A/B tests
An A/B test shows two versions of your site to different visitors at the same time and counts which one gets more of them to call, email, send a form or click your main button. You pick a saved version, Briks runs it beside the live one, and when the numbers are clear you keep the winner.
Before you start
- Two versions. One side of a test is always the version that's live when you start. The other is any saved version you pick. So make the change you want to test first, then decide which version should be live.
- Real visitors. Each version needs about 100 visitors before the result means much. A preview address is kept out of search engines and rarely gets that many, so most tests belong on a site in production.
- Edit rights. Editors, Admins and Owners can start and end tests.
- A quiet moment. A test can't start while something is building, and a site runs one test at a time.
Start a test
Make the version you want to test
Where: Studio
Ask for the change, such as a new hero headline. It becomes the newest version and goes live.
Decide which goes live
Where: Preview toolbar
If you'd rather the old version stays live while you test, press undo. Either way works: one side is live, the other is the one you pick.
Pick the other version
Where: Versions
Open Versions, find the version that isn't live and press Test against live.
Wait for it to publish
Where: Studio
Briks publishes both versions together and replies “The A/B test is live: new visitors are split between the two versions, and each visitor keeps the one they first saw.”
Find the results strip
Where: Preview
A strip headed A/B test appears above the preview, marked Running.
How visitors are split
When you press Test against live, choose how many new visitors see the version under test: Half (50%), A quarter (25%), A tenth (10%), or Custom for any share from 5 to 95 percent. Then press Start test. The strip says how the split stands, for example “90% see v11, 10% see v12”. A smaller share is gentler on a bold change, but takes longer to give a clear answer.
Each visitor keeps the version they first saw, through a cookie that lasts 30 days. Every page they open comes from their version, so someone who reads the home page, leaves and comes back next week still sees the same one.
What counts as a conversion
| What a visitor does | Counts as |
|---|---|
| Taps a phone number link | A conversion |
| Taps an email address link | A conversion |
| Sends any form on the site | A conversion |
| Clicks the site's main action, such as Book a visit | A conversion |
| Opens the site | A visitor |
| Does two of these, or one twice | One conversion |
Each visitor counts once as a visitor and at most once as a conversion. Counts from one internet address are also capped at 20 an hour, so nobody can tilt a test by reloading.
Read the results
The strip refreshes every 15 seconds while the studio is open. Each version has its own box: the share of its visitors who converted, then a line such as “18 of 410 visitors converted”. The version that was live when you started is marked “live before the test”. Under the boxes, one line sums it up:
| The line says | What it means |
|---|---|
| “Collecting visitors. Each version needs about 100 before the result means much.” | At least one version has fewer than 100 visitors. Keep waiting. |
| “No clear winner yet.” | There's enough traffic, but the difference could still be chance. |
| “Version 12 is very likely better.” | There's at least a 95 percent chance that version converts better. |
The same line gives the Change in conversion (how much better or worse the version you picked does, in percent) and the Chance that it converts better.
End the test and keep the winner
Choose the version
Where: A/B test strip
Press Keep v11 or Keep v12.
Wait for the publish
Where: Studio
Briks publishes the kept version for everyone and replies, for example, “The test has ended. Version 12 is live for everyone.”
Carry on
Where: Studio
The kept version is the live one now. Your next change starts from it, and undo and redo work again.
Keeping a version is the only way to end a test. Starting and ending one hold no credits, because both publish saved versions without building.
What's locked while a test runs
Both versions are published together, so anything that would publish something new waits until you keep one. While a test runs you can't:
- ask for a change (“An A/B test is running on this site. End it by keeping one version, then ask for your change.”);
- undo, redo or press Make live;
- start another test;
- save code from the File Explorer, publish from the Visual Editor, or Replace a picture on the Media tab;
- use Apply to the live site, Publish or Back to preview;
- save the project as a template, or change its address.
Questions still work: you can ask Briks about the site at any time. See Asking questions.
Worked example: a plumber tests two headlines
Harbour Plumbing is a sample business: a two-person firm in Bristol whose site is in production. Its hero headline reads “Plumbing and heating repairs across Bristol”. The owner suspects a promise about speed would bring more calls.
Change the hero headline to: A plumber at your door within two hours. Keep everything else the same.
v12 went live with the new headline.
The owner pressed undo so the original headline (v11) was live again, opened Versions, and pressed Test against live on v12.
Half of new visitors saw v11 and half saw v12. Taps on the header's phone link, sent contact forms and clicks on Book a visit all counted.
v11, live before the test: 4.4%, 18 of 410 visitors converted. v12: 7.8%, 31 of 398 visitors converted.
“Version 12 is very likely better. Change in conversion: +77.4%. Chance v12 converts better: 98%.”
The owner pressed Keep v12.
“The test has ended. Version 12 is live for everyone.” Their next change, adding Sunday call-out prices, was made from v12.
Can I test two versions when neither is live?
No. One side is always the version that's live when the test starts. Make the other one live first with Make live if you need to.
Can I change the split?
Yes, when you start the test: half, a quarter, a tenth, or any share from 5 to 95 percent. A running test keeps its split; to use another, end it and start a new one.
Who sees the results?
Everyone in the workspace who opens the project's Studio tab, Viewers included. Only Editors, Admins and Owners can end the test.