Duplicating an Experiment

Learn what Pertento copies when you duplicate an experiment, and what it deliberately leaves behind.

2 min read

Duplicating an experiment gives you a new draft that starts from an existing one, which is the quickest way to run a second iteration of a test without rebuilding the variants. Use Duplicate from the actions menu on any row in the Experiment Library, and give the copy a name.

What is copied

The duplicate keeps the parts that represent your work:

  • Experiment type, description and configuration.
  • Every variant that is not a deployed rollout, including its visual changes, global CSS and global JavaScript.
  • URL targeting rules.
  • Device targeting rules.

What is not copied

Everything that belongs to the original run is deliberately left behind, because carrying it over would attach one experiment’s history to another:

  • All results. The copy has no sessions, no conversions and no statistics. It is a new experiment, not a continuation.
  • Start and end dates. The duplicate has never run.
  • Deployed variants. A rollout is not a test arm, so it is skipped rather than copied. If you duplicate a deployed experiment you get its test variants back, not the rollout.
  • Archived and deleted state. The copy arrives clean.
  • Cookie targeting rules. These are not carried over and need to be set again on the copy.

Traffic weights are reset

Any variant that had an explicit weight of zero, and the Original, come back unallocated rather than pinned to a number. This matters: a variant paused at 0% in the original would otherwise silently receive no traffic in the copy, and you would be waiting on results from an experiment that was never going to produce them.

Check the traffic split on the duplicate before you launch it.

The copy is always a draft

However the original ended up, the duplicate is created as a Draft. Nothing goes live and no visitor sees it until you launch it yourself, so it is safe to duplicate a running experiment in the middle of the day.

Notes

  • Give the copy a name that distinguishes it from the original. "Hero test" and "Hero test" are indistinguishable in the library a month later.
  • Re-check cookie targeting and traffic weights on every duplicate before launching.
  • Duplicating never changes the original in any way.