Skip to main content
A/B testing lets you try more than one version of an email and find out which performs best. In Norbelys, each email step can have several versions (A, B, C…), and two or more versions turn that step into a test.

​
Create a version

In the email editor, the variant rail shows your versions. Use “new version” to add one — start blank, duplicate the current email, or generate with AI (“write a genuinely different version”). Give each version genuinely different copy so the test tells you something.
Content is immutable once it’s sending — editing a version that’s live actually creates a new version. This keeps your results honest, so a “winner” always refers to exactly the copy that was tested.

​
Choose how versions are sent

Open the A/B distribution settings for the step. You have two modes:

Manual distribution

You set the share each version gets. Enter a weight per version, or press Split equally. Everyone is divided by your percentages.

AI distribution (Smart)

Test on a sample of your audience, then automatically send the best version to the rest.

​
How AI distribution works

Smart A/B distribution: a test group receives each version, the winner is chosen by reply rate, and it's sent to the winner group.

Test a sample of your audience, then the best version by replies is sent to everyone else.

1

Pick a test group size

A slider splits your audience into a test portion and a winner portion — for example 30% test, 70% winner.
2

Choose the deciding metric and threshold

Set what “best” means (default: replies) and how many sends per version are needed before deciding.
3

Let it run

The test group gets versions in the shares you set. A live table shows each version’s sends and score, with Leading and Winner badges.
4

The winner locks

Once every version reaches the threshold, the winner locks automatically and the rest of your audience gets it. You’ll see “Version B won” — and you can Reset test if you want to run it again.

​
Why replies, not opens?

The default winning metric is replies, and that’s deliberate. Opens are unreliable — mail providers and privacy proxies fire your tracking pixel automatically, so a high “open rate” can be mostly machines. Replies are a real human action, so they tell you which version actually worked. You can choose a different metric, but replies are the honest default. See Analytics for more on why opens can’t be trusted.
Archiving a version is always safe — even the winning one — because a winner is a stable pointer to the copy that was tested, not something that can be accidentally edited out from under your results.