One metric that matters, and everything feeding it.
Act as a clear-headed analytics lead who insists decisions come first. The task is designing a KPI tree for my team or product. Context: [the north star or goal, the levers I believe drive it, team responsibilities, current metrics].
1. Break the north star into input metrics two levels deep.
2. Assign an owner and cadence to every input metric.
3. Flag metrics that are easy to game and add a guardrail metric.
4. Define what we stop measuring once this tree exists.
Make it specific to my context and immediately usable. Do not invent current prices, laws, permissions or other missing facts; label what needs checking.
Act as a clear-headed analytics lead who insists decisions come first. The task is writing a requirements brief for a new dashboard. Context: [decisions this dashboard must support, audience, data sources available, refresh needs].
1. List the five questions the dashboard must answer in order.
2. Sketch each panel: question, metric, dimension, time grain.
3. Define what "good" and "bad" look like so color means something.
4. Name what this replaces so we do not add dashboard number twelve.
Make it specific to my context and immediately usable. Do not invent current prices, laws, permissions or other missing facts; label what needs checking.
Act as a clear-headed analytics lead who insists decisions come first. The task is translating a business question into a data specification. Context: [the business question, tables or fields I know exist, time range, comparison needed].
1. Restate the question as a measurable definition with edge cases called out.
2. Specify filters, cohorts and the time window explicitly.
3. List the ambiguities the analyst will hit and my answer to each.
4. Write the sanity check that proves the result is right.
Make it specific to my context and immediately usable. Do not invent current prices, laws, permissions or other missing facts; label what needs checking.
Act as a clear-headed analytics lead who insists decisions come first. The task is writing an experiment readout for stakeholders. Context: [hypothesis, variants, key metrics with results, sample size and duration, surprises].
1. State the decision this experiment informs up front.
2. Present results as decision-relevant deltas with uncertainty in plain words.
3. Address the most likely misreading before someone makes it.
4. End with a clear recommendation and what we do if it is wrong.
Make it specific to my context and immediately usable. Do not invent current prices, laws, permissions or other missing facts; label what needs checking.
Act as a clear-headed analytics lead who insists decisions come first. The task is planning a data quality audit for a critical dataset. Context: [the dataset and its source, known issues, who uses it, decisions depending on it].
1. List the failure modes to check: missing, duplicated, stale, inconsistent.
2. Define a concrete test or query for each failure mode.
3. Rank findings by decision impact, not by count.
4. Write the fix, owner and monitoring rule for the top three.
Make it specific to my context and immediately usable. Do not invent current prices, laws, permissions or other missing facts; label what needs checking.
See retention as behavior, not as one big average.
Act as a clear-headed analytics lead who insists decisions come first. The task is planning a cohort analysis. Context: [the product or service, signup or start event, the behavior that defines retention, time grain].
1. Define the cohort event and the retention event precisely.
2. Choose the cuts that matter: channel, plan, geography or behavior.
3. Write the three hypotheses this analysis should confirm or kill.
4. Specify how results will be presented so trends are obvious.
Make it specific to my context and immediately usable. Do not invent current prices, laws, permissions or other missing facts; label what needs checking.
Act as a clear-headed analytics lead who insists decisions come first. The task is writing a canonical definition for one contested metric. Context: [the metric, the systems involved, current conflicting definitions, who argues about it].
1. Write the definition as an exact formula with inclusions and exclusions.
2. Document the source tables and update cadence.
3. List the known limitations and common misuses.
4. Add worked examples: three real cases and how each is counted.
Make it specific to my context and immediately usable. Do not invent current prices, laws, permissions or other missing facts; label what needs checking.
Act as a clear-headed analytics lead who insists decisions come first. The task is designing a weekly one-page metrics brief for leadership. Context: [the business, top-level goals, available weekly metrics, last quarter's trend].
1. Choose five metrics that together tell the health story.
2. For each: current value, week-over-week delta, one-line so-what.
3. Include one watch item that is not yet a problem.
4. Keep it to one page and define the exact layout.
Make it specific to my context and immediately usable. Do not invent current prices, laws, permissions or other missing facts; label what needs checking.
Act as a clear-headed analytics lead who insists decisions come first. The task is reviewing and fixing a survey before it launches. Context: [my draft questions, who receives it, decisions the results will drive, sample size].
1. Flag leading, double-barreled and unanswerable questions.
2. Rewrite the worst five questions neutrally.
3. Reorder to protect the most important answers from fatigue.
4. Define the response rate below which results are unusable.
Make it specific to my context and immediately usable. Do not invent current prices, laws, permissions or other missing facts; label what needs checking.
Act as a clear-headed analytics lead who insists decisions come first. The task is investigating an unexpected metric change. Context: [the metric, when it changed and by how much, recent releases or campaigns, segments available].
1. List the boring causes first: tracking change, seasonality, mix shift.
2. Order the investigation steps from cheapest to most expensive.
3. Specify the segment cuts that isolate the cause fastest.
4. Define what evidence would confirm each hypothesis.
Make it specific to my context and immediately usable. Do not invent current prices, laws, permissions or other missing facts; label what needs checking.
Pressure-test the forecast before it sets targets.
Act as a clear-headed analytics lead who insists decisions come first. The task is sanity-checking a forecast I was given. Context: [the forecast values, the method if known, historical actuals, known upcoming changes].
1. Compare the implied growth rates against history and call out breaks.
2. List the assumptions the forecast silently depends on.
3. Run three what-if stress cases and show the range.
4. State whether to trust, adjust or rebuild, with reasons.
Make it specific to my context and immediately usable. Do not invent current prices, laws, permissions or other missing facts; label what needs checking.
Make the data land as a narrative, not a spreadsheet.
Act as a clear-headed analytics lead who insists decisions come first. The task is turning analysis findings into a stakeholder story. Context: [the findings, the audience and their priorities, the decision I want, time allotted].
1. Open with the decision or stakes, not the methodology.
2. Structure as situation, complication, resolution using my numbers.
3. Choose the three charts that carry the story and kill the rest.
4. Write the one-sentence takeaway I want repeated afterward.
Make it specific to my context and immediately usable. Do not invent current prices, laws, permissions or other missing facts; label what needs checking.