The min. sample size for an A/B test can be [calculated](https://www.evanmiller.org/ab-testing/sample-size.html) from the picks at which rate to avoid type II and I errors, as defined by [statistical power and significance](Power%20and%20Significance.md), $\beta-1$ and $\alpha$, and the [effect size](Effect%20Size.md). The math behind Even Miller's sample size calculator is described in this [Intuition Behind A/B Sample Sizing](https://towardsdatascience.com/intuition-behind-a-b-sample-sizing-cb7e9c4fb992) article. Typical values are 95% significance ($\alpha = 0.05$) at 80% power ($\beta = 0.2$), because making a false negative mistake (rejecting the new feature when it had the desired effect) is less costly than a false positive.