A/B tests

Two pages, one address

Split the traffic between two landees and see which one gets its goal clicked more often.

  • Visitors are split at random, half to each
  • You choose what counts as a win — a click on a specific link
  • The result is broken down by device, country and source

A tie so far. Each variant needs 300 more visits before the difference means anything.

A/B tests
One landee against another on a shared address. The winner is whichever gets its goal clicked more often.
Create a test
Funnels + A/B tests

What you can put against what

A funnel shows where people drop off. A test shows which path is better. Here is every combination the system allows, and when to reach for each.

A
vs
B

Page against page

Which look and wording works better.

When one screen changes: the headline, the button label, the order of blocks.

A
vs
B

Page against funnel

Whether the funnel is worth the extra steps.

The goal is the same button: on the page in A, on the last step in B.

A
vs
B

Funnel against funnel

Which shape of the path works better.

Three steps against five, a different order of screens, a different first step.

branch A
branch B

A split inside one funnel

Which screen in the middle of the path is better.

Cheaper than a test: no second site, and exactly one step changes.

from a phone
everyone else

A conditional rule is not a test

Who gets to see what.

One thing from a phone, another from Instagram. The audiences differ by construction, so the branches are not compared with each other.

So the numbers can be trusted

  • Until the smallest branch has a hundred visits, no winner is named — the screen says how many are still missing.
  • A variant is not changed mid-test: half the data would be about one thing and half about another.
  • One difference at a time. A test plus a split inside it, and you can no longer tell which one worked.
  • When the funnels differ, you are measuring the page and the path together, not the page alone.