GrowthBook 0.4.0 & August Update

August was quite a month for GrowthBook! We’re closing in on 1500 GitHub Stars, making us one of the most popular open source A/B testing platforms.

We’re also continuing to add many highly-requested features to the platform, based on your feedback in our Slack channel and on GitHub, and we just released version 0.4.0! Here are the major features of the 0.4.0 release:

Per-metric risk thresholds

There’s a trade-off between low. Now you can decide this tradeoff on a per-metric basis. So you can reduce risk and take your time with Revenue while moving quickly and accepting more risk with less important metrics.

Experiment-level dimensions

GrowthBook has always had the ability to define user-level dimensions to drill down into experiment results. But that didn’t work great for things like “browser” or “traffic source” since a user may use multiple devices or have many different sessions on your site over time.

Now GrowthBook also supports exploring risk and ending an experiment more quickly, with experiment-level dimensions for drilling down by these point-in-time attributes. All you need to do is select additional columns in your experiment query. Learn more about dimensions

Export configurations

Back in version 0.3.0, we added support for configuring self-hosted GrowthBook using a config.yml file instead of through the UI. This made it easier to version control and share configuration between environments. With the 0.4.0 release, you can now export settings that were created in the UI to a config.yml file at any time.

Detect variation ID mismatches

GrowthBook now detects and warns you when the variation IDs defined for an experiment differ from those returned from the database when fetching results. This is really useful for identifying typos and bugs in your data.

Custom rules when importing past experiments

When importing past experiments from a data source, GrowthBook applies a bunch of rules to make sure we keep the list of experiments clean. One of those rules requires a minimum runtime of 5 days in an attempt to exclude experiments that were stopped early. You can now customize this minimum length as needed.

Have ideas for other import rules or tweaks to how this feature works? Let us know!