This is how you do it, but don't send 100 and then a 100, use a much bigger sample size to make sure you are climbing up the hill correctly. In other words, do

This is how you do it, but don't send 100 and then a 100, use a much bigger sample size to make sure you are climbing up the hill correctly. In other words, do A/B testing in a highly statistically significant manner until you find what works for your service. You can grab things from other sites like Linked In to try to climb up the hill faster.