Real experiments. Real outcomes. Actionable patterns. Browse A/B tests with problem-to-solution framing, results, and recommendations for what to test next.
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. Worth noting: external research flags sticky CTAs as context-dependent — they help when the primary action is buried below the fold, but can hurt on shorter pages where the original CTA is already visible.
Ambiguity > absence. A vague benefit callout can create more friction than no callout at all: visitor diagnostics showed users were drawn in by the badge (time-on-page up, bounce rate down) but exit rate rose and FAQ-section attractiveness spiked — a signature of users searching for answers and not finding them. The same concept won at a sister brand whose variant used descriptive benefit-framed copy ("we'll help you find the right plan if this isn't a fit"); the variant in this test used short labelled-badge copy that raised more questions than it answered. The lesson is not that benefit guarantees fail — it's that surfacing one with insufficient context can backfire by introducing uncertainty the page doesn't resolve.
Problem: Coupon and promo code fields on checkouts can distract users — they leave to hunt for codes, reducing completion rates.
Test the variable users actually complain about — not the variable that's easiest to redesign. This test is a textbook case of treating form when the problem is content. Cross-brand qualitative research had consistently flagged three specific confusion themes: (1) pricing structure is opaque — users can't predict what they'll pay; (2) plan names are brand-driven rather than benefit-driven, so the names themselves don't communicate what the user is buying; (3) no side-by-side comparison — vertical layouts force users to scroll and remember instead of compare in parallel. Visual hierarchy is a presentation improvement; it does nothing about pricing opacity, naming clarity, or comparison difficulty. The test reached its planned sample size and produced a directionally-negative result at the noise floor — because organizing unclear content doesn't make the content clearer. The transferable insight isn't about visual hierarchy specifically; it's about the importance of mapping qualitative complaints to the test variable. If the user research says 'I don't understand what this plan costs,' the test should manipulate cost-clarity. If it says 'I can't tell these plans apart,' the test should manipulate differentiation. Layout tests are appropriate when the complaint is about layout — not when they're a default reflex.
Problem: Visual elements on the listing aren't doing enough to communicate value, build trust, or guide users toward the next step.
Context: Multi-step processes on the product can overwhelm users if they can't see how far along they are or how much is left.
Context: Multi-step processes on the home landing can overwhelm users if they can't see how far along they are or how much is left.
Context: Users on the checkout need validation from others before committing — without visible proof of success, they hesitate.
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.
Context: How prices are displayed on the product directly influences perceived value and willingness to buy.
Context: Each additional form field adds friction to the checkout, increasing the chance users abandon before completing their submission.
Context: Friction during the checkout process causes users to abandon right when they're closest to converting.
Problem: How prices are displayed on the pricing page directly influences perceived value and willingness to buy.
Problem: Multi-step processes on the listing can overwhelm users if they can't see how far along they are or how much is left.
Context: The primary call-to-action on the listing isn't converting at its potential — design, copy, or placement may be the bottleneck.
A CTA's click rate is not its conversion contribution. This test surfaced one of the most consistently underweighted patterns in CRO: behavioral diagnostics almost always tell a more honest story than the topline. The aggregate result looked like a tiny non-significant lift (+1%); the diagnostic revealed that of every 100 button clicks, only 6 reached the next funnel step. Two failure modes converged: (1) copy intent mismatch — the chosen label read as 'create account' rather than 'shop,' so a large share of clicks came from users trying to log in / manage their account from support and customer pages; (2) extra modal step before the destination page added friction without value. The aggregate lift was partially cannibalization from higher-converting paths. The transferable pattern: when introducing a global navigation element, validate the click→conversion ratio per source page, not just the topline. High clicks from low-intent pages creates a false signal of engagement that can mask poor performance.
Problem: The registration experience on the product asks too much too soon, causing potential users to drop off.
Context: Key actions on the content page disappear as users scroll, creating a gap between intent and the ability to act.
Problem: How prices are displayed on the pricing page directly influences perceived value and willingness to buy.
Context: The headline on the product may not resonate with what users actually care about or address their top objections.
Build on what's already been learned. Save your own experiments, surface winning patterns, and make every test count.