This may happen if your sample size is too small.
The statistical test to determine whether your results are significantly different has meaning only for predefined sample sizes that are large enough. If your sample (number of visitors tested) is too small, this test loses its value.
This is why we recommend that you calculate the sample size you need to attain before trusting this statistical test (ex: http://www.evanmiller.org/ab-testing/sample-size.html). We also recommend you wait a few extra days even if your test seems large enough to take into account behavioral differences between the days of the week.