Guides · Landing pages and conversion
A/B testing for a small site: when you have enough traffic and when you don't
A/B testing means showing two versions of a page to visitors at the same time and counting which one gets more enquiries or sales. On a small site it only works if you have enough visits to tell a real difference from luck. If you do not, test one big change at a time, or skip testing and fix the obvious problems first.
This guide is for an owner who wants to know whether testing is worth doing yet. It covers how much traffic you need, what to do when you have less, and the mistakes that make a test lie. It is one part of our wider conversion rate optimisation work, where every page change is tested against what was there before.
What is A/B testing, and what does it need?
You build a second version of a page with one thing changed. Half your visitors see the original (version A) and half see the new one (version B), chosen at random and at the same time. After enough visits, you compare the share of each group that enquired.
Three things have to be true for the answer to mean anything:
- The split is random and simultaneous. Running version A in June and version B in July tells you about the seasons as much as the pages.
- You measure one outcome that matters. An enquiry, a booking or a sale. Clicks and scroll depth are not the business.
- You wait for enough visits. This is where small sites fail, and it is the next section.
Most testing tools split traffic and count enquiries for you. The tool is the easy part. The hard part is having a clear idea to test and enough visitors to test it on. A tool with no traffic behind it will still produce a winner, and that winner will mean nothing.
It also helps to write down your guess before you start. Something like: "Removing the menu will lift enquiries because visitors will have one thing to do." A test with a reason behind it teaches you something even when it loses. A test run on a hunch teaches you nothing either way.
How much traffic do you need to A/B test?
There is no single number, because it depends on how big the difference between your two versions is. A big change shows up on fewer visits. A small tweak needs far more, typically thousands of visits per version once your starting rate is a few percent.
Here is a worked example, using round numbers. Say your page converts at 4% and gets 500 visits in a version. That is 20 enquiries (500 × 4%). If version B gets 23 enquiries from its 500 visits, it looks 15% better (23 ÷ 20). But three extra enquiries is well inside what chance alone produces on numbers this small. Run the same test again and the result could easily flip.
That is the trap. Small numbers feel like proof because a percentage sounds precise. They are not. If you want a feel for how your own traffic and enquiry rate play out, put them through the ad spend calculator and see how many enquiries you really get in a month.
What does a test with enough traffic look like?
For Jim's Mowing, we split their Meta ads at random for eight weeks between their main website and a quote page we built. From 3 August to 29 September 2026, 121 of 1,053 visits to our page led to an enquiry (a form or a tap to call), against 19 of 477 visits to the main website, measured in Google Analytics. That is 11.5% against 4%. See the results for the full case.
Two things made that test readable. The gap was large, nearly three times the enquiry rate, so about 1,500 visits was enough. And the traffic was paid, so it arrived steadily every day. A small gap on a trickle of organic visits would have taken far longer to call.
This result is from one client with a specific page. It does not mean your change will triple anything. It shows what a clear answer needs: a big difference, steady traffic and a fair split.
What should you do if you do not have enough traffic?
Most small business sites sit here, and it is not a dead end. You have three honest options.
- Test bigger changes. Replace the whole page with one that has a single job, instead of changing a button colour. Large differences show up on small numbers.
- Compare over a longer window. Four to eight weeks is common. Stopping a test after three days because one version leads is the most common way to get a false winner.
- Fix the obvious problems without a test. A slow mobile page, a form with ten fields, a headline that does not match the ad. You do not need a split to know a broken thing is broken.
Our guide to landing page design lists the principles we start from before any testing, and they are a good checklist for the fix-first route.
If you are not running paid traffic at all, we would usually put the money into getting steady visits first and test after. Testing a page nobody visits is not a good use of an afternoon.
There is a middle path too. Look at what visitors do on the page you have now. Watch a few session recordings, check where the form is abandoned and read the questions people ask when they ring. Ten minutes of that often shows the one change worth making, and you can then test it on the traffic you do have.
The aim is not to run tests. The aim is more enquiries from the visitors you already pay for. Testing is one way to get there, and for a small site it is often not the first one.
What mistakes make a test lie?
- Stopping early. A lead on day three is noise. Decide the length before you start.
- Changing five things at once. If B wins, you do not know why, so you cannot repeat it.
- Judging on the wrong outcome. More enquiries that never become customers is not a win. Check the result against what you actually sold.
- Mixing traffic sources. Visitors from a cold ad and visitors who typed your name behave differently. Keep the split inside one source.
- Testing before the basics work. If your tracking is off, every result is wrong. Check that a test enquiry shows up in your analytics before you start.
- Running tests on top of each other. Two tests on the same page at once muddy both results. Finish one, then start the next.
A losing test is not a failure. It tells you your guess was wrong, and that saves you from rolling out a change that would have cost you enquiries. Keep a short log of what you tested, how long it ran and what happened, so the next person does not repeat it.
It also helps to know your baseline. Our guide on what a good conversion rate is explains why a rate only means something against the same traffic and the same definition of a conversion.
When is it worth paying for help?
Testing is worth paying for when you spend steadily on ads and your page is the weak link. It is not worth it if your traffic is thin, your offer is unclear or your follow-up is slow. Those are bigger leaks than any page test.
If you do hire someone, ask them to show a result where the split was random and the outcome was enquiries or sales. Our guide on how to choose a CRO agency has the questions to ask. And if you already run ads and want a second opinion on where they leak, we offer a free ads audit.