Landing page optimisation has a reputation for marginal gains because most of it tests marginal things. Button colour, headline punctuation and image choice produce small effects that usually fail to reach significance.
The large effects come from a different category of change, and testing in the right order saves months.
Start with message match
The single most common cause of poor conversion is a mismatch between what brought the visitor and what the page says.
Someone who clicked an ad about reducing onboarding time should land on a page whose heading is about reducing onboarding time, not on a generic product page. The mismatch is often invisible internally because the people who built both know they are related.
Audit this before testing anything: list your top traffic sources, look at the exact wording of the ad, link or search query, and compare it with the first thing on the page. Fixing a genuine mismatch frequently produces a larger gain than a year of button tests.
Then the offer
The second-largest lever is what you are asking for and what you are giving.
A page asking for a 45-minute demo converts worse than one offering a 10-minute call, which converts worse than one offering a self-serve trial. None of these is universally right — they attract different volumes of different-quality leads — but the choice affects conversion more than any design change.
Consider lowering the commitment one step and measuring downstream rather than at the form. More leads at lower quality is only an improvement if they convert later.
Then the page structure
A conventional structure exists because it works, and deviating from it needs a reason.
- A heading stating the specific outcome, not the product category.
- A subheading explaining who it is for and how it works, in one sentence.
- A visual showing the actual thing — a screenshot, not a stock photograph of a meeting.
- The call to action, above the fold, repeated further down.
- Three to five benefits, each tied to a capability rather than an adjective.
- Evidence — specific customers, specific numbers, specific quotes with names and roles.
- Objection handling, as a short FAQ.
- A final call to action.
Pages that underperform usually omit the evidence or the objection handling, which are the two sections that address why someone would hesitate.
Then the form
Every field costs conversion, and the cost is not linear — the fourth field costs more than the second, and anything requiring the visitor to look something up costs disproportionately.
Ask only for what determines the next step. If a human will call, you need enough to make the call worthwhile. If the lead enters a nurture sequence, an email address is enough.
Two mechanical improvements are reliably worth making: show validation errors inline rather than after submission, and never clear the form on an error.
Only then, the details
Button copy, colour, image choice, layout variations. These matter, and they matter least, and they need the most traffic to resolve.
Button copy is the most worthwhile of them: specific, first-person phrasing describing what happens (“Start my free trial”) generally outperforms generic labels (“Submit”). The effect is real and modest.
How much traffic a test needs
This is where most optimisation programmes quietly fail. Detecting a 10% relative improvement on a 5% baseline conversion rate requires several thousand visitors per variant to reach a reasonable confidence level.
Most B2B landing pages do not receive that in a month. The practical consequences:
Below a few hundred conversions a month, A/B testing will not produce reliable answers. Use qualitative research instead — session recordings, user testing with five people, and simply asking visitors what stopped them.
Test big changes, not small ones. A completely different page has a chance of producing a detectable effect; a different button colour does not.
Do not stop a test early because it looks good. Early results fluctuate wildly, and calling a winner at day three is how organisations accumulate a portfolio of changes that did nothing.
Qualitative methods when traffic is thin
Three that produce more insight than an underpowered test.
Session recordings. Watch twenty sessions on the page. Where people hesitate, scroll back and leave is usually obvious and usually surprising.
Five-second tests. Show the page to someone for five seconds and ask what the product does and who it is for. If they cannot answer, no amount of optimisation below the fold matters.
Exit surveys. One question to visitors leaving without converting. “What stopped you from signing up today?” produces unflattering and actionable answers.
Common mistakes
Optimising the page while the traffic is wrong. A page converting at 1% from an irrelevant audience is a targeting problem.
Measuring form submissions rather than outcomes. A change that doubles submissions and halves qualified leads has made things worse.
Removing information to reduce length. Long pages often convert better for considered purchases, because the visitor has questions. Remove what is not useful, not what is long.
Running several tests at once on the same page without the traffic to separate them.
Social proof that actually persuades
Most landing pages include evidence and most of it is wasted, because it is generic. Four things separate persuasive proof from decoration.
Specificity. “Reduced onboarding time from nine days to two” works; “great product, highly recommend” does not. A number the reader can check against their own situation is worth a dozen adjectives.
Resemblance. A testimonial from a company like the visitor’s is worth several from companies unlike it. Where traffic is segmented, the proof should be too.
Attribution. A name, a role and a company. An anonymous quote reads as invented, because a meaningful share of anonymous quotes are.
Placement. Proof works best immediately after the claim it supports, not collected in a logo strip at the bottom where nobody reads it.
Logos of recognisable customers are the exception to the specificity rule — they work as a trust signal with no words at all, provided you have permission to use them.
What to do with the result of a test
A finished test has three possible outcomes and teams handle two of them badly.
A clear winner. Implement it, and write down what you think it tells you about the audience. The hypothesis is more valuable than the variant, because it informs the next ten decisions.
A clear loser. This is useful information and is routinely discarded. Knowing that shorter copy hurts conversion for your audience saves you from three future attempts at it.
No detectable difference. The most common outcome, and the one that gets misreported as a small win. If the test did not reach significance, it did not produce an answer. Keep the simpler variant and test something bigger next time.
The offer behind the page
Page optimisation has a ceiling set by what is being offered. A well-built page for a weak offer will convert worse than a plain page for a strong one, and teams frequently spend months on the former.
If conversion is poor despite a clean page and matched traffic, the next thing to examine is the offer itself — what the visitor receives and what it costs them to get it. Our notes on B2B lead magnets cover which formats attract people who will actually buy.
Frequently asked questions
What is a good landing page conversion rate?
It varies enormously by offer and traffic source. A low-commitment content download might convert at 20–30%; a demo request from cold paid traffic might convert at 1–3%. Compare against your own history rather than a benchmark.
Should the form be above the fold?
For high-intent traffic that already knows what you do, yes. For cold traffic, a form before any explanation converts badly — the visitor needs the reason first.
How long should a test run?
At least two full weeks to cover weekly cycles, and until the required sample size is reached. Whichever is longer.
Do video backgrounds help?
Rarely enough to justify the page weight. A short product demonstration video in the body frequently does help, because it shows the thing working. See Core Web Vitals for the performance cost.
