You test until you have enough data to be actionable. The more data you have the more confidence you will have in your conclusions. So where you feel comfortable making changes will vary based on personality, the size of your campaigns, and you need to be accurate. Once you make a change, based on the results of your test, you should continue testing for both confirmations that you made the right change, as well as for further enhancements.
Most people do a way too little testing and make many of their decisions based solely on anecdotal evidence. This, of course, leads to many bad decisions. I suggest that you study the principles of a valid test. Learn the difference between anecdotal evidence and empirical data.
As a rule of thumb, the same test repeated 103 times will give you roughly 80% probability with a 20% margin of error, and 312 tests will get you to about 90% probability with a 10% margin of error. It’s up to you to decide how much accuracy you need to make a change. Generally, you need at least 30 repeats to have a better than 50% probability of being correct.