Landing page testing attracts more theory than almost any other job in marketing. Frameworks, heatmaps, color psychology, diagrams with more arrows than a flight map. Meanwhile, the teams actually running these tests have small budgets, tight timelines, and, more often than anyone admits, a page someone else already broke. I've been running these tests for about ten years, mostly on B2B and fintech pages, mostly in VWO with GA4 keeping score. The fintech runs below are from PingPong, where I was Marketing Manager, Ads and Lifecycle, 2021–2023. This is the sequence I run, in the order I run it, including one test I ruined myself.
The work starts before the test does
Before I change anything on a page, I answer three questions: what is this page for, what did the ad or email that sent people here promise, and does the visitor's intent match what the page delivers? Most underperforming pages fail right here. People arrive expecting one thing and the page hands them another.
I used to tell teams that fixing this mismatch solves half the conversion problem. Honest version: I can't prove the half. It's a rule of thumb I have never measured. What I can point to is the Google registration program I ran, where cost per signup fell from $90+ to about $35 across roughly three months; I matched the headline to the ad, one of the first levers I pulled there, and it shipped before any layout change did. Audience and bidding tightened in the same window, so the headline carries part of that fall.
Simplify before you optimize
Most pages over-explain and bury the essentials. The pages that work have one call-to-action, one core message, and one value proposition visible before any scrolling. My check is the 3-second stare test: show the page to someone outside the project for three seconds, then ask what's on offer and what they're supposed to do. If they can't answer, nothing below the fold will rescue you. So the first “test” is usually subtraction.
Test the angle before the headline
The angle decides whether the reader sees themselves on the page: budget pressure, compliance. I test the angle before the phrasing. Ten headline variants on the same angle mostly produce noise. Two genuinely different angles produce behavior you can see without squinting at a significance calculator.
Layout tests only matter once the message is settled. A page with the wrong message underperforms in any layout; a strong message survives a mediocre one. Many teams run this backwards and spend a quarter polishing UI around copy nobody believed.
One variable at a time (I learned this the expensive way)
Early in my time at PingPong I was testing headline copy, and mid-flight a well-meaning teammate “refreshed” the hero image on one variant. Nobody logged it. Two weeks of paid traffic later, variant B was winning comfortably, and I was a day away from crediting a headline for work the photo was doing. I caught it in the QA screenshots, threw that window out and ran the test again. The rerun was my call. Same media budget, a low five figures, spent twice, for one usable answer.
Since then the rule is boring and absolute. Testing copy? Freeze the visuals. Testing visuals? Freeze the copy. Change something mid-test, log it. Better, let it wait its turn.
The fintech test that changed our segmentation
On one fintech page I ran three angles head-to-head in VWO: speed, cost savings, and compliance reassurance. My money was on cost savings; on paper this was a price-sensitive segment. Compliance won, and it wasn't close. I won't print the exact rates here; they live in a former employer's dashboards and aren't mine to publish. The sample was small, so I read the result as directional. But the gap was wide enough that I rebuilt the page and rewrote that segment's messaging across email and paid social, both channels I owned at the time. That audience was managing fear.
Once you lock the angle, layout gets mechanical: how long the page runs, where the proof, form and CTA sit. Get the angle wrong and none of it matters.
Where AI actually fits
AI is worth having in the grunt phases. Sifting a batch of Hotjar session recordings used to cost me about an hour per page; now I export the funnel data and recording summaries and ask Claude one question: “Where do people stall, and what is on screen when they do?” On one page it flagged that a compliance disclaimer was rendering above the form on mobile and pushing the CTA out of the first screen, something three of us had scrolled past for weeks.
What I don't let it do is pick the angle. The angle lives in the audience, and every time I've asked a model to guess it cold, it picks the safe, wrong one. Speed and savings, every time. The compliance-fear insight came out of a test.
The whole workflow, stripped down
- Find where the ad and the page disagree on what was promised
- Reduce friction. One CTA and one message the reader sees in three seconds
- Discover the winning angle, test one emotion at a time
- Rebuild the hero around it. Put the angle in the headline
- Reorganize the page so every section serves the message
- Refine the UI last, one frozen variable at a time, with a change log
Does the page say something the user cares about, clearly? Are we honoring the promise that attracted them?
Open your highest-spend ad and its landing page side by side, then fix whichever one is lying.