The experiment workflow: Roll out your experiment

  • Updated

This article will help you:

  • Launch your experiment to your users

Once you've defined your experiment goals, set up your variants, and allocated your traffic, launching your experiment is as simple as clicking Start Experiment, in the upper-right corner of your screen.


Notice the Inactive toggle switch just above the Start Experiment button? When you click Start Experiment, that switch will toggle to Active automatically. However, you can also use this toggle to activate your instrumentation without starting your experiment. By switching this toggle, you can:

  • Test the flag without actually launching your experiment
  • Rely on fetch() to get user assignment data from the Amplitude servers
  • Trigger automated exposure events

However, this will not set an experiment start date; clicking Start Experiment does that.

All changes must be saved before clicking Start Experiment.

NOTE: Start Experiment only activates the experiment once. Changing the start date will not automatically trigger the experiment to activate on the new start date.

Once the experiment is running, the button is relabeled Complete Experiment. You will be able to click this button again when you reach the experiment's end date, or when the experiment hits statistical significance. At that point, you can do one of three things:

  • roll out the winning variant
  • roll back everything and return to a pre-experiment state, or
  • continue the experiment

You can always revisit this decision after you've made it.

What happens when your experiment ends?

If you roll out your experiment to all users:

  • Percentage rollouts are set to 100%
  • Sticky bucketing is set to false
  • The rollout weights change, to 1 for the variant you're rolling out and to 0 for all other variants

If you roll out your experiment to “custom,” the automatic changes listed above will not occur. You will have to apply changes manually after confirming your rollout decision.

If you roll back your experiment:

  • The flag is turned off
  • Percentage rollouts are set to 0%

If you opt to continue running, your experiment, you can enter a new end date.

Now that you’ve rolled out your experiment, the next step is to learn from it.