The prompt
Score 10 of 10, Excellent
Use it when A number went up or down and you need a careful write-up of what the data does and does not show before anyone acts on it.
- {{audience}}
- {{metric_name}}
- {{metric_definition}}
- {{change}}
- {{periods}}
- {{data}}
- {{known_events}}
Full prompt and why it works
Prompt
Help me explain a change in a metric to {{audience}}, using only the data below. The goal is an explanation they can trust, not a story that merely sounds confident. Metric: {{metric_name}}, defined as {{metric_definition}} Change: {{change}} Periods compared: {{periods}} Data: """ {{data}} """ Known events in the period, such as launches, outages, pricing or tracking changes, and holidays: {{known_events}} Work through it in this order, with a heading for each part: 1. What changed: the size of the change in absolute and relative terms, and whether it is outside the normal variation visible in this data. 2. Where it changed: break the change down by the segments in the data, and name the segments that account for most of it. 3. Possible explanations: up to 4, each with the evidence for it in the data, the evidence against it, and your confidence (high, medium or low). 4. What would confirm it: for each explanation, the one extra cut of the data or check that would confirm or rule it out. 5. Summary for {{audience}}: three sentences in plain words. Keep the whole write-up under 450 words. Show each calculation on one line so it can be checked, for example "(1,240 - 1,050) / 1,050 = 18%". Do not present correlation as cause, and do not use data or events that are not given above. If the data includes a tracking or definition change, flag it first, because that alone can explain a change. If the data is not enough to tell the explanations apart, say so plainly instead of picking one.
Variables
- {{audience}}
- Who the explanation is for, for example "the weekly leadership meeting".
- {{metric_name}}
- The metric's name.
- {{metric_definition}}
- How it is calculated, for example "paid orders divided by sessions".
- {{change}}
- What moved, for example "down from 3.1% to 2.6%".
- {{periods}}
- The periods compared, for example "the week of October 6 against the four weeks before".
- {{data}}
- A table or CSV with the metric by period and by segment, such as channel, platform or region.
- {{known_events}}
- Anything that happened in the period. Write "none known" if nothing did.
Why it works
- Who it's for. Saying the explanation must be trusted by a specific audience pushes the model toward evidence and confidence levels instead of a tidy story.
- What to avoid. It forbids presenting correlation as cause and puts tracking changes first, the two most common ways a metric write-up goes wrong.
- If unsure. When the data cannot separate the explanations, it says so and names the check that would.