You run an A/B test on a subscription product and see a 5% increase in subscription upgrades in treatment, but overall sample downloads fall by 2%. The team is encouraged by the monetization lift, but you are concerned the experiment may be improving upgrades at the expense of core product usage.
How would you decide whether to ship this change? Walk through how you would define the primary metric and guardrails, assess whether the observed trade-off is statistically and practically meaningful, and make a launch recommendation.