A rollout serves a change to a share of your visitors. There is no second version and nothing to compare, so there is no winner.
Use one when the question is already answered. A test that found a winner stops when you stop it, and the change stops with it. A rollout is that same change without the test around it, so it stays on the page while your engineers build it properly. It is also the way to put a change live for a tenth of your traffic first and watch nothing break.
How it differs from a test
| Test | Rollout | |
|---|---|---|
| Versions | Two or more, compared | One |
| Metrics | Required, one decides it | None |
| Result | A verdict | No verdict |
| What it reports | A full report | How many visitors it reached |
| How it ends | You stop it, or a schedule | You switch it off |
A rollout is judged by nobody. No metrics, no significance, no traffic split check. Those all exist to choose between versions, and a rollout has one version. If you want to know whether a change works, run a test.
How to set one up
- Open Rollouts in the sidebar and click New rollout.
- Type a Name, such as "Green checkout button". Save it.
- On the Pages tab, say which pages carry the change.
- On the Changes tab, open the editor. It is the same editor a test variation uses, and it opens the page you named under Pages.
- On the Audience tab, narrow who it reaches, or leave it to reach everybody who matches your pages.
- On the Settings tab, set the Share of traffic.
- Press Switch on.
A new rollout starts as a draft and reaches nobody until you switch it on.
What each label means
| Where | Label | What it means |
|---|---|---|
| Settings | Name | What you will recognise it by in a list next year. Up to 120 characters. |
| Settings | Share of traffic | A share of the visitors its rules match, not of everybody. From 0.01% to 100%. |
| Pages | Page to edit | The page the editor opens. Up to 2048 characters. Your homepage if you leave it empty. |
| Pages | Page rules | Which pages carry the change. Up to forty rules. Any one of them matching is enough. |
| Changes | The change | Authored on your own site, in the same editor a test variation uses. |
| Audience | Audience | Who it reaches, built the same way as a test's. See Audiences. |
On the rollouts list:
| Column | What it shows |
|---|---|
| Rollout | Its name, and whether it is a draft, live or off. |
| Changes | How many changes it carries. |
| Reached | How many visitors have seen it. "Nobody yet" until one does. |
A share of zero is refused. A rollout reaching nobody is one that is switched off, so use Switch off rather than setting the share to nothing.
Good to know
- Switch off is the undo. It is one press and it takes effect on the next page load. Reach for it first if something looks wrong, before you start editing.
- A rollout does not take part in mutual exclusion and does not reserve traffic from any test. A visitor can be in a test and see a rollout at the same time.
- Because of that, a rollout changing the same element a running test changes will muddy that test. Check what is live before you switch one on.
- Reached counts visitors, not page views. One person seeing the change five times counts once, so the number is smaller than your traffic and is meant to be.
- The share is of matching visitors, not of all your traffic. Ten per cent of a rollout scoped to your pricing page is ten per cent of pricing page visitors.
- Each visitor gets a consistent answer. Somebody in the share on Monday is in it on Tuesday, on the same browser.
- Write the name for somebody reading it in a year. "Green checkout button" tells them what is on the page. "Test 4 winner" does not.
Related articles
- Your first A/B test: finding out whether a change works first.
- Audiences: narrowing who a rollout reaches.
- Install the tag: the script that serves the change.