<b>I ran a test that could NEVER have won. 40 days wasted.</b>

<b>I ran a test that could NEVER have won. 40 days wasted.</b>

Painful one. I tested a tiny CTA tweak on a page that gets ~300 conversions a month.

Forty days in: 'no significant difference.' Of course not.

Here's what I skipped: I never calculated the minimum detectable effect. With my traffic, that test could only ever detect a 20%+ swing. A small CTA tweak realistically moves 2-3%. It was statistically impossible to ever read.

Low-traffic pages can only detect BIG changes. On them, stop testing micro-tweaks — test radically different versions, or don't test at all and just use judgment.

— Check your monthly conversions on the test page
— Under ~1,000? Only test big, bold variants

Go run an MDE check before your next test. Kill it if the math says no. Report back.