You work on a B2B communication platform and want to test a new shared-inbox routing experience that automatically suggests the best teammate to answer an incoming call or conversation. The team believes the change will improve first-response speed and answered-call rate by reducing internal handoff friction. However, accounts are collaborative: one agent’s treatment can change workload and outcomes for other agents on the same account. You need an experiment design that measures impact while handling likely interference.
How would you design this experiment so that the estimated treatment effect is credible despite cross-user interference within accounts? Explain how you would choose randomization, define success and guardrails, size the test, analyze results, and decide whether to ship.