1. Interpreting results and making the call

The interviewer has walked you through the whole test, and now asks the last question: "The primary metric is up, one guardrail is down, and one segment went the other way. Do you ship?" Interpreting results is the step that answers it. You take the confidence intervals the analysis produced for the primary metric, the guardrails, and the planned segments, read each one against a threshold the team set before launch, and turn them into one decision you can defend out loud. It starts only after the test passes its trust checks: if the arm counts are off or logging is broken, there is nothing to interpret yet, so fix the test and rerun it.

This is where almost every A/B case question ends, usually in the last 5 to 10 minutes of the round. For a mid-level role, the bar is mapping an interval to a clear call and refusing to read "not significant" as "no effect". For a senior role, interviewers also listen for how you check a guardrail against a tolerated margin, whether you trust a segment result, and how you plan for an effect that may fade after launch.

The running example is a food-delivery app testing a new restaurant ranking on its home feed, and the tutorial starts from intervals that are already computed, the work covered in Hypothesis Testing and Confidence Intervals. It also assumes you know what primary and guardrail metrics are.