TL;DR. Nova paid churn has roughly doubled in 6 weeks (~20 → 55/wk), and as a share of the base the rate went from ~2%/wk to ~4.6% (~6.6% including past-due). It is one narrow cohort: new-ish, monthly, Grow ($99), native-Nova customers hitting their first or second charge — and ~43% of it is failed payments. Recovery is broken (~32% of renewals fail, only ~13% recover vs a healthy 30–70%). Two levers cover almost all of it: fix Paddle dunning and drive activation before the day-37 first bill.
| Week | Paid churn | of which involuntary |
|---|---|---|
| Jun 29 | 26 | 0 |
| Jul 6–20 | ~20/wk | 0–2 |
| Jul 27 | 29 | 8 |
| Aug 3 | 46 | 14 |
| Aug 10 | 55 | 18 |
Churn rate ≈ 2.1%/wk → 4.6%/wk (55 on ~1,205 subs), ~6.6%/wk including past-due exits. Real worsening, not just a bigger base.
Stripping out trial-window cancels (≤14 days), the real paid-churn cohort is 295 cancels over the last 45 days (~46/wk) — and it's remarkably uniform. It is essentially the same customer over and over:
| Dimension | The cohort | Share |
|---|---|---|
| Billing cadence | Monthly (annual almost never churns) | 99% |
| Base | Native Nova — not Classic/Verifone migrants | 98% |
| Plan tier | Grow ($99) — only 5 of ~205 priced were higher | ~98% |
| Age at churn | Day 15–45 (first monthly bill ~day 37) | 58% |
| Age at churn | Day 46–90 (2nd–3rd bill) → ~88% gone by 90d | 30% |
| Reason | manual (voluntary) | 53% |
| Reason | payment_failure (involuntary) | 43% |
| Geography | Paid in a non-USD currency (higher card-decline) | 30% |
| Exit survey | Skipped; among answers, "Not using it enough" leads | ~85% skip |
The takeaway: churn is not spread across the base — it's concentrated in new, monthly Grow customers who hit their first real charge before the product has proven its value, and nearly half exit via a failed payment. That narrowness is good news: two targeted fixes address almost all of it.
Age-at-churn is bimodal — two spikes at the monthly bills:
| Age at churn | Cancels (45d) |
|---|---|
| 15–45d (first monthly bill, ~day 37) | 170 |
| 46–90d (2nd–3rd bill) | 89 |
| 90d+ | 36 |
~⅔ of paid churn happens at the first two renewals — a value-realization problem at the bill.
Comparing the churned paid cohort with retained payers on activation (both corrected for billing-unpublish, which nulls publishedAt when a paid plan lapses):
| Signal | Churned | Retained |
|---|---|---|
| Completed onboarding | 98% | 99% |
| Completed builder tour | 86% | 94% |
| Ever published a page | 69% | 86% |
| Avg pages created | 3.0 | 9.6 |
| Avg pages published | 2.0 | 6.6 |
| Connected a custom domain | 19% | 47% |
| Set up a lead-capture form | 56% | 69% |
| Never built anything | 10% | 6% |
Onboarding completion is a vanity metric. 98–99% of both cohorts finish onboarding and the tours — it does not predict retention. The real fork is depth and commitment: retained payers build ~3× more pages and are 2.5× more likely to connect their own domain (47% vs 19%) — the "I'm running my real business on this" signal. Churned payers mostly did try it (69% published, only 10% never built a page) but stayed shallow and hit the day-37 bill unconvinced.
They go dormant in week 1 — long before the bill. Anchoring last-edit activity to signup day (not cancel day) shows the drop-off is immediate: the median churned payer's last edit was day 1. They set something up in the first session and never came back.
| Last time they touched the product | Users | Share |
|---|---|---|
| Never built anything | 29 | 10% |
| Week 1 (days 0–7) | 168 | 57% |
| Week 2 (8–14) | 22 | 7% |
| Week 3 (15–21) | 20 | 7% |
| Week 4 (22–30) | 13 | 4% |
| After day 30 (past the bill) | 43 | 15% |
Two-thirds either never built or last touched in week 1, and 74% of everyone who built anything had gone dark ≥2 weeks before the day-30 bill; only ~15% were still active past the first charge. So this is not "used it, hit renewal, decided it wasn't worth it" — it's a failure to form a habit in week 1, with the bill simply formalizing a decision made ~30 days earlier. The intervention window is unambiguous: week 1 is the whole game, and a "paid, no page edit in 7–10 days" flag catches the drop-off while it's still recoverable. (Edit activity, not logins — a set-and-forget customer looks the same here, but averaging ~3 pages / 19% domain and citing "not using it enough" limits that slice.)
Failed-payment churn is native Nova, not Verifone→Paddle migrants (only 7 of the 295 are migrants), and 100% Paddle (Nova bills entirely through Paddle). It concentrates in Grow ($99) and in the recent signup cohorts now hitting their first/second charge — which explains the sudden late-July onset. So it's the everyday Nova base failing at renewal, not a migration regression. Current past-due pipeline: ~403 subs ≈ ~$10k/mo, Grow-dominant.
From the Paddle recurring-renewal export (2,956 renewals since Jun 15):
| Renewal outcome | Count | Share |
|---|---|---|
| Completed first try | 2,010 | 68% |
| Recovered after retry | 124 | 4% |
| Still stuck in past-due | 403 | 14% |
| Canceled (gave up) | 419 | 14% |
~32% of renewals fail on first attempt, and only ~13% of failures recover — versus a healthy dunning benchmark of 30–70%. The involuntary churn is recoverable failures we're not recovering.
Track A · Involuntary / dunning Owner: Paddle/billing (Aziz) — the fast, ~$10k/mo win:
Track B · Voluntary / first-renewal Owner: Product / CS: