For AI agents: the complete documentation index is available at https://docs.clickmax.io/en/llms.txt, the full documentation bundle is available at https://docs.clickmax.io/en/llms-full.txt, and this page is available as Markdown at https://docs.clickmax.io/en/guides/ab-testing-in-practice.md.
  • English

      • How do I test two versions and pick the winner?

      An A/B test replaces opinion with data. But a badly built test produces data that looks like an answer and is not — which is worse than not testing at all.

      Before you start

      • Role: Owner, Admin or Editor.
      • Tracking has to be working.
      • You need volume. With little traffic, the result is noise.

      1. Test one variable at a time

      If you change the headline, the image and the button together and version B wins, you do not know what worked — and you cannot repeat the win.

      Start with what has the most impact: the headline and the value proposition move the result far more than a button colour.

      2. Define the metric up front

      Decide what "winning" means before you start. It is usually the sale, not the click: a variation can produce more clicks and fewer purchases.

      3. Build the test

      Follow create a funnel A/B test.

      4. Let it run

      Two mistakes destroy the validity of a test:

      • Stopping early. In the first few days one version always looks like it is winning. That is chance.
      • Ignoring the weekly cycle. Monday behaviour differs from Saturday behaviour. Run whole weeks.

      5. Decide on a criterion

      Small differences on low volume are not a conclusion. If the two versions are level, the most likely explanation is that the variable you tested does not matter — move on to another hypothesis instead of forcing a winner.

      6. Apply it and carry on

      Publish the winner and start the next test. Conversion gains come from many small tests, not from one big discovery.

      7. Write down what you learnt

      Record what was tested, on how much volume, and what the result was. Without a record, months later the team tests the same thing again.

      If it didn't work

      • The versions came out level: the variable you tested is probably not the decisive one.
      • The numbers do not match your sales: confirm that the metric is the sale, not the click.
      • Traffic did not split as expected: check the test setup in A/B test.