#ab-testing

An A/B test of an app (1% of users in each arm) estimated the change in daily visitors as +1.2%, with 95% confidence interval (−0.2%, +2.6%). The team wants a new experiment that detects a change of 1.0% with 80% power, using a two-sided test at $\alpha = 0.05$. The standard error shrinks like $1/\sqrt{n}$. By what factor must the size of each arm grow?