How it works
An A/B test compares variants of a defined experience against a measurable outcome. Decide what changes, who sees each version and what success means before running it. Separating visitors consistently helps prevent a person from switching between versions during one task.
A realistic example
A contact page could compare two clear button labels while keeping the offer and form unchanged. Measure completed valid requests rather than clicks alone; a label that attracts clicks but confuses visitors may not improve the journey.
What to check
Check assignment, event collection, device mix and whether errors affect one variant. Keep a record of duration and sample size. Inspect the whole flow in both versions, including keyboard use and translated text.
Limits and next steps
Small samples and early stopping can produce misleading apparent winners. Test a specific hypothesis rather than several unrelated changes together. Preserve privacy requirements and avoid interpreting a correlation as proof of why people behaved differently.