. Amplitude's cohort export is asynchronous, so use Request Cohort Download to start the export, then poll Check Cohort Download Status until the job reports JOB COMPLETED (wait a few seconds between polls, and give up after roughly 5 minutes with a Slack error notice if the job never completes). Once complete, call Download Cohort File to pull the CSV of at-risk user IDs and their user properties. If the cohort is empty this week, post a one-line "no at-risk accounts today" message to the CS Slack channel so the team knows the check ran, then stop.

Step 3 — Classify each at-risk user. For every row in the CSV, an AI customer success analyst reads the user's Amplitude properties (last-active date, session counts, feature usage, support signals, plan tier) and assigns exactly one risk tier: drop-in-usage (login/session frequency has dropped sharply), feature-abandonment (adopted a core feature but stopped using it), or support-frustration (recent support tickets, negative signals, error rates). Also draft a two-sentence "why they might be leaving" note per account, grounded in the specific properties you saw, not generic filler.

Step 4 — Tag each user in Intercom. Match each Amplitude user ID to the corresponding Intercom contact (assume they share the same external ID; if a contact can't be resolved, skip it and note it in the digest tail). Then use Intercom's Tag a Contact operation to apply the tag matching the risk tier — "churn-risk-drop-in-usage", "churn-risk-feature-abandonment", or "churn-risk-support-frustration" — so downstream Intercom campaigns and audiences pick them up automatically.

Step 5 — Post the Slack digest. Use Slack Bot's Send a Message to post a ranked digest to the configured CS channel. Include the total count of at-risk users, a short overall summary, and then the top 20 accounts ranked by severity. For each of the top 20 show: account/user identifier, risk tier, the two-sentence hypothesis, and a direct link to the Intercom contact page (https://app.intercom.com/a/apps/{workspace}/users/{intercomContactId}). Keep the message scannable with Slack markdown headers and bullets. If any Intercom lookups failed, list them at the bottom under "Could not resolve in Intercom" so the CS team can fix the mapping.

Inputs the user should be able to configure on the prompt page: the Amplitude cohort name (default "Churn Risk"), the Slack channel to post to, the Intercom workspace slug used in contact URLs, and the maximum number of accounts shown in the ranked list (default 20).