原文核对：2026-10-13 · 中文正文 · 英文提示词与演示
场景定位
结合指标定义、导出、分群、上线背景和讨论，形成有来源的根因简报，区分确认驱动与假设。
- 难度
- 中级
- 时间跨度
- 1 小时
适合用于
- 调查异常 KPI 变化的产品、增长或运营团队。
- 需分群、队列、渠道、地区或产品切分的问题。
- 要求确认因素与合理假设分开的审阅。
相关工具
检查导出、计算切分并绘图。
读取指标定义、仪表盘、上线和源文件。
Google Drive ↗检查相关人员背景与近期变化。
Slack ↗整理带来源和局限的简报。
起步提示词
英文原文 · 可复制后按实际材料修改
Explain why [KPI] changed during [time window].
Use the KPI dashboard, metric definition, source exports, segment cuts, launch or campaign context, and stakeholder threads I provide. Break down the movement by relevant segment, cohort, channel, geography, and product surface.
Return a root-cause brief with:
- charts and material changes
- confirmed drivers
- hypotheses to investigate
- caveats and data-quality issues
- source links
- open questions
- recommended actions
Do not treat correlation as proof, and do not change the source data.先定义指标
根因分析先确定稳定定义、比较窗口和权威数据，再提供仪表盘、导出及上线或活动背景。
- 说明 KPI、期间、预期方向和支持的决策。
- 提供定义、仪表盘、导出、切分和背景。
- 检查数据质量，提出有用拆解。
- 分别审阅确认驱动和假设。
- 改仪表盘或流程前，与指标负责人验证行动建议。
用图表检查变化，但不能把分群相关性当作因果证明；保留来源、局限和开放问题。
质疑根因
提交管理层前查找反证和替代解释。
官方提示词
英文原文 · 可复制后按实际材料修改
Challenge the root-cause brief.
For each confirmed driver or hypothesis, list:
- supporting source and calculation
- counterevidence or alternative explanation
- segment or cohort where it does not hold
- data-quality limitation
- next check that would increase confidence
Revise the recommendation only when the evidence supports it, and keep all changes visible.展开英文原文对照
Give ChatGPT KPI dashboards, metric definitions, exports, segment cuts, launch context, and stakeholder threads, then ask it to separate confirmed drivers from hypotheses in a source-backed root-cause brief.
Define the metric before explaining movement
Root-cause work starts with a stable definition, comparison window, and source-of-truth data. Give ChatGPT the KPI definition, dashboard, exports, and context around launches or campaigns before asking it to explain the change.
- State the KPI, comparison period, expected direction, and decision the analysis should support.
- Attach the dashboard, metric definition, exports, segment cuts, and relevant context.
- Ask ChatGPT to inspect data quality and propose the most useful breakdowns.
- Run the starter prompt and review confirmed drivers separately from hypotheses.
- Validate the recommended actions with the metric owner before changing a dashboard or process.
Use charts to make the movement inspectable, but do not treat a segment correlation as proof of cause. Keep source links, caveats, and open questions with the brief.
Challenge the root cause
Ask ChatGPT to look for counterevidence and alternative explanations before the brief goes to leadership.
官方提示词
英文原文 · 可复制后按实际材料修改
Challenge the root-cause brief.
For each confirmed driver or hypothesis, list:
- supporting source and calculation
- counterevidence or alternative explanation
- segment or cohort where it does not hold
- data-quality limitation
- next check that would increase confidence
Revise the recommendation only when the evidence supports it, and keep all changes visible.