To Know If Your Landing Page Works, You Need These Four Numbers

A/B testing isn't guesswork. Here's the math that tells you when you've seen enough visitors to trust a winner.

The 5-second version

  • Landing page tests require four inputs: baseline conversion rate, minimum lift you care about, sample size per variation, and traffic split rules.
  • A page converting at 3% needs roughly 13,900 visitors per version to detect a meaningful 20% improvement with standard statistical confidence.
  • Running tests without these numbers wastes traffic on inconclusive results and delays the decision to roll out or kill a variant.

Most business owners run A/B tests on landing pages by launching a variant, watching the click-through rate climb, and picking a winner within days or weeks. The result feels obvious, so the test feels done. Then, three months later, the "winner" stops performing, or the improvement never materializes at scale. What went wrong was not the test; it was the sample size.

Prestige Technologies' 2026 guide lays out the four numbers every landing-page test requires: baseline conversion rate, minimum detectable effect, planned sample size per variation, and consistent traffic allocation. Without these, you are not running a test; you are watching traffic and guessing when to call it.

The Four Inputs That Define Your Test

Your baseline conversion rate is what your current page converts at today. If you send 10,000 visitors a month and 300 convert, your baseline is 3%. This is your starting line.

Your minimum detectable effect is the smallest lift that moves the needle for your business. A 1% lift sounds great but might be noise. A 20% relative lift is meaningful: it moves your 3% page to 3.6%. That is the threshold you design the test to catch reliably.

Your sample size is how many visitors each variation needs to see before you can trust the result. The guide shows the math: a page at 3% baseline, aiming to detect a 20% relative lift, requires about 13,900 visitors per variation. That is per version, not total. A two-variation test (control and challenger) needs nearly 28,000 combined visitors before the result is statistically sound.

Traffic allocation is your rule for who sees what. Typically you split 50/50 between control and challenger. The rule keeps you honest: you send new visitors to both versions equally until the sample size is reached, then you stop, check the math, and act. No peeking; no stopping early because one side looks better.

13,900 visitors per variation needed to detect a 20% lift on a 3% baseline landing page (Prestige Technologies, 2026)

Why Sample Size Stops You From Fooling Yourself

The first week of a test often shows wild swings. One version may jump 15% ahead. That is not a signal yet; it is a small sample doing what small samples do. By week three, the gap narrows. By the time you reach your target sample size, the true difference emerges, and it is often smaller (or zero) than the early days suggested.

Running without a pre-set sample size kills money in two ways. First, you launch a false winner and watch it underperform once you scale. Second, you shut down a real winner too early because you gave up waiting. The math prevents both mistakes.

The Math in Practice

Prestige Technologies' example assumes standard statistical settings: 95% confidence and 80% power. Those are industry defaults. At those settings, a 3% baseline page needs 13,900 visitors per version to reliably catch a 20% relative lift (moving from 3% to 3.6% conversion). If your page converts at 10%, the number shrinks. If you care about a 10% lift instead of 20%, the number grows because a smaller win is harder to prove.

The point: the numbers are not arbitrary. They come from your own baseline and your own goal. Know them before you launch the test, and you stop wasting traffic on hunches.

Who Actually Hits Sample Size

Not every landing page gets 28,000 visitors in a month. If yours gets 5,000, your test will take six weeks. If it gets 1,000, it takes half a year. That is not a problem; it is the schedule. The alternative is guessing faster, which costs more money than waiting.

For pages with very low traffic, reduce your ambition: instead of detecting a 20% lift, test for a 50% lift. The math shrinks the sample size. You are less likely to catch tiny wins, but you speed up your ability to catch big ones. The tradeoff is explicit and honest.

The discipline of knowing your four numbers and respecting them turns a landing-page test from a guessing game into a revenue lever. You stop second-guessing winners. You stop killing tests too early. You scale what actually works.

Questions owners ask

How do I know when to stop running an A/B test?

Stop when you have reached the calculated sample size per variation that your baseline conversion rate and desired lift require. According to Prestige Technologies' 2026 guide, a 3% baseline page needs about 13,900 visitors per variation to detect a 20% relative improvement with standard statistical settings; different rates and lifts shift that number.

What if I stop a test early because one version looks better?

You risk a false positive: the apparent winner may lose its edge as more data comes in. The sample-size math is built to avoid that trap, so stopping before you hit the target number undermines the whole test.

Does every change to my landing page need an A/B test?

Only if the result will change your business decision. Tests require visitor volume and time; if you are redesigning the entire page or trying a new offer, the stakes are high enough to test. Minor tweaks may not move the needle enough to justify the sample size.

What is a 'minimum detectable effect' and why does it matter?

It is the smallest improvement you actually care about hitting. If you decide a 20% lift in conversions is your win threshold, the test is sized to catch that lift reliably; smaller improvements get buried in noise.

Sources