# Model Answer: Bad or Misleading Data ## How to Approach Use STARL: - **Situation:** context and decision. - **Task:** what you owned. - **Action:** decision, discovery, correction. - **Result:** impact and recovery. - **Learning:** safeguards implemented. The key is to own the decision and show systemic improvement. ## Example Answer **Situation:** I owned acquisition budget for a consumer product. Our weekly dashboard showed a 15% drop in ROAS for a large campaign cohort. The dashboard showed ROAS falling from 2.0 to 1.7, which suggested those campaigns had become inefficient. **Task:** I needed to protect margin quickly while maintaining growth. **Action:** Based on the dashboard, I paused about 20% of the long-tail campaigns that appeared least efficient and reallocated budget to stronger performers. Two days later, I noticed that total daily revenue had fallen without the expected margin improvement. That made me question whether the original data was accurate. I started a reconciliation with analytics and engineering. We compared our warehouse metrics against source platform reporting and finance billing data. We found that a recent ETL change joined a cost adjustment table incorrectly. Some ad groups had multiple cost rows per day, so cost was duplicated in the warehouse. Revenue was correct, but cost was overstated, which made ROAS look worse than it was. **Impact:** We paused viable campaigns for several days. That caused a short-term revenue dip and required team time to investigate, backfill data, and ramp campaigns back up. I owned the mistake because I acted on a dashboard without triangulating a surprising metric shift against source systems or guardrails. **Recovery:** We restored the campaigns with a controlled ramp, corrected the ETL, backfilled the dashboard, and communicated the issue to stakeholders with a postmortem. **Safeguards:** We implemented several changes: - Data quality checks for duplicate cost rows. - Source-to-warehouse reconciliation for spend and revenue. - Dashboard annotations when metric definitions or pipelines change. - Alerting when cost changes materially but source platform spend does not. - A launch checklist requiring owners for critical metrics. - Decision guardrail: for large budget changes, verify surprising metric movement against at least one independent source. **Learning:** The lesson was not "never trust data." It was that important decisions need metric hygiene, source reconciliation, and guardrails, especially when a metric moves suddenly and the action is hard to unwind. ## Why This Works This answer: - Owns the decision. - Explains why the data looked credible. - Shows how the issue was detected. - Quantifies impact directionally. - Implements systemic safeguards. - Demonstrates mature product judgment. ## Common Mistakes - Blaming the data team. - Saying "I should have checked more" without a process fix. - Hiding the impact. - Describing a trivial mistake. - Not explaining how trust was restored.