## Demonstrate Customer-Centric Judgment About AI Describe one real situation in which you had to decide whether or how to use an AI capability in a customer-facing product or workflow. Explain the customer need, the evidence you gathered, the options considered, and how you balanced usefulness with accuracy, privacy, transparency, or another material risk. The example may end with launching, narrowing, delaying, or rejecting the AI approach. The quality of judgment matters more than claiming that AI was always the right solution. ### Constraints & Assumptions - Identify the affected customer group rather than referring to “users” as one uniform population. - Separate observed customer evidence from internal enthusiasm for a technology. - Explain which failures were tolerable, which required human review, and which would block release. - Do not claim a result without describing how it was measured or observed. - Be precise about your own decision and the work owned by others. ### Clarifying Questions to Ask - Does the interviewer prefer a product launch, an internal tool with customer impact, or a decision not to deploy? - Which dimension should receive the most depth: customer discovery, technical evaluation, or risk management? - May the example discuss confidential details at an abstracted level? ### What a Strong Answer Covers - A specific customer problem and the evidence showing it mattered. - A comparison between an AI approach and at least one non-AI alternative. - Evaluation criteria tied to customer outcomes, including important failure slices. - Privacy, disclosure, human-control, and fallback decisions appropriate to the example. - The candidate's role in resolving disagreement or changing the plan. - A measured outcome, limitation, and lesson that would affect the next decision. ```hint Start with the customer decision, not the model Explain what customers needed and what failure would harm them before discussing an AI technique or tool. ``` ### Follow-up Questions 1. What evidence would have caused you to choose the non-AI alternative? 2. Which customer segment experienced the greatest benefit or risk? 3. How did you evaluate plausible-looking but incorrect output? 4. What control did customers have over the AI-assisted result? 5. If the outcome metric improved, which guardrail could still have blocked the launch? 6. What would you change about the decision process today?