On this page5 sections
Published October 22, 2026.
Write the hypothesis before the build
A useful hypothesis names the audience, the change, the expected behavior, and the measure. "A shorter form will increase completed requests from paid search visitors" is testable. "Improve conversion" is not.
Record the baseline rate and the observation window before the variant goes live. Without a baseline, the result becomes a debate about whether the new page "feels better."
Keep the variant and the control comparable
Change one meaningful element at a time. If the page also gets a new traffic source, a new offer, and a new form, the result cannot be attributed. Record every difference between the two versions.
Use the same audience definition and the same qualification rules. A test that sends sales-qualified leads to one variant and general traffic to the other measures lead quality, not the page.
Log the decision, not just the number
Record the result, the confidence the team has in it, and the decision: adopt, reject, iterate, or stop. If the team overrides the data for a business reason, write the reason down.
A result that is not adopted should still be logged. It prevents the team from rerunning the same test next quarter.
Set an expiry date
Markets, offers, and traffic mix change. Record the date after which the result should be retested. A two-year-old conversion win is not evidence for today’s page.
Review the log quarterly and archive tests that no longer apply.
Related reading
For the supporting routine, see this guide and this guide. To discuss the workflow, contact us.
