What a Good Creative Test Actually Looks Like

Most creative tests we inherit are not tests. They are four finished ads, made at the same time by the same people, put into one ad set to see which wins. One of them wins. Nobody learns anything, because the four differed in a dozen ways at once.

Change one thing

A test needs a variable. Same body, four openings. Same opening, four offers. Same script, one read by a creator and one by a voiceover. When only one thing differs, the winner tells you something you can apply to the next fifteen ads.

Give it enough budget to be believed

We do not read a result under roughly fifty conversions per variant, and we do not stop a test before it has run through a full weekend. Consumer buying behaviour on a Tuesday is not the same as on a Sunday evening, and a test killed on Thursday afternoon usually just tells you what Thursday thinks.

Write the prediction down first

This is the part everyone skips. Before the test runs, write one sentence saying which variant you expect to win and why. When you are right, you have found a rule. When you are wrong, you have found something better, because a wrong prediction is the only thing that changes how a team makes ads.

Keep a record that outlives the account

We keep a single running document per client: the variable, the prediction, the result, the date. After six months it is worth more than the ad library. New people read it on day one and stop repeating tests we ran in March.

Three tests a month, run properly, beat twenty run badly. The limit is rarely budget. It is patience.