The problem with testing at low volume
With twelve conversions in a test, a 30 percent difference between variants is well within the range of pure chance. Call it a winner and you have made a permanent decision from a coin flip.
This happens constantly, and it is worse than not testing, because it manufactures false confidence and then compounds on it.
Rule one: one variable
Change the headline or the image or the offer or the landing page. Not two. If you change two and results move, you have learned nothing about either.
This is boring and it is the entire difference between testing and guessing.
Rule two: thirty conversions per variant, minimum
Not clicks — conversions. Below roughly thirty per side, the result is not readable at small-business effect sizes. If your volume means that takes six weeks, the test takes six weeks.
If you cannot reach thirty in a reasonable window, do not test that thing. Test something with more volume, or make the change on judgment and move on.
Rule three: test big differences
At low volume you can only detect large effects. So test $99 against free, not $99 against $95. Test a completely different hook, not a reworded one.
Small refinements are for accounts with volume. At $2,000 a month, test only things that could plausibly change results by half.
What to do with a result
Implement the winner everywhere it applies, write down what you learned with the date, and start the next test. An untested variant kept running "for balance" is just a worse ad you are still paying for.