I have a question about percentage splits as they ...
# ask-questions
I have a question about percentage splits as they relate both to Feature Flags and to Experiments: If we have a new site element that we've built as a feature flag and want to run an A/B experiment for all traffic and split at 50/50 - should my feature flag be set to 100% and the experiment split set to 50%/50%? What happens if the feature flag is set to 30% and then we run an experiment with a 50/50 split? @fresh-football-47124
Hi Matt - The experiment name and the user id will be hashed to get a value from 0 to 1. This decimal is deterministic and will remain the same for this user (assuming nothing else changes). If you are running an experiment at 30%, with a 50/50 split, you'll get 15% of your traffic in each variation. What this actually does is select users who's hashed value is from 0 to 0.15 will get A (control probably) and then 0.5 to 0.65 will get B. As you ramp up exposure the decimal ranges will increase all the way to 0 to 0.5 and 0.5 to 1 -this way we avoid having users switch variants on percentage changes (and also why we recommended only adjusting the overall exposure)