New setup dashboard: first 4 weeks

Skipping is new: about 1 in 4 skips at least one step on day one (24.8%), resolving the checklist instead of leaving it to nag.

Teammate row counts each design's own invite actions: staff records before the 2 July invite revamp; invitations, direct adds and invite links after. The checklist's team step is not comparable across designs, so it is not shown.

Two earlier claims were corrected on 31 July after excluding the scam-signup waves. The client-creation lift (previously reported as significant) drops to +2.4 points, z 1.4: 63 of the 78 scam signups in the after cohort mass-created fake clients on day one. The teammate row flips from slightly up to slightly down: roughly half of the raw after-period "team actions" were scam accounts adding fake staff. Note creation and booking-link sharing sit slightly below baseline but within noise. The card decline is genuine — scam accounts click no cards — and has persisted a full month.

Top 2 cards take about 4 in 10 clicks. The bottom 4 feature cards earn roughly 2 clicks a day each across all users: show them only when relevant.

• Mobile setup completion still edges desktop (57.9% vs 56.5%), up 12.9 points on the old design.
• The mobile client-creation gap persists at a month (23% vs 31% desktop); mobile improved 3.8 points but that stays inside noise even at n = 309.
• The card decline is worst on mobile (28% to 10%), and mobile users skip more (31% vs 21%). The step panel likely pushes the carousel below the fold on phones.

Day N is the 24-hour window N days after each person's own signup; incomplete windows excluded (after cohort n = 977 for day 7). Day 7 always lands on the same weekday as signup, which is why it runs slightly above day 3 in the baseline.

• 14-day conversion, first mature read (signups 7 to 16 July, n = 674): 4.79% organic before, 4.30% after — within noise (z 0.5), and that cohort includes the weak launch week. The with-promos gap (7.92% vs 5.19%) is June's offer pushes, not the dashboard.
• 7-day conversion, mature cohorts (n = 1,064): 3.22% organic before, 2.44% after — within noise (z 1.1). Week 2 on its own hit 3.9% at 7 days (20 of 512), above the June baseline.
• Week 1 remains a real, isolated dip; the mature pre-June account control stayed normal through July, and the old design had a comparable one-week crater (8 June, 0.39%) with no release in sight.
• Stripe intent is holding at 7 days: 10.2% of after-release signups click connect Stripe within a week, against 6.0% before. The PostHog Stripe-completion event still never fires within 7 days of signup — completion truth stays with the Customer.io field.
• Scam signups do not change this verdict: none of the 142 excluded accounts subscribed or clicked a Stripe CTA, so they only shrink the denominators — spam-excluded 14-day organic is 5.1% before vs 4.6% after, still flat.

What this says about the April "3+ steps → subscription cliff": V3 pushed roughly 30 to 40% more people past 3 real steps, and conversion did not move. So far the cliff reads as intent-revealing rather than causal — people who were going to convert always did more setup. Caveat: at n = 674 the 14-day read can only rule out changes larger than about ±2 points, so a small effect either way is still possible; the weeks 2 to 4 windows maturing by 7 August will roughly double that sample.

• The old 'get started' PostHog dashboard (904713) now over-reads: feature completions rose about 5 times for mechanical reasons.
• Any metric counting the team step as a completion shows a false decline.
• No holdout: this is before-and-after, not a controlled test. Regional mix was stable (72 to 77% Americas both periods).