Following guidelines are recommended for proper test execution:

1. 95% statistically valid

The test outcome needs to be 95% statistically valid. That still means 1 out of 20 times there is a false positive; i.e. the test outcome indicates an improvement - but in reality there is none.

2. Run for at least two business cycles (2 to 4 weeks)

Often there are differences between weekends and midweeks and day and night. To rule this out it is generally recommended to test full weeks.

3. At least 100 conversions per variation

With low numbers your outcome will be unreliable - a few conversions more and your outcome would be different if you have just a few conversions.

4. Hypothesis based on analysis

If you don't have a sound hypothesis you are just throwing mud on the wall. A proper analysis will highly increase the chances of a successful test.