Guides
Managing experiments
Creating, editing, starting, pausing and completing experiments, and what each status means.
Creating an experiment#
Go to Experiments and choose New experiment. You set the name, the hypothesis, the traffic allocation and the variations. New experiments always start as Draft. They are created in the project selected in the top bar.
Targeting#
In the Traffic step, Who can enter limits a test to certain visitors. Leave it empty to include everyone. All rules you set must match.
| Rule | How it works |
|---|---|
| Devices | Desktop, mobile or tablet, based on the browser. None selected means all devices. |
| utm_source, utm_medium, utm_campaign | The value in the page address must equal what you enter. Case does not matter. |
| Referrer contains | The page the visitor came from must contain this text, such as facebook.com. |
| New or returning | New means this is the visitor's first browser session on your site. Returning means they visited in an earlier session. |
Visitors who do not match see the control page and are not counted. Rules are checked only when a visitor first joins, so people already in the test keep their version if you change the rules later.
Statuses#
| Status | Visitors assigned? | Notes |
|---|---|---|
| Draft | No | Set up and test your code. Drafts are the only experiments that can be deleted. |
| Running | Yes | New visitors are assigned and events are recorded. |
| Paused | No | Assign calls return assigned: false, so everyone sees your default page. Assignments are kept and resume when you resume. |
| Completed | No | Final. Read only, and cannot be restarted. Results stay available. |
Use the buttons at the top of the experiment page: Start experiment (draft), Pause and Complete (running), Resume and Complete (paused). Any other change, such as completed back to running, is rejected.
Note
Starting an experiment for the first time sets its start date and counts against your organization's monthly experiment limit, if one is set. Resuming a paused experiment does not count again.
Editing#
On the Settings tab you can change the name, the description and the traffic allocation. The Edit test tab lets you make safe changes, even while the test runs:
- Traffic split. Change the weights. Active weights always total 100.
- Pause a variation. It gets no new visitors. The control cannot be paused, and at least one variation must stay on.
- Who can enter. Change the audience rules.
- Add a goal. It counts from the moment you add it. You can make it the primary goal.
Every change applies to new visitors only. People already in the test keep their version, so your results stay clean. Variation URLs cannot be changed and variations cannot be added or removed after creation. If you need different pages, create a new test.
Changing traffic allocation#
Traffic allocation is the share of visitors who enter the experiment. Visitors outside it are not assigned and are not counted. While running:
- Raising it adds new visitors. Visitors who were already in keep their variation.
- Lowering it stops some new visitors from entering. Visitors already assigned keep their variation and keep being counted.
Deleting#
Only drafts can be deleted, from the experiment header or the Settings tab. Once an experiment has run, complete it instead so its history stays intact.
Audit log#
Every create, update, status change and delete is written to the audit log with who did it and what changed. A view of the audit log in the dashboard is coming later.
Coming later
Scheduling and duplicating experiments are planned.