Stop A/B testing your buttons

You have run 40 A/B tests on the same checkout button. Round corners. Square corners. Red. Green. “Buy now.” “Place order.” The lift is always 1.5% or it is dead. You are running out of variants to test.

The button is not the problem

If you cannot break a conversion ceiling with copy and color experiments, the ceiling is structural. The page is asking the wrong question, the flow is asking it in the wrong order, or the user is missing context the page assumes they already have.

Buttons are easy to A/B-test because they are visible, small, and isolated. Structure is hard to A/B-test because the unit of change is a flow, not an element. Most teams test buttons because buttons are testable. That is selection bias, not strategy.

What to test instead

• Order of steps. Does the email field come before the address? Why? Try swapping.
• What is on the page at all. Cut a section. Cut three. See what happens.
• The mental model. Is this a form or a configuration? A purchase or a subscription? The visual treatment should match the model.
• What comes before the page. Most checkout problems are upstream — in the cart, the PDP, the landing.

Earn the right to test structurally

Structural tests are scary because they cannot run on a 50/50 split for 6 hours. They require a 5% canary for two weeks, careful instrumentation, and a rollback plan. Build that muscle. The teams that ship the biggest wins are the ones that earned the trust to run structural tests at all.