Your AI CEO ships the cohort table you keep meaning to build.
You'd think this needs a data analyst and a Mode dashboard — Astra ships a fresh cohort table to your Lark every Monday at 9am.
Astra runs a weekly cohort retention analysis by pulling signups from Stripe and behavioral events from PostHog, then bucketing users by signup week and tracking week-1, week-2, week-4, week-8, and week-12 retention. Every Monday she ships a Lark message with the full cohort triangle, week-over-week deltas highlighted, and the one cohort that's leaking worst with a hypothesis on why — usually traced to a specific traffic source or a product change that shipped during their signup week. You stop wondering whether retention is actually improving. You see the curve, you see which acquisition channels produce sticky users, and you get a concrete "this week, fix X" call-out rooted in real numbers. No SQL. No data hire. No more pretending the screenshot from 3 months ago still represents your product.
Stripe customer.created events for the cohort definition. PostHog session events for the activity definition (default: any session = retained, configurable to feature use).
Group by signup week. Compute week-1, w2, w4, w8, w12 retention as % of cohort still active. Compute WoW deltas vs the prior cohort.
Tag each cohort with dominant acquisition channel (GA4 source) and any product changes that shipped that week (Vercel deploy log + Linear).
Identify the cohort with the worst retention vs the 8-week trailing average. Generate hypothesis: which channel? which product change? which onboarding step?
Lark message Monday 9am with the cohort triangle (rendered table), the worst cohort highlighted, the hypothesis, and one concrete action for this week.
A Lark message every Monday 9am with the full cohort retention triangle, WoW deltas highlighted, the worst-performing cohort called out with a hypothesis, and one concrete action for this week.
Weekly Monday 9am, plus on-demand reruns when you ask.
We'll spin up your workspace, hand the prompt to Astra, and you see the answer in 60 seconds. Free.
Try this with AstraClear answers about wallet credit, usage, subscriptions, and how Tycoon charges for work.
Yes. Astra works with Mixpanel, Amplitude, GA4 events, or even just Stripe (login = active). With Stripe-only she can do payment retention (still on a paid plan w8 vs paid w1) which is the sharpest business signal anyway.
Default: any session in week N counts as retained. You can override to a specific event (used core feature X, sent message, completed task) — that's typically the more honest number. Astra will show both side-by-side if you want.
Example: "Cohort week of March 4 has w4 retention of 31% vs trailing 8-wk avg of 47%. 68% of that cohort came from a Twitter post about feature X. Retention drops sharpest at day 9 — that's when feature X's free tier limit hits. Likely cause: misleading expectations from the Twitter copy. Action: rewrite the Twitter post or extend trial."
Yes. Tell Astra the filter ("only Pro plan", "only users from organic search", "only users who hit feature X in week 1") and she re-runs the cohort against that segment. You can have multiple recurring versions side by side.