Where a member came from predicts how long they stay

Where a member came from predicts how long they stay

We obsess over what happens after the join. The invite-source data says a lot is decided before it.

What the data shows
When joins are tagged by invite source (specific friend invite, partnered server, public listing, viral link, paid promo), retention diverges sharply by origin. Members from a personal invite or a topically-aligned partner server retain far better than those from public discovery listings or broad viral links — often a multiple difference in 30-day survival, consistently across studied servers.

Why it happens
A friend invite carries pre-built social tie and pre-qualified intent — the member arrives with a reason and a relationship. Discovery-listing and viral joiners arrive curious but unattached, so they default to the baseline drift rate. The source encodes intent, and intent is most of retention. Telegram shows the same pattern: members from a trusted cross-post or referral outlast members from a folder-listing or mass shoutout.

The caveat
Source is correlated with everything (the kind of person who clicks a viral link differs broadly), so it's a bundle of confounds, not a clean lever. And invite-tagging is imperfect — many joins are untracked.

Open question: should acquisition strategy weight retained members per source over raw joins — and would a server grow slower-but-bigger by ignoring its highest-volume, lowest-quality channel entirely?