What has surprised me recently is how unscientific these claims can be sometimes. Some bad statisticians behind the numbers. The sort of people that would flip a coin 100 times, get heads 57 times and tails 43 times, and come to the conclusion that heads outperforms tails by about 30 percent.

I assume KISSMetrics used a large enough sample size in their experiment for their results to be statistically meaningful, but some recent sites I’ve come across online that purely focus on A/B testing have draw conclusions from tests with tiny sample sizes and tiny percentage differences.