You work on a collaborative work-management product and want to evaluate a change to a shared surface, such as task notifications, inbox ranking, or workload recommendations. A standard user-level A/B test may be hard to run cleanly because users interact inside teams and projects, exposure can spill across collaborators, or the rollout may need to happen by workspace or over time.
How would you decide whether to use a quasi-experiment instead of a randomized test in this situation? Walk through the conditions that would push you away from randomization, what quasi-experimental design you would choose, and how you would judge whether the evidence is strong enough to act on.