We recommend using a daily swap interval (to have the versions swap every day at midnight) and let it run for at least 2 weeks so each version sees a different day of the week. But you can set intervals down as low as 30 minutes if you’d like.
2 weeks is a good minimum time frame but the longer a test runs and the higher the statistical significance level, the more accurate your results will be.
Theme Scientist has a built-in statistical significance calculator to help you decide how confident 1 version or the other is performing better. The more traffic and the longer the test runs, the more accurate your results.