Most tests you start manually and stop when the verdict is in. Some
tests benefit from a schedule, a Black Friday hero, a launch-day
homepage, a seven-day validation on a limited feature. That is what the
Schedule section handles.
Availability. Scheduled starts, ends, and max-run-days are on the
Starter plan and above. Free users can still start and stop
manually. See pricing. Three fields
Under Schedule in the editor, you can set any combination of:
- Start date and time. When to flip the experiment from Draft to
Live. Defaults to “start when I click Start now”.
- End date and time. When to auto-stop. Blank means run until you
stop it.
- Stop after (days). A safety cap, labelled “Stop after” in the
editor. If set, the test auto-stops that many days after it actually
starts. It counts 24 hour days from the start instant, not calendar
days, so a cap that spans a daylight-saving change lands an hour off
the original local time.
The start and end fields use your project timezone, which is set
once per site under Settings → General. If your
project is America/New_York, “9 AM” means 9 AM Eastern. The editor
labels each field with the zone and its current UTC offset.
What gets stored is the exact moment your wall clock time resolves
to, not the wall clock time itself. If you schedule 9 AM Eastern in
October and the experiment starts in November, after the clocks change,
it starts at 8 AM Eastern. Changing your project timezone later
re-displays an existing schedule, it never moves it.
Use Schedule for later, not Save changes
Fill the schedule fields, then click Schedule for later at the
top-right. That one button does everything: it saves the experiment and
flips its status to Scheduled.
Save changes does not apply a schedule. On a draft it saves
everything else and silently drops the start, end, and stop-after you
just typed. The editor warns you when this happens, with a “Saved, but
not scheduled” message. If you see it, click Schedule for later to
actually schedule the test.
Auto-stop lands in Paused
When the end date arrives (or max-days is hit), the experiment
transitions to Paused, not Archived. Nothing is deleted. The
results page stays available and any team member with the right role
can review, then choose to Resume or Archive.
Auto-stop does not notify anyone. There is no email and no alert when a
test stops itself. The experiment moves to Paused and waits. If your
day-seven test needs a day-eight look, put that reminder in your own
calendar.
What happens to bucketed visitors on auto-stop
Existing bucketed visitors stop seeing the variant on their next
pageview, once the new configuration reaches them. Allow a few minutes
end to end: up to about a minute for the stop itself, then up to about
three more while the previous configuration ages out of the caches
between us and the visitor’s browser. Historical data is preserved and
the results page keeps working.
Scheduled + activation triggers
A scheduled experiment respects its activation trigger like any other.
If your test is set to On click and cron starts it at 9 AM, the
first exposure fires when the first eligible visitor clicks, not at 9
AM sharp.
Corner cases
Start date in the past, rejected. Use Start now instead.
End date before start, rejected.
Less than five minutes between start and end, rejected. The whole
window could otherwise elapse between two runs of the job that starts
and stops experiments.
End date without a start, allowed, but it cannot be scheduled.
Moving an experiment to Scheduled requires a start time in the future.
An end date on its own takes effect when you click Start now, and
becomes the auto-stop.
Stop after and end date both set, whichever comes first wins. 
