Skip to content

Latest commit

 

History

History
36 lines (24 loc) · 1.86 KB

a_b_testing_best_practices.md

File metadata and controls

36 lines (24 loc) · 1.86 KB

A/b Testing best practices

What will you need to run a successful experiment?

  1. An admin feature flag while you are developing the feature.

    This flag has only one goal: Is the feature ready to be released to all users?

Want to learn more about feature flags, see our docs on developing a feature.

  1. An Unleash experiment flag

    This is the flag that will handle the split to all users. Don't use this flag as a regular feature flag!

Want to learn more about adding/using an experiment to Eigen, see our docs on adding an experiment.

How to properly QA the release?

It's important to ship experiments with the least updates and bug fixes to them as possible. This is important because it can have an impact on the experiment results.

Below is a non exhaustive list of areas to QA

  • The control variant functionality.
  • The experiment variant functionality.
  • Tracking for both the control and experiment variants.
    • The tracking event is triggered only once per screen!
    • This is important to make sure that we don't end up triggering the same event multiple times and breaking the split ratio.

Want to learn how to switch between different variants, see how to add an override

Ready for release checkboxes!

  • My feature has been QA'd and it was approved by product, design, and data!
  • My feature toggle (the admin feature flag available in features.ts) readyForRelease is set to true.
  • My feature toggle is available in Echo and is set to true
  • My feature experiment flag is enabled for all users in unleash.artsy.net
    • The standard strategy is ON for all users in production
    • The different variants split is also available in production (you can use copy to environment to get the split from staging)