Lead Data Scientist @ GrowthBook
Only 10–30% of experiments produce a clear winner — and that’s not a problem, it’s reality. This article shows how high-performing teams design experiments to learn faster and make better decisions, even when results are neutral.
Most holdouts measure only shipped features. Ours measure everything—including failed experiments. This technical deep dive reveals why we chose reality over clean rooms, and how we built it.
Many teams ship features weekly but struggle to measure their true cumulative impact. Holdouts in GrowthBook provide a simple way to maintain a control group across multiple features, answering the critical question: What did all this shipping actually do to our key metrics?
GrowthBook provides you with 3 different types of experimentation for different purposes: Bandits for picking a winner among many, Safe Rollouts for releasing safely, and Experiments for learning.
Discover how Multi-Armed Bandits revolutionize A/B testing in GrowthBook. Learn to optimize experiments automatically, reduce opportunity costs, and make data-driven decisions faster.
GrowthBook's updated Bayesian engine enables new capabilities and improved estimation in small sample sizes.
Running experiments can be a messy business and you often want to make changes mid-experiment. Making changes to running experiments
CUPED is now available for the Frequentist engine in GrowthBook 2.0!
Experiment analysis queries in GrowthBook now run up to 2X faster! Read about the improvements
In under two minutes, GrowthBook can be set up and ready for feature flagging and A/B testing, whether you use our cloud or self-host.
