​

V4 hero CTA revision: start free before connecting accounts

​

Decision

Change only V4’s signed-out hero CTA from Connect your accounts free to Get started free. The destination remains /sign-up. Creating an account is the next step; connecting a financial institution comes later. Keep V3, both headlines, the V4 product demonstration, page layout, navigation, lower-page CTAs, pricing, and authentication behavior unchanged. Extract the V4 hero into components/landing-v4/hero.tsx without changing its presentation. This tests the initial commitment expressed by the button, not a new page design.

​

Evidence used in the September 5, 2026 review

Source: user-supplied Landing V4 Decision Brief, Part II, pinned window [2026-09-02 14:48:19, 2026-09-04 23:30:00) UTC. These reported counts were not independently re-queried in PostHog. The shared artifact initially served the old report; the attached HTML contains the reconciled tables. Desktop completed signups were 15/151 vs 14/140, despite lower V4 hero clicks. The priority is more completed signups per visitor, not moving clicks between routes. V4’s five signups through /sign-in + Google must remain in the outcome. The signup differences are inconclusive; no uplift is promised by this change. The local V4 hero CTA is fully above the fold at 402 × 700 (top approximately 402 px, bottom 458 px). Both variants already have a sticky mobile signup button. This weakens the missing-button explanation. Android Chrome’s reported V4 LCP p75 is 3.91 s vs 2.90 s for V3; investigate the actual LCP element and loading phases separately before attributing the difference to animation. Remaining report limitations: CTA-path subtotals exceed their unique totals; channel signup counts do not sum to the reconciled signup totals; activation uses signup denominators rather than exposed-visitor denominators; replay findings come from event streams, not visual review. None justifies changing the signup form, traffic allocation, or lower-page content in this revision.

​

Revision and release boundary

V4 events now include landing_revision: 'hero-cta-v2' in addition to the existing variant: 'v4'. The property is event-local, never a persistent PostHog person/superproperty that could leak onto V3 or signup events. View, engagement, scroll, section, and showcase dedupe keys include this revision so a browser that previously saw the old V4 can record the new exposure. This is a named iteration of the existing landing-page-conversion-v01 test arm, not a newly randomized old-V4 vs new-V4 experiment. The PR does not change the feature flag, traffic split, or production deployment. Before production rollout, the experiment owner must archive the original read and record the exact release time. The original September 19 sizing/read plan does not transfer automatically to the new revision. After release:

1. Start a new analysis window at the recorded production release. Include only randomized control/test exposures. V4 must also carry this revision; compare with V3 first exposures from the same time window. Exclude unbucketed visitors, QA, and anyone already registered at exposure.
2. For the primary acquisition analysis, exclude visitors previously exposed to the original experiment in either arm, symmetrically. First-exposure assignment stays fixed; report crossovers separately. Prior anonymous activity is not proof of an existing account—label any “no activity in 30 days” restriction as an additional new-visitor segment.
3. Count user_signed_up after exposure within a predeclared, equal follow-up window, regardless of whether the user entered via /sign-up, /sign-in, the hero, navigation, or pricing. Do not require a click/page-view event to qualify a signup. Do not require landing_revision on signup events; join outcomes to the person’s qualifying exposure.
4. Judge completed signups per exposed visitor, with mobile specified in advance as the key segment. Check account connection + chat and subscriptions per exposed visitor with equal cohort age. Report raw counts and uncertainty.
5. Use hero clicks and all-auth-page arrivals as diagnostics. A rise in hero clicks without more completed signups is not a win. Correct the report’s overlap/count issues before interpreting finer path or channel breakdowns.
6. Recalculate sample size using the chosen eligibility, baseline, follow-up, and meaningful effect size. Precommit a stopping rule. Do not pool original V4 results with this revision or claim a before/after difference isolates the wording’s causal effect. That requires concurrent old/new V4 allocation.

​

QA

Use /?landing_variant=test and /?landing_variant=control on nonproduction. Check 375 × 667, 402 × 700, and desktop. Confirm the revised hero label, retained 56 px hit target, signup destination, unchanged secondary links, and visible keyboard focus. Check signed-in dashboard behavior through component validation. Verify revision stamping and deduplication, including a preexisting old V4 session key. No signup/account creation is needed for browser validation. Validated locally: At 402 × 700 the unchanged V3 CTA is at y=557–613. Keyboard focus is visible on the revised V4 button and Enter opens the actual /sign-up account-creation page. The four component regression tests cover revision exposure/dedupe, signup attribution/navigation, analytics-disabled navigation, and signed-in dashboard behavior. Changed-file lint and the full TypeScript check pass. The extracted hero JSX matches staging after normalizing the one changed label and equivalent palette references. V3, navigation, and flag-resolution code have no diff. The companion navigation-skill PR documents the hero entry point and explains that clicking it opens signup rather than connecting a bank. Sync that skill to S3 with the corresponding UI release, per docs/SILVIA-SKILLS-SYNC.md.