You work on a chat platform’s growth team. The team has built a new server recommendation module shown in the server discovery entry points and believes it will increase the rate at which users join new servers and improve early retention. However, recommendations can create network effects: if treated users join and become active in servers, they may change the experience for control users in those same servers. You need an experiment design that can estimate impact despite that interference risk.
How would you design this experiment so that you can measure whether the new recommendation feature should ship, given the possibility of network interference across users and servers? Be explicit about your hypothesis, metrics, randomization strategy, power and MDE, analysis plan, and how you would handle common experimentation risks.