Tell me about a time you were wrong about something at work.
Being wrong and recognizing it quickly is a signal of strong analytical judgment — not weakness. The best answers name a specific, consequential mistake, explain how you discovered you were wrong, describe what you did about it, and connect it to a habit or process you changed as a result.
How to think about it
What the interviewer is actually testing
This question probes intellectual honesty, self-awareness, and growth mindset — three traits that matter enormously on data teams where the cost of a confident-but-wrong analysis can be significant. Interviewers also want to see that you update on evidence rather than defending a position out of ego.
How to structure a strong answer
Pick a substantive mistake. A typo in a dashboard is too small. A flawed assumption that drove a bad business decision, an overconfident model evaluation, a data pipeline that silently introduced bias — these are the right scale. The stakes signal that you’ve shipped real work.
Be clear about what you believed and why. The interviewer wants to understand your reasoning process, not just that you made an error. Explaining your original logic (even though it was wrong) shows how you think.
Describe how you found out. Did a colleague catch it? Did downstream metrics reveal it? Did a stakeholder point it out? How you discovered the error matters — ideally you have some examples of catching your own mistakes, which shows rigor.
State what you did immediately. Did you communicate proactively? Fix and re-deliver the analysis? Quantify the impact of the error? Responsible behavior after a mistake is often what the interviewer remembers most.
Close with the lasting change. Not “I was more careful after that.” Something concrete: “I now always sanity-check cohort distributions before aggregating” or “I added a data validation step before any analysis goes to stakeholders.”
Skeleton example: “I reported that a new onboarding feature increased 30-day retention by 12%. Six days later I realized I had included users who saw the feature rollout page but did not interact with the feature itself in the treatment group. I went back to the PM immediately, corrected the segmentation, and the real lift was 4% — still positive, but the timeline changed. I built a template for experiment segmentation review that we now use team-wide.”