Two onboarding flows, two retention curves: a side-by-side

Two onboarding flows, two retention curves: a side-by-side

A 2025 case study did something rare — ran two onboarding flows as parallel cohorts on the same 50,000-member Discord for 90 days and plotted the full retention curves, not just an endpoint.

What the data shows
— Flow A (passive: land in #general) lost 60% of members within 72 hours, then flattened to ~15% at day 30.
— Flow B (guided: 3-step setup, role pick, intro prompt) lost only 44% in 72 hours and held ~24% at day 30.
— Both curves became nearly parallel after day ~10 — the entire retention gap was set in the first 72 hours.

Why it happens
Retention curves for communities are front-loaded: the make-or-break window is the first few days, and interventions after that move little. Flow B's gain came almost entirely from reducing early abandonment, after which the two cohorts decayed at the same rate. This matches the broader pattern that lifecycle outcomes are largely determined at entry.

The caveat
Parallel cohorts are far stronger than before/after, but it's still one server and one niche. The curves flattening together suggests onboarding affects activation, not long-term loyalty — which is a ceiling, not a cure. Ninety days won't capture slow late-stage churn.

Open question: if the entire retention gap is decided in 72 hours, how much of the community-building energy spent on month-two engagement is simply misallocated?