Help me analyze [BUSINESS QUESTION] using [DATASET DETAILS] for [AUDIENCE]. The analysis period is [DATE RANGE], and these are the metric definitions or business rules: [METRIC DEFINITIONS]. I will use [ANALYSIS TOOL]. Work as a senior analytics partner. First, restate the decision this analysis should support and identify the primary metric, comparison baseline, segmentation, and grain of the data. Distinguish a count, rate, average, median, and cumulative measure where relevant. Do not treat correlation as causation. Then produce: 1. A short analysis plan with the exact cuts, cohorts, exclusions, and comparison periods to use. 2. The SQL query, spreadsheet formulas, or tool-specific steps needed. State assumptions about table names, date fields, joins, time zones, nulls, duplicates, and denominator handling. For SQL, use CTEs and qualify every column after joins. For spreadsheets, give formulas with cell or column references I can adapt. 3. A plain-English findings template with: headline, evidence, likely drivers, limitations, and recommended next action. 4. A validation checklist covering row counts, join multiplication, missing values, outliers, metric reconciliation, and whether percentages use the right denominator. If results are not provided, do not invent findings; tell me what the output would reveal. Flag any claim that cannot be supported by the available fields. Before answering, self-check that the query/formulas match the stated metric definitions and that every suggested comparison has a valid baseline. Ask up to 3 clarifying questions only if a required input is missing.
Fill in
| Placeholder | What to enter | Example |
|---|---|---|
| [BUSINESS QUESTION] | State the decision you need to make and the question the analysis must answer. | Did the August onboarding change improve first-week activation for self-serve customers? |
| [DATASET DETAILS] | List relevant tables or sheets, key fields, their grain, and any known data-quality issues. | BigQuery tables users, signup_events, and product_events; one row per event in product_events; user_id joins all tables. |
| [AUDIENCE] | Name who will read the analysis and what they need to decide. | The product manager deciding whether to keep the new onboarding flow. |
| [DATE RANGE] | Give the reporting period and any comparison period. | August 1-31, compared with July 1-31. |
| [METRIC DEFINITIONS] | Define each important metric, inclusion rule, exclusion, and target. | Activation means completing a project within 7 calendar days of signup; exclude internal accounts and users with missing signup dates. |
| [ANALYSIS TOOL] | Name the tool you are using, such as BigQuery, Snowflake, Excel, or Google Sheets. | BigQuery SQL. |
How to use
- Paste the prompt and replace every bracketed input with your question, schema details, and metric rules.
- Run the query or formulas on a small date range first and compare totals with a trusted dashboard.
- Check every join, date boundary, and rate denominator before using the output in a decision.
- Follow up with: “Here are the results. Write the findings section, separating observed facts from possible explanations.”
Variations
Dashboard audit
Use this when a recurring dashboard has numbers stakeholders do not trust.
Audit this dashboard metric: [METRIC]. Its current definition is [CURRENT DEFINITION], sourced from [TABLES OR SHEETS], and the dashboard shows [UNEXPECTED RESULT]. Create a diagnostic plan ordered from most likely to least likely cause. Include checks for date filters, time zones, late-arriving records, duplicated joins, changed event instrumentation, null treatment, and denominator logic. Give SQL or spreadsheet checks for each issue. Do not diagnose a cause without evidence; state what result would confirm or rule out each hypothesis.
Experiment analysis
Use this when you need to evaluate an A/B test without overstating its result.
Analyze an experiment with this setup: [EXPERIMENT DETAILS], [PRIMARY METRIC], [GUARDRAIL METRICS], and [RESULTS OR DATA SCHEMA]. Produce a decision memo with assignment and exposure checks, sample-ratio mismatch checks, the correct unit of analysis, effect size, confidence interval or uncertainty explanation, and segment findings labeled exploratory. Note any novelty, contamination, multiple-comparison, or duration risks. Recommend ship, iterate, or hold only from the evidence provided. Do not invent statistical outputs or causal conclusions.
Executive report
Use this when leaders need an analytics update rather than raw queries.
Turn these verified analytics results into a one-page report for [AUDIENCE]: [RESULTS]. The business goal is [GOAL], and the comparison baseline is [BASELINE]. Write a title, three key takeaways, a compact metric table, drivers supported by the data, risks or caveats, and two proposed actions with an owner and measurement plan. Put absolute values beside percentage changes where available. Separate facts, interpretations, and open questions. Flag any conclusion that the supplied results do not establish.
Tips
- Write the metric definition before asking for a query; “active users” can mean a login, a meaningful event, or a billing event.
- State the table grain for every source, because many inflated metrics come from joining a user-level table to an event-level table.
- For rates, provide both the numerator and denominator eligibility rules, especially when users can enter a cohort late.
- Ask for a reconciliation check against a known total before sharing a new analysis.
FAQ
Can AI write SQL from a screenshot of a dashboard?
It can help infer a query, but it cannot verify the semantic layer, filters, or source tables from a chart alone. Provide the metric definition and schema before trusting the SQL.
Should I ask for SQL or an analysis plan first?
Ask for both. A plan exposes wrong cohorts and comparisons before you spend time debugging a query.
How do I avoid misleading percentage changes?
Require the model to show base counts, absolute changes, and the denominator for every percentage.