Decide with a number, not with an opinion
Show two versions, count the conversions, and get a clear verdict: is the difference real, or is it noise that will vanish next week?
A verdict, not a leaderboard
Every comparison carries a p-value and says plainly whether the gap is significant at 95%. Declaring a winner on thirty visitors is worse than saying nothing.
How many more people you need
When the result is not conclusive yet, the screen estimates the sample per variant that would settle it — so you know whether to wait or to stop.
Assignment that never drifts
A person always gets the same variant: the choice is a hash of the experiment key and their identifier, computed the same way in the browser and on the server.
How it works
Create the experiment
A key your code will use, two to five variants with their weights, and the custom event that counts as a conversion. It starts as a draft: nobody is exposed yet.
Read the variant in your code
Ask the tracker which variant this visitor gets, and render accordingly. The exposure is recorded at that moment, not before.
Start, then stop when it is decided
Results count from the start, so a test relaunched later never inherits the figures of the previous run.
Frequently asked
Which statistical test is used?+
A two-proportion z-test at 95%, two-sided, against the first variant as control. The p-value and the sample needed are shown next to each variant.
Does the same person always see the same version?+
Yes, as long as their identifier is stable — the visitor identifier within a day, or the identity you provide with identify(). The experiment key is part of the hash, so two tests running at once do not split the same people the same way.
Can I run an experiment without writing code?+
No. PulseTrack decides who sees what and counts the results; rendering the variant stays in your own code, where it belongs.
Which plan includes experiments?+
All of them, including the free plan. Only your monthly event allowance applies.