Typically A/B tests are run until they reach a statistically significant result. The article doesn't include the sample size, but with a sample size as small as 500 users for each test an increase from 1.6% to 3.2% would be a change larger than 1 standard deviation, and thus can be assumed to be more than just random noise.