Test one variable, or you learn nothing

Test one variable, or you learn nothing

Teams love to "test ads", then ship four variants that differ in headline, image, body, and call to action all at once. When one wins, they have no idea why, so they cannot repeat it. That is not testing, that is gambling with a spreadsheet. A test only teaches you something if you can attribute the result to a single deliberate change.

Decide what you are actually trying to learn before you build the variants. Is it the angle (the core argument)? The hook (the opening line)? The format (static versus video, long versus short)? The offer (what you ask the click to do)? Pick one layer, hold everything else steady, and let the data answer one question at a time. Angle is almost always the highest-leverage thing to test first, because a better argument beats a better headline on a worse argument every time.

Respect the maths. A test needs enough conversions, not enough clicks, to call a winner, and in B2B with long sales cycles and small audiences that can take a while. Calling a winner after twenty clicks is reading tea leaves. If you cannot reach a meaningful number of conversions in a reasonable window, test higher-funnel signals you can actually measure, like click-through rate or cost-per-click on the hook, and accept that those are proxies, not the truth.

The output of a good test is not "version B won". It is a written sentence you can reuse: "naming the specific job title in the opening beat the generic pain statement". That sentence compounds. Run enough clean tests and you stop guessing, because you are building a library of what is true for this audience.

INTERVIEW EWOUD: What is your default first test on a new account, and why that one first?

INTERVIEW EWOUD: How many conversions do you want before you trust a result? Give me your real threshold, not the textbook one.