You run an A/B test on a growth surface in a business software product, and the treatment shows a positive lift on the primary conversion metric. However, the confidence interval is wide, so the estimate ranges from a small win to a much larger one, and stakeholders are pushing to call the experiment successful.
What would you do if an experiment had a positive result but the confidence interval was wide? How would you decide whether the result is actionable enough to ship, extend, or rerun?