Changing your site at random is playing the lottery. A/B testing means testing two versions of the same page at the same time to measure which converts better. No intuition, no opinions: only real data.
Understanding the principle in 30 seconds
You keep your current page (version A) and create a variant (version B) with a single element changed. Your visitors are split randomly between the two. After a sufficient number of visits, you compare the conversion rates and apply the winning version.
Fundamental rule: never test several elements at once. If you change the headline AND the button colour, you will never know which of the two made the difference.
The 4 elements to test first
- Your page headline (the highest-impact element in 80% of tests)
- The copy and colour of your call-to-action button
- The length and fields of your contact form
- The images or photographs used in the hero
The tools for starting without a budget
To analyse your visitors' behaviour before testing, start with Microsoft Clarity (free). It records sessions and generates heatmaps so you can see where people click and at what moment they abandon your page.
For A/B tests proper, VWO and AB Tasty offer plans accessible to small companies. Both let you launch a first test in under an hour, without touching your code.
How to read your results without going wrong
The main beginner's mistake: stopping the test too soon. If you have 50 visitors on each version and B looks 5% better, that is not statistically significant. You need at least 100 to 200 conversions per variant to draw a reliable conclusion.
The key concept here is statistical significance. Optimizely and HubSpot offer free calculators for checking whether your result is reliable. Aim for confidence above 95% before declaring a winner.
To track the conversions of each variant, connect your tests to Google Analytics. You will be able to measure goals (clicks, form submissions) and compare the two versions in your dashboard.
The mistakes that distort your tests
Seasonal bias, site outages or advertising campaigns launched mid-test can distort your results. Nielsen Norman Group recommends noting external events (sales, emails sent, press mentions) on your test calendar so you can contextualise any unusual spike in traffic.
A good A/B test runs for at least 2 weeks, even if you reach statistical volume sooner. That captures the differences in behaviour between weekdays and weekends.
Where to start today
Start with the page that already generates traffic but few conversions. Generally that is your homepage or your contact page. Identify the most visible element, change only its copy, and launch your first test. The first test is always imperfect. That is normal.
If your page does not convert at all, A/B testing will not solve the structural problems. Read why your landing page does not convert first, before moving on to optimisation through testing.