Q2 churn concentrates in week-one drop-off
Self-serve churn ran at 5.8% a month in Q2 — well above the 1.6% we lose on the sales-assisted book — and it is not spread evenly. Six in ten self-serve accounts that left never finished week-one onboarding. Fix that one drop-off and the model says most of the lost revenue comes back.
source → billing export (Stripe), 2026-06-28+1 more
source → billing export (Stripe), 2026-06-28
Churn by cohort, six months
The two books have diverged since March. Sales-assisted churn holds near 1.6%, while the self-serve line has climbed almost every month, tracking release cycles that added signup steps without revisiting activation.
monthly logo churn rate by cohort, %
source → billing export (Stripe), 2026-06-28
Where the revenue and the risk sit
The self-serve book is most of the logos and a third of the revenue, but nearly all of the churn. Losing it quietly, one account at a time, is the failure mode — no single month looks alarming on its own.
| Cohort | Accounts | MRR | Monthly churn | State |
|---|---|---|---|---|
| Self-serve | 1,240 | $61k | 5.8% | degraded |
| Sales-assisted | 180 | $128k | 1.6% | healthy |
| Blended | 1,420 | $189k | 4.1% | degraded |
source → billing export (Stripe), 2026-06-28
What we ruled out
Price was not the driver: exit-survey “too expensive” held flat at 12% quarter over quarter, and downgrade-then-cancel paths accounted for fewer than 40 accounts.Downgrade-then-cancel = an account that dropped a tier before churning within 60 days; a proxy for value-fit, not price sensitivity. Nor was it one bad release — churn rose across three deploys, not a single regression. The signal that actually moved was activation. Accounts that reached their first published artifact in week one churned at 1.9%; those that did not churned at 14.2%.
source → product analytics funnel, W23–W26
The moment it surfaced
The activation split was not the question we started with — it fell out of the session halfway through, when the cohort averages refused to explain the spike.
The May spike doesn't line up with any release. What else moved?
Nothing on the release side — but splitting churned accounts by week-one activation does it: 1.9% for activated accounts vs 14.2% for the rest. The averages were hiding the split.
source → product analytics funnel, W23–W26
Decision needed this sprint. The onboarding rebuild competes with the billing migration for the same two engineers. If it slips past July, the Q3 self-serve plan carries roughly $14k of avoidable MRR loss that this analysis says is recoverable.
Next steps
- Make the week-one activation event the single north-star metric for the self-serve cohort.
- Rebuild the first-run flow to reach a published artifact in under five minutes, targeting 60% week-one activation.
- Trigger a lifecycle nudge for accounts still stalled before activation at day two and day five.
- Re-run this cohort read at the end of Q3 and compare against the 5.8% baseline.
- — 7/8 sections carry a verifiable agent layer, 6 cite sources
- — raw values kept behind every figure — the page stays re-queryable
- — deterministic layout from a declarative palette, zero executable code
- — stable section anchors — citations don’t rot when the page is edited
- — typeset: modular scale, single measure, one accent, body contrast ≥ 4.5:1
- ○ zero composition warnings