Test offers and read the results
An A/B test runs two versions of an offer at once and splits traffic between them, so you can see which earns more. Set one up under Upsell → A/B testing. The Analytics tab shows what every offer earned. Test one change at a time, and give it at least two weeks.
What this does
Guessing at discounts is expensive. A test answers the question with your own customers instead of an opinion.
Create a test
- Go to Upsell → A/B testing.
- Select Create test.
- Name it, and pick the placement to test on.
- Choose your success metric — what counts as winning.
- Set a target confidence, and a start and end date.
- Set how traffic splits between the two versions.
- Save.
Test one thing at a time
The app says it on the page, and it matters: test one change at a time, and run it for at least two weeks.
Change the discount or the layout or the wording. Change two and a result tells you nothing — you will not know which one moved it.
Two weeks is the minimum because order volume swings by day. A test that runs Friday to Monday measures the weekend, not the offer.
Read the results
Each test shows both versions side by side:
- Views — How many customers saw it
- Conversions — How many took it
- Revenue — What it earned
The stronger version is marked Winner, and the original is marked Control.
Above the list, four figures cover everything: Active tests, Completed, Upsells added, and Revenue.
The Analytics tab
Separate from testing, Upsell → Analytics shows how offers are doing overall — revenue, conversion rate, and which offers performed best.
Use analytics to decide what to test. Use tests to decide which version wins.
Good to know
Do not stop a test early. An offer that looks ahead after three days often is not. Wait for the end date.
A losing test is still useful. Learning that a 20% discount earns no more than 10% saves you 10% on every future sale.
Views count impressions, not customers. One customer who reloads the page counts twice. Compare the two versions against each other, not against your order count.
Tests need traffic. With a handful of orders a week, a two-week test will not tell you much. Improve the offer instead, and test once you have volume.
Troubleshooting
- The test shows no data — Why: The offer's app block is missing from that page — Fix: Add it, then restart the test
- Both versions look identical — Why: More than one thing changed — Fix: Test one change at a time
- Views but no conversions — Why: The offer is not tempting, or the product does not fit — Fix: Try a different product before a bigger discount
- The test ended with no winner — Why: Not enough traffic to tell them apart — Fix: Run longer, or accept there is no real difference
- Analytics and test numbers differ — Why: They measure different things — Fix: Analytics covers all offers; a test covers only its own
- The test will not start — Why: The placement has no active offer — Fix: Publish an offer on that placement first
FAQ
How long should a test run?
At least two weeks. Shorter than that and you are measuring which days of the week fell inside the test.
Can I test more than two versions?
A test compares two. To try three ideas, run one test, keep the winner, then test that against the third.
What should I test first?
Whether a discount helps at all. Run your current offer against the same offer with no discount. Customers who have just bought are often willing without one.
Does testing slow anything down?
No. Each customer sees one version, chosen when the page loads.
What if both versions perform the same?
That is a real answer. Keep the cheaper one — usually the smaller discount — and test something else.
Related articles
Updated on: 21/08/2026
Thank you!