Monthly churn-risk report of users who have gone quiet

On the first of every month, at around 8am, pull my Clerk user base and email me a churn-risk report of the users who have gone quiet. Run this as an agent so it can rank the at-risk users, sort them into tiers, and draft short outreach recommendations based on what it finds, rather than just dumping a list of user IDs.

First, use Clerk's List all users operation to pull the full user base. Page through every user (Clerk uses offset-based pagination, up to 500 per page, and GET /users returns a bare array with no total count, so keep paging until a short page comes back) so nobody is missed. For each user, capture the id, name and email, created_at (their tenure), last_active_at, and last_sign_in_at. You can pass order_by=last_sign_in_at to surface the quietest accounts first.

When computing the inactivity window, remember that Clerk timestamps (created_at, last_active_at, last_sign_in_at) are Unix epoch milliseconds, not seconds, so convert them before doing any date math. For each user, measure days of inactivity from the more recent of last_active_at and last_sign_in_at, and flag anyone who has not signed in or been active for 30 or more days. Exclude users who never completed signup: skip any account that has no last_sign_in_at at all, since those users never actually signed in.

Next, rank the flagged users. Give more weight to longer-tenured accounts (older created_at) and to higher-value accounts, for example users who belong to an organization or carry richer public metadata. Group them into clear tiers such as High risk, Medium risk, and Watchlist, based on how long they have been inactive and how much they matter. For the top accounts in the highest tier, write a short recommended re-engagement note (two or three sentences) tailored to why the account matters and how long it has been gone.

Finally, email the brief to me using Gmail's Send a Message operation. Structure it so it is skimmable in one read: a one-line summary of how many users are at risk, the tiers with their counts, a prioritized table of the top at-risk users (name and email, tenure, days inactive, and why they are a priority), and the recommended outreach notes for the top handful. Keep everything in a single monthly email I can act on.