ANALYTICS & REPORTING

Data dump to three decisions

Use when you have a pile of numbers and need to get to what should actually change.

data-dump-to-three-decisions.md

Download .md

You are turning a data dump into decisions. Nothing else.

Data:
{{DATA_DUMP}}

Business context: {{BUSINESS_CONTEXT}}
What we can actually change: {{LEVERS_AVAILABLE}}
Constraints: {{CONSTRAINTS}}

Output exactly three decisions. For each:
- Decision, stated as an action someone can take on Monday
- The specific numbers that led you here, quoted from the data
- What we are betting will happen, as a measurable prediction with a date
- What would tell us we were wrong, and by when
- Effort (hours or days) and who does it
- Confidence: High, Medium, or Low, with the reason

Then two short lists:
- Things that look interesting in this data but should not change anything yet, and what would need to be true for them to matter
- Numbers in this dump you do not trust, and why

Rules:
- Three decisions, not five, not "consider also".
- Order by expected impact divided by effort, and show that ordering reasoning in one line.
- Do not propose anything outside the levers I listed.
- If the data does not support three decisions, give fewer and say what data would unlock the rest. Do not pad.

Fill in before running

Replace each placeholder with your own detail. The more specific you are, the less the model invents.

{{DATA_DUMP}}
{{BUSINESS_CONTEXT}}
{{LEVERS_AVAILABLE}}
{{CONSTRAINTS}}

Getting a better result

Listing the levers honestly is what stops it recommending things you cannot do.
The falsification line is the point of the exercise - keep it and check it on the date.
Run it on the same dump twice a quarter apart to see whether the decisions changed.

Questions about this prompt

When do I use this rather than writing an analysis?

When the analysis already exists and nothing has changed because of it. This is deliberately narrow: three decisions someone could act on Monday, ordered by impact over effort. If you still need to understand what the data says, do that first, because this prompt will not explore, it will commit.

What do I need in front of me?

The data, the business context, the constraints, and an honest list of the levers you can actually pull. The levers field is the whole trick. Without it you get recommendations needing a team you do not have. With it, every decision is something your organisation could start this week.

What comes back?

Three decisions, each with the numbers behind it quoted from the dump, a measurable prediction with a date, what would prove you wrong and by when, effort, owner and confidence. Then a not yet list and a numbers I do not trust list. The falsification line is the point of the exercise.

What is the mistake that costs me here?

Never going back on the date you set. A prediction with a falsification condition is only worth writing if somebody checks it, and re-running the same dump a quarter later to see whether the decisions changed is the cheap version of that. Do not let it pad to five decisions either.