Once an experiment is running, you have three controls: Pause, Resume, and Complete. Each does something specific to your data and to the visitors currently bucketed into variants.
Pause
Pausing freezes the experiment. New visitors stop being assigned to variants — they all see the control. Visitors already bucketed keep their assignment in case they come back, so when you resume, they continue to see the variant they were in.
Use pause when you spot something wrong: a variant breaks the page on mobile, conversion drops noticeably, you need to fix something and re-test. Pausing doesn't lose your data — when you resume, the experiment picks up where it left off.
Resume
Reactivates a paused experiment. Same variants, same goal, same traffic split. Existing visitors stay in their bucket; new visitors are split as before. Use this when the issue that caused the pause is fixed.
Complete
Marks the experiment as finished. New visitors stop seeing variants entirely — everyone sees the control. The data is locked in for your records, but you can no longer collect more conversions. This is one-way.
Complete an experiment when:
- You've reached statistical significance and want to ship the winner.
- You ran for a full business cycle and the result is conclusive — even if it's "no winner".
- The test is no longer relevant (you redesigned the page, you're moving on).
Shipping the winner
Completing the experiment doesn't automatically ship the winning variant — that's a separate step. After completing, take the changes from the winning variant (text edits, CSS, layout) and apply them as the permanent design. Then archive the experiment for history.
Don't stop early
The biggest mistake is hitting Complete the moment a variant looks ahead. Read understanding significance first — small samples lie loudly, and stopping the test early will burn you.