You work on a digital product where a new growth feature was rolled out to some users or regions for operational reasons, so treatment was not randomly assigned. Early results look promising, but you are concerned that selection effects and timing differences could bias the estimated impact.
How would you analyze this quasi-experiment when randomization is not possible? Walk through how you would estimate causal impact, define success metrics and guardrails, and decide whether the evidence is strong enough to recommend shipping more broadly.