Landing Page Testing Framework: Systematic Optimization

Landing page testing has better conditions than email testing — traffic can be split evenly, tests can run until they resolve, and there is no fixed audience size. It has one problem instead: most landing pages get far less traffic than the people testing them believe.

The consequence is a lot of tests that cannot resolve, followed by a decision made on noise, followed by a change that does nothing. Doing the arithmetic first avoids all of it.

The order that stops you wasting tests

Each stage changes the numbers for everything below it, so testing out of order produces results that expire.

Offer first: the thing being exchanged. This has the largest effect by a wide margin, and every subsequent test is measured against whatever it settles on.

Layout second: one column or two, form position, page length, how much page the ask gets. Large effects, and cheap to test because the variants are structurally different.

Copy third: headline formula, then the supporting line, then the button. Moderate effects, needing more traffic than the first two.

Detail last: colours, images, spacing. Small effects, and on most landing pages below the resolution available.

What to do when the traffic is not there

Most landing pages, most of the time. Three honest options.

Test sequentially: run version A for a fortnight, version B for a fortnight, compare. Not controlled — traffic mix varies between periods — but for a large effect it resolves, and running A-B-A-B removes most of the objection.

Test across pages: apply the same change to several landing pages at once and look at the aggregate. This works well for structural changes like removing navigation.

Or decide on judgement. For a change with strong prior evidence — removing site navigation, moving the form above the fold for a small ask — implementing it without a test is a reasonable use of limited traffic.

Measure the right conversion

The form submission is the obvious metric and it is not always the right one. A page that raises submissions while attracting people who never engage afterwards has made things worse in a way the test reports as a win.

For any offer test in particular, look at what those cohorts do over the following months, not just what they did on the page. That is the difference between optimising the page and optimising the business, and they diverge more often than is comfortable.

Write down what did not work

A test showing no difference is a real finding and it is the one nobody records, so the same test gets proposed again a year later and consumes another fortnight to reach the same answer.

Four lines per test: what was tested, over what period, the numbers, and what was decided. It takes a minute and it is the only thing that stops a team relitigating settled questions — and it protects the decisions that were made for reasons the numbers alone do not show.