A/B testing splits your traffic between two or more versions of a bundle so you can see which earns more. Unlike page variations, this is a real split test: Atlas divides the traffic itself and reports the result. By the end of this you will have a test running, and you will know what happens when you end it.
In this article:
Before you start
The bundle must be saved. Until it is, the A/B test tab shows a banner rather than the controls. A variant is a copy of the bundle, so there has to be a stored bundle to copy.
ℹ️ A bundle targeting specific products or specific collections can be tested exactly like one targeting all products. Only the save matters.
You need enough traffic. A test on a product with a handful of visitors a week will not produce a usable answer in any reasonable time.
Step 1: Enable A/B testing
Open the bundle and go to the
A/B testtab.Select
Enable A/B testing.
Step 2: Create a variant
Select Add variant. A variant is a full copy of the bundle that you then change.
Change one thing. If the variant differs in its discount, its badge, its colours and its wording all at once, a win tells you nothing about which change caused it.
Good single things to test:
Test | Why it is worth knowing |
Which offer is preselected | Often the single largest effect available |
Two tiers against three | More choice is not always better |
Discount depth, 10% against 20% | Tells you the shape of your demand curve |
Badge on the middle tier against the top | Moves customers between tiers |
A free gift against a deeper discount | Which your customers actually value |
Save the variant with Publish Variant. Remove one with Delete variant, which asks Are you sure you want to delete {variant}?
Step 3: Let it run
Atlas splits traffic between the original and your variants automatically. There is nothing to configure and nothing to point at the variant.
⚠️ Leave it alone while it runs. Editing a bundle that is in a test changes the thing being measured, and the results before and after your edit are no longer comparable.
Run it long enough to cover a full weekly cycle. Weekend and weekday buyers often behave differently, and a test that ran Tuesday to Thursday can point the wrong way.
Step 4: Read the results
The AB test analytics view compares bundle variant performance to find your highest-converting offer.
Metric | What it tells you |
| The single best summary. It combines conversion and order value. |
| Percentage of visitors who placed a bundle order |
| Average revenue per bundle order |
| Extra revenue beyond the base product price |
| Share of bundle orders that included a free gift |
| Share that included an upsell |
| How each variant compares against the |
The variant currently ahead is marked Winner.
💡 Judge on revenue per visitor, not conversion rate. A variant with a bigger discount will almost always convert better and can still earn you less per visitor. Revenue per visitor is the number that pays you.
If there is nothing yet you will see No variants data available. A bundle that is not in a test returns This bundle is not part of an A/B test.
Step 5: End the test
Select
End A/B test.Choose the variant you want to keep with
Choose variant.Confirm.
What ending a test destroys
⚠️ Once you select your winning variant, all other variants will be deleted and all traffic will be directed to the selected variant. The losing variants are deleted permanently and cannot be recovered.
Before you end a test, write down the numbers for every variant, including the losers. Knowing that a 20% discount earned less per visitor than 10% is worth as much next quarter as the winner is today, and after you end the test that comparison is gone.
Once ended, all traffic goes to the variant you chose, and it becomes the bundle.
Testing well
Test one change at a time. Otherwise you learn that something worked without learning what.
Give it enough traffic. A difference of a few percent on a hundred visitors is noise. The cart drawer test view is explicit that at least 1,000 views are needed for reliable results, and the same order of magnitude applies here.
Run whole weeks. Partial weeks skew towards whichever days you happened to include.
Do not peek and stop early. Ending a test the moment one variant edges ahead is how you convince yourself of things that are not true.
Keep a record. Atlas deletes the losing variants when you end the test, so your own notes are the only lasting record of what you learned.