Run an A/B test between two campaigns
Split new shoppers between two versions of a campaign up to a number you set, then compare the results side by side.
Last reviewed November 7, 2026
How a test works
A test copies each base campaign into a hidden variant. While it runs, new shoppers on the same lead source go to the two variants at random, nudged toward whichever is behind so the split stays even. A shopper still has to pass the variant's checks, and one both variants turn down falls through to your other campaigns.
Make the difference first
You cannot open or edit a variant once the test exists, so build the two versions before you create it.
- Duplicate the campaign you want to test.
- Change one thing in the copy, such as the discount code. It does not need to be published.
- Use the original and the copy as the two base campaigns. Both must have the same objective and lead source, and neither can be archived.
If you leave Variant B on Same as Variant A, both variants are identical.
Create and start the test
- Open Recovery tests under Tests in the sidebar (the A/B tests tab on Campaigns leads there too) and press Create A/B test under Campaign A/B tests.
- Enter a name, then choose a Base campaign and a Variant name for each side.
- Under Number of shoppers, set the size of the test, from 10 to 1,000,000. It starts at 500.
- Press Create test. The test shows Not started.
- Press Start.
Start publishes both variants and switches their triggers on straight away, without asking again. If a variant's setup is incomplete, the start fails and names what is missing.
Read the results
| Column | What it counts |
|---|---|
| Sent | Shoppers the test sent to this variant |
| Runs | Runs in this variant |
| Engaged | Runs where the shopper engaged |
| Recovered | Runs with a credited order, or that ended positive or handed to your team |
| Recovery rate | Recovered divided by Runs |
| Revenue | Credited revenue from this variant's runs |
Recovered is wider than its name suggests, because a positive reply or a hand-off counts even with no order. To compare orders, use Revenue.
When it ends
The test shows Finished when it reaches your number. Stop ends it early and shows Stopped. Either way, new shoppers go back to your ordinary campaigns, and runs already open in the variants carry on.
Variants stay hidden and there is no button to promote a winner. To roll one out, make the same change in your main campaign and publish it.