CRO & UX Research

A research sprint produces more observations than a team can act on, so each is written as a testable hypothesis: because we observed X, changing Y for audience Z should move metric M. A hypothesis without an observation behind it goes to the bottom of the backlog.

We score the backlog with ICE (impact, confidence, ease) or PIE (potential, importance, ease). Impact comes from the page's traffic and conversion value, confidence from how many methods pointed at the same problem, ease from your build capacity. The scores are rough by design; they make the order of work defensible and keep the loudest opinion from choosing the next test. Lost tests re-score the backlog too; they often disprove a whole family of hypotheses.

An A/B test answers one question: is the difference between variants unlikely to be chance? Before a test starts we fix the primary metric, the conventional 95 percent confidence level, the statistical power we accept (usually 80 percent) and the minimum detectable effect, the smallest improvement worth building. Together these give the sample size: conversions, not visits, per variant. Smaller effects need far more conversions than teams assume.

Duration follows from the sample size and the business cycle: whole weeks, usually two to four, so weekdays and weekends are both represented. We do not stop early because a line looks promising; stopping on a good day is the most common way to ship a winner that does not exist. Overlong tests blur instead, as cookie churn, campaigns and seasonality creep in.

When the calculation says a valid test would take many months, we do not run it. Low-traffic pages get research-led changes: problems that replay and user testing show unambiguously are fixed, shipped and measured before and after in GA4, same segments, seasonality in mind. With some traffic we test bigger changes, pool pages sharing a template, or start with the busiest funnel step.

Landing pages are where CRO and paid media meet. The first screen must repeat the promise of the ad that brought the visitor (message match), say who the offer is for, and make the next step obvious on a phone. We work from the Google Ads and social advertising campaign structure, so variants align with ad groups and audiences.

Forms are the most common leak and the easiest to measure. Form analytics shows which field people abandon; replay shows why: a date picker that fails on iOS, a phone field that rejects local formats, a misplaced error message. Typical fixes: fewer fields, inline validation, the right mobile keyboard, a visible privacy note and, for clinics in Türkiye and MENA, a WhatsApp or call option beside the form.

A slow or hard-to-operate page converts worse before any persuasion is involved. Google documents Core Web Vitals (LCP, INP and CLS) as page-experience signals (Google Search Central), and visitors feel the same things: a late hero, a slow button, a layout that shifts under the thumb. Accessibility works the same way: sufficient contrast, visible focus states, labelled fields and thumb-sized tap targets help every visitor and remove errors replay otherwise shows as abandoned forms. Speed fixes go to the technical SEO work, so both disciplines gain.

Every programme starts with a measurement audit; a test judged on unreliable data is worse than no test. Research precedes hypotheses, hypotheses precede tests, and the sample-size calculation sits in every experiment plan so nobody is surprised by a test that cannot conclude. How many tests run at once depends on your traffic and build capacity, not a template.

CRO and SEO do not conflict when tests run the way Google describes (Google Search Central): variants on the same URL or a rel=canonical from the variant to the original, 302 redirects for redirect tests, nothing shown to crawlers that users do not see, and tests ended once they conclude. We keep the H1, core content and internal links stable unless they are the subject of the test. A page that answers the question clearly, loads fast and is easy to act on is what search engines reward and what converts.