Each record reports a winning outcome. That label is not proof that the result is independently reviewable or transferable; use the evidence tier and missing-data disclosure on each record before drawing a conclusion.
Sticky mobile CTAs can compress time-on-page meaningfully (~15% faster) without sacrificing engagement signals — users converted at a directionally higher rate AND moved through the page faster, suggesting reduced hesitation rather than rushed clicks. The result was shipped via 90/10 holdout monitoring rather than traditional 50/50 A/B inference — the high baseline (~85%) and limited mobile traffic made full A/B underpowered, so the team chose a holdout-validated rollout as the deliberate methodology. Bayesian P(variant > control) was ~0.90, supporting the directional ship call.
The story behind this win is the iteration discipline. The first attempt at this homepage redesign changed two systems at once (messaging + routing) and produced an ambiguous result: the entry metric moved slightly positive while downstream metrics moved meaningfully negative. The team correctly identified that the routing change — which inadvertently replaced direct links to a personalized plan-search experience with modal-driven entry into a generic flow — was the downstream killer. The iteration restored the original routing and kept ONLY the homepage hierarchy changes. All funnel metrics moved directionally positive in lockstep (entry +2.38%, mid-funnel +7%, conversion +11.81%) — none stat-sig individually but consistent enough across the funnel to justify shipping. Element-level diagnostics confirmed the mechanism: the segment CTAs the team intended to promote saw a 26-30% lift in unique-visitor interaction, while the unchanged hero banner stayed flat (as expected). Two key behavioral observations: (1) page-length reduction surfaced a 4x lift on a previously buried bottom-of-page zip code input — proving the secondary lesson that 'less page' can mean 'more conversion real estate'; (2) desktop strongly outperformed mobile, with the suspected cause being mobile's lead-with-form pattern (zip code above hero) — putting the form before the message creates friction. The broader transferable insight: when a messy test confounds multiple variables, the right move is to isolate one variable in the next test, not to abandon the hypothesis.
Use reported outcomes to generate hypotheses, then plan the sample, guardrails, analysis method, and stopping rule for your own context.
View Plans & Pricing