We tested it using Optimizely, what we use for A/B tests, but rather doing a true A/B test we made into a sequential test (by manually setting all traffic to go to one variation versus another). It took a couple days to test and it's tough because since it's sequential traffic spikes will screw everything up and you'll have to start all over. So basically we compared two weeks with "average" traffic: running $5 plans one week and free plans the next week. Obviously if anyone noticed we allowed them to switch their plan to free.
