Evaluate experiments continuously with mSPRT, auto-stop guardrails, and CUPED. Explore GDPR-compliant A/B testing with EU hosting from Growth.
During the trial one test runs in parallel so you can try the feature for real. Starter does not include A/B testing. Growth allows three concurrent active tests, and from Pro the number is unlimited and CUPED analysis is added. Finished tests do not count against the limit.
That depends on your baseline rate and the effect you want to detect. Rough orientation: at a 3% conversion rate and a targeted 20% relative lift you need a few thousand sessions per variant. Sequential evaluation ends the test as soon as evidence is sufficient and CUPED lowers the requirement further — neither guarantees a fast answer.
Yes, that is exactly what sequential evaluation is for. With classic t-tests every peek raises the chance of a false positive; mSPRT keeps the error rate intact under continuous monitoring.
Guardrails check revenue per session, bounce rate and frustration signals against the control group daily. When a threshold breaks, the test stops automatically and you receive a Watchtower alert including the statistical details.
CUPED uses the same visitors' behaviour from before the test as a covariate and removes pre-existing differences. It helps most with returning visitors and high-variance metrics such as revenue per session. With purely new traffic and no pre-period the effect stays small.
No. The variant is derived from a hash of visitor ID and test ID. There is no separate assignment cookie and no additional consent purpose beyond your existing tracking.
Google Optimize was discontinued in 2023. Pure testing tools give you the result but not the context. In BigHoot the test, funnel, heatmap and session replay share the same sessions — you see not only that variant B wins, but where the control variant loses people.