PlanetScale growth experiments
A/B testing for PlanetScale’s marketing site, GrowthBook on top of the existing analytics, with variants assigned in edge middleware and served as prerendered static pages.
- Company
- PlanetScale
- Date
Setup
Analytics was already in place; what was missing was a way to test against it. I set up GrowthBook and wired it into our own analytics, a private package with its own server, rather than a separate third-party pipeline.
The site was static and had to stay that way. The usual client-side variant swap costs a flash of the original and a second render behind it. Assignment happens in edge middleware instead: a cookie decides which variant a visitor is in, and the request is rewritten to a separate prerendered route for it. Every variant is built ahead of time, so the response is static either way.
Most of the work after that was verification: confirming each visitor saw the variant they'd been assigned, every time, and that exposures and conversions were measured correctly as they moved through Segment, our CDP. Tracking runs server-side at the proxy, next to where assignment already happens. It held up better than the client because it isn't dropped by ad blockers, so the counts are closer to complete.
Experiments
Three, once the setup held: one on the homepage, one across a few product pages, and one on signup.
The homepage tested the control against a shorter, summarized version. The summarized version performed better and became the default.
Signup tested the standard self-serve flow against sending people to contact instead.