X'inhu A/B test?

An A/B test (split test) shows visitors two versions of a page or element at random, the control and a variant, and measures which produces more of a target outcome, usually a conversion. Random assignment is what makes it an experiment rather than a before-and-after comparison: both versions face the same traffic mix, seasonality, and news cycle at the same time, so the difference is attributable to the change.

Most A/B testing failures are procedural. Decide the success metric and required sample size before starting, then run until you reach it; checking daily and stopping the moment the result looks good ("peeking") is the classic way to ship noise. Test one meaningful change at a time, expect most tests to lose or tie (industry win rates run well under a third), and confirm results with statistical significance rather than eyeballing a gap. Low-traffic sites often cannot feed a proper test at all; below a few thousand conversions-relevant visits per variant, bigger swings and qualitative feedback beat micro-tests.

Good test candidates come from your data: a funnel step with heavy drop-off, or a high-traffic landing page with a weak conversion rate.