Finding the Aha Behavior: A Host Cohort Audit
Objective: Given a synthetic export of 30 Airbnb hosts' first-14-day behaviors and whether each host was still active at day 90, apply the lesson's cohort-comparison method to identify which single behavior is the real activation event, and set the time window that captures it.
You're a growth analyst on Airbnb's host-growth team. Leadership wants one activation metric for new hosts instead of five loosely tracked onboarding steps. You've pulled a 30-host synthetic export with first-14-day behaviors and each host's day-90 status.
Compare the retained-at-day-90 cohort against the churned cohort, find the earliest behavior that actually separates them, and write a one-line activation definition an engineer could instrument tomorrow.
Before you start
What you'll need
Free path (everything below is enough to finish)
No account friction, pivot tables handle a 30-row cohort split easily
The process
1 step
Step 01 of 01
The lesson's Step 1 says: pull the retained-past-day-30 cohort and the churned-before-day-30 cohort, then look for the earliest behavior that separates them. The activation event is not the behavior most hosts do, it is the behavior that correlates with sticking around.
22 of 30 hosts synced their calendar in week 1, but only 11 hosts uploaded 5+ photos in week 1. Which behavior actually separates the day-90-retained hosts from the churned ones, and why does the more 'popular' behavior lose?
Procedure
- Import the export and freeze row 1.
- Split hosts into two groups: day_90_active = TRUE (18 hosts) and FALSE (12 hosts).
- For each behavior column (calendar_synced, photos_5plus, first_booking_accepted, pricing_tool_used, messaged_guest_first), compute the % TRUE within each group.
- Rank behaviors by the gap between the two group percentages, largest gap first.
- Confirm the winning behavior happens early enough (within 14 days) to be useful as a leading indicator, not a lagging one.
Behavior gap analysis (n=30 hosts) calendar_synced: 73% (active) vs 67% (churned) -> 6pt gap photos_5plus: 61% (active) vs 58% (churned) -> 3pt gap first_booking_accepted: 89% (active) vs 25% (churned) -> 64pt gap pricing_tool_used: 44% (active) vs 33% (churned) -> 11pt gap messaged_guest_first: 39% (active) vs 17% (churned) -> 22pt gap Winner: first_booking_accepted, 64-point gap, occurs at a median of day 9.
Healthy
One behavior shows a gap of 40+ points between the retained and churned cohorts, and it happens inside the target window.
Unhealthy
The 'busiest' behavior (calendar_synced, done by almost everyone) shows only a 6-point gap. Building onboarding around it would optimize for attendance, not value.
What this means
Accepting a first booking, not syncing a calendar, is the aha moment: it's the point where a host experiences real income, which is what makes them stick. Calendar sync is a setup step everyone does regardless of outcome.
So what do I do about it?
| Symptom | Action | Effort |
|---|---|---|
| Onboarding funnel currently celebrates 'calendar synced' as the activation milestone | Re-point the activation event to first_booking_accepted, and redesign onboarding to accelerate getting a first booking (better default pricing, instant-book nudges) | half day |
| No single frozen definition exists, each team cites a different onboarding step | Write the definition down: 'Activated = accepted first booking within 14 days of listing published' and circulate it as the one number every team reports against | 30 min |
Final deliverable
A one-line activation definition (event + threshold + window), backed by the cohort gap analysis, ready to hand to engineering for instrumentation as a single `activated` event.
See a reference example
HelloFresh, new-subscriber activation definition (excerpt) Candidate behaviors tested against day-30 retention (n=45 synthetic subscribers): box_customized: 8pt gap delivery_rescheduled: 4pt gap second_box_ordered: 71pt gap <- winner Frozen definition: 'Activated = ordered a second box within 21 days of first delivery.' Owner: Retention pod. Reviewed quarterly against fresh cohort data.
Success criteria
You're done when you can:
- Correctly computes the retained-vs-churned percentage gap for all 5 candidate behaviors
- Selects the behavior with the largest gap, not the most frequent behavior
- Writes a frozen definition with an explicit event name, threshold, and time window