A/B testing
Stop Guessing Which Version Works
A new headline, a shorter form, a different call to action. Test it on real visitors, and get an answer you can trust, including “your site is too quiet to tell”.
Try it
How Long Would a Test Take?
The same sample-size rule the product uses before a test starts. Move the sliders to your own page's numbers.
This test needs
21
weeks
About 5,855 visitors for each version. Longer than a test runs (6 weeks), so only a bigger difference would show. A running test like this is flagged as slow: test a bolder change, or a busier page.
A Finished Test, as You'd See It
Sample data, judged by the product's own statistics.
Job page: “Apply in one minute” vs “Apply now”
2,943 visitors split at random, 28 days
Apply rate, A and B
Chance B is better, day by day
How Tests Are Run
Built to give small sites honest answers.
Split at random
Each visitor is shown one version and keeps seeing it. A first-party cookie holds only a random number: no tracking cookies.
Safe to check every day
Results use a sequential test, so looking every morning doesn't crown a false winner. A plain 95% check looked at daily gets it wrong about one test in five.
Ends when it knows
A test stops once the answer is clear, or at its time limit. Keep the winner in one click, or undo it.
Judged on placements
Applications from each version are followed to interviews and placements. That's what's reported, as ranges, never a single invented number.
Tests say what caused what. Everything else says what happened next.
A test is a controlled comparison, so it can say a version caused more applications. Placements and fees that followed are reported as “went on to”, with ranges, because a handful of placements can't prove a cause.
See What You Could Test First
Your private preview shows your own pages on the new site, where every block can be tested.
Start free trial