---
name: A/B Testing
description: Measure what actually works.
---
# A/B Testing

Most of what you "know" works, doesn't — until you test it. A/B testing replaces opinion with measurement. But bad testing produces confident wrong answers.

## Discipline
- **One variable at a time.** If you change headline, image, and CTA together and the variant wins, you've learned nothing about which mattered.
- **Pre-register the hypothesis and metric.** Decide before launch what counts as success. Picking the metric after seeing results is p-hacking.
- **Wait for statistical significance.** Stop early and you read noise as signal. Calculate the sample size needed before you start; running longer than that wastes time, shorter invalidates.
- **Bigger effects need smaller samples.** Tiny optimizations need huge traffic to trust. If you're small, test big changes, not button shades.

## Avoid
- The "winner" that doesn't replicate. Usually a methodology bug, not a real effect.