Choose the page and the metric
Pick the page with the most traffic that has a clear goal, such as a demo request or a download. Define one primary metric before you start: the share of visitors who complete that goal. Add one supporting metric, such as click-through on the main button, to see where the change acted.
Do not use bounce rate alone. A visitor can stay and still not act. Use your analytics tool, such as Google Analytics or PostHog, to record the baseline for the last four weeks first.
Check whether you have enough traffic
Be honest about volume. If your page gets 1,500 visits and converts at 3%, that is 45 conversions a month. With numbers that low, a small change in wording will not give you a clear answer in a reasonable time.
As a rule of thumb, aim for at least 100 conversions per variant before you trust the result. If you cannot reach that in four weeks, do not test small tweaks. Test a bold change, or use qualitative methods such as five user sessions and recordings in Hotjar. Our chapter on testing systematically explains the wider routine.
Write a hypothesis
Use this format: "Because [observation], we believe [change] will [result], which we will see in [metric]." For example: "Because recordings show visitors scrolling past the headline, we believe stating the outcome in numbers will raise demo requests."
A hypothesis forces you to say why. Without it, you are only trying random ideas.
Pick what to change
Change one element in each test. In order of typical impact:
- Headline: lead with the outcome the visitor wants, or with the problem you solve. Compare these two against each other.
- Subhead: say who it is for and how it works in one sentence.
- Hero image: a product screenshot against a picture of people. Test whether showing the real product helps.
- Button text: "Book a demo" against "See how it works".
- Proof line: a customer logo or number directly under the headline.
Start with the headline. If you want to know what a good above-the-fold should contain, read the hero has five seconds to answer three questions.
Set up the test
Use a testing tool such as VWO or the testing features in Unbounce or Webflow. Split traffic evenly and randomly between the original and one variant. Check that both versions load at the same speed and look right on mobile before you start.
Do not send paid traffic to one version only. A split that is not random gives you an answer about your traffic, not your page.
Run it for full weeks
Run the test for at least two full weeks, so that every weekday appears twice. Do not stop on the first day you see a winner. Early results swing a lot and often reverse.
Decide the end date before you start and write it down. Do not change the page, the ad copy or the offer while the test runs.
Read the result by device and source
When the test ends, split the result by mobile and desktop, and by traffic source. A headline can win on desktop and lose on mobile, where less of it is visible. If the result is mixed, ship the version that wins on your biggest segment and note the exception.
Log the learning
Write down the hypothesis, the variants, the dates, the numbers and what you will do next. A shared document in Notion is enough. When you test a similar page later, start from the winning pattern and test one new thing, instead of repeating the same test.
Common mistakes
- Testing a new headline and a new image together, so you do not know which one worked.
- Ending the test when it looks good after three days.
- Using too little traffic for a small change.
- Measuring clicks on the button but not what happens after.
- Forgetting mobile, where most visitors may be.
How you know it works
Each test ends with a decision and a written note. At least half of your tests teach you something, even if the variant loses. Over a quarter, the conversion rate of your top page rises, and you can name the two or three changes that caused it.
