Landing page split test — the clean protocol

Landing page split test — the clean protocol

Most LP tests are unreadable because the setup confounds variables. Here's how to get a result you can trust.

The protocol:
— Hold the offer constant. Test LPs UNDER a single fixed offer, never across offers with different payouts.
— Even weighting at start: 50/50 (or 1/N). Do NOT let the tracker auto-optimize weights until you have enough data — early auto-optimization starves the eventual winner.
— Define the metric before you look: conversion rate to the tracked event, not click-through to offer. CTR to offer rewards aggressive pre-sells that don't convert downstream.
— Set a minimum sample: aim for ~100 conversions per arm, or compute it for your baseline CR and minimum detectable lift. Don't call winners at 8 vs 11.
— One variable per test. New headline OR new hero image, not both. Two changes = no causal read.

Guardrails:
— Run arms simultaneously, never sequentially. Sequential tests confound LP changes with day-of-week and traffic-quality drift.
— Re-randomize, don't reuse yesterday's split-cached users.

Log the test ID, variants, dates, sample sizes, and result. That log is your LP knowledge base.

Run this before every launch.