Stop Pivoting in Excel: 12 Data Analysis Prompts That Turn Raw Sheets into Business Decisions

Stop Pivoting in Excel: 12 Data Analysis Prompts That Turn Raw Sheets into Business Decisions

You export a CSV, open it in Excel, and freeze. The client wants a forecast, the ops lead wants to know why churn spiked, and you're still trying to remember how to write a XLOOKUP that doesn't break on the next refresh. Most analysts don't fail because they can't count. They fail because translating a business question into a formula, a pivot, or a SQL query is slow, and the clock is always ticking.

This collection is different. These are prompts I actually run against raw tables — messy exports, duplicate rows, mixed date formats — and each one produces something a stakeholder can act on: a trend, a forecast, a ranked list, a fix. No fluff, no "as an AI language model." Just the prompt, why it works, and a real example with the output you should expect.

A quick note on tooling: these work in ChatGPT, Claude, Gemini, or any agent that can read tabular data. For anything touching production data, keep PII out of the prompt and paste only the columns you need. The prompts below assume you describe your schema in plain English — that's the whole trick. AI doesn't need your real column names; it needs the shape of the data.

1. The Schema Translator (Start Here, Always)

Before any analysis, dump the structure. This prompt builds a mental model of your table so every later prompt is accurate.

Prompt:

You are a data analyst. I'm giving you the column names and 3 sample rows of a table. For each column, tell me: (1) likely data type, (2) what business metric it probably represents, (3) common data-quality problems to check (nulls, duplicates, mixed formats). Output a Markdown table. Then list the 5 analysis questions this table can actually answer.

Example:

This one step prevents the classic failure where the AI invents a "profit" column you never had.

2. Clean the Mess (Nulls, Duplicates, Mixed Dates)

Prompt:

Here is my table. Identify every data-quality issue. For each, give me the exact Excel formula or Google Sheets formula to detect it, and a second formula to fix or flag it. Prioritize issues that would distort a revenue or churn calculation.

Real output for a sales export:
- Duplicate detection: =COUNTIF(A:A, A2)>1
- Null check: =ISBLANK(C2)
- Date normalization: =DATEVALUE(TEXT(B2,"YYYY-MM-DD"))

One caveat worth knowing: TEXT with a fixed format string can silently misparse ambiguous dates like 03/04/2026. For those, force the locale explicitly or clean at export time. This is the kind of trap that turns a "clean" dataset into a wrong forecast.

3. The Pivot, Explained Before You Build It

Prompt:

I want to see monthly revenue by product category. Before giving me the pivot, tell me which fields go in Rows, Columns, and Values, and which aggregation is correct (SUM vs COUNT vs AVERAGE) and why. Then give me the Google Sheets QUERY formula as an alternative.

Sheets alternative:

=QUERY(A1:E1000, "select C, sum(D) where B is not null group by C pivot month(A) label sum(D) 'Revenue'", 1)

This teaches the reasoning, so next time you don't need the prompt.

4. Trend Detection with a Sanity Check

Prompt:

Using this monthly series, compute the month-over-month percent change. Flag any month where the change exceeds ±20% and suggest 2 plausible business explanations. Do NOT claim causation — list what data we'd need to confirm each explanation.

That last sentence matters. AI will happily tell you a marketing campaign caused a spike. This prompt forces it to separate correlation from cause — the same discipline a good analyst applies.

5. Forecast with Confidence Bands (Not Just a Line)

Prompt:

Forecast the next 3 months of this series using a simple method appropriate for a small dataset. Show the point forecast AND a low/high range. State your method and its assumptions in one sentence. If the series has fewer than 12 points, say so and lower your confidence.

A single forecast number is a lie of precision. A range is a decision tool. For real forecasting work, exponential smoothing (FORECAST.ETS in Excel) is a legitimate built-in — it exists, it handles seasonality, and it's a better default than a straight trendline.

6. The SQL Translation

Prompt:

Convert this business question into a SQL query for PostgreSQL: "Top 10 customers by total revenue in the last 90 days, excluding refunds." Assume tables orders(id, customer_id, amount, created_at, status). Explain each clause briefly.

Output:

SELECT customer_id, SUM(amount) AS revenue
FROM orders
WHERE created_at >= NOW() - INTERVAL '90 days'
  AND status <> 'refunded'
GROUP BY customer_id
ORDER BY revenue DESC
LIMIT 10;

INTERVAL '90 days' is standard PostgreSQL syntax and portable to most warehouses. Run it, don't trust it blindly — check that status values match what you assumed.

7. Cohort / Retention Analysis

Prompt:

Build a monthly cohort retention table from this signup-and-activity data. Rows = signup month, columns = months since signup, values = % of cohort still active. Explain how you defined "active" and flag any definition that could inflate retention.

Retention definitions are where analysts quietly lie to themselves. If "active" means "logged in once," your retention looks great and means nothing. The prompt makes the definition explicit and auditable.

8. Anomaly Hunt

Prompt:

Scan this dataset for anomalies: values more than 3 standard deviations from the mean, sudden level shifts, and impossible values (negative prices, future dates). For each, state whether it's likely a data error or a real business event, and what to check next.

This catches the duplicate-row bug that doubles your revenue before it reaches a board deck.

9. The Executive Summary

Prompt:

Turn these findings into a 5-bullet executive summary. Each bullet: one number, one implication, one recommended action. No jargon. Assume the reader has 30 seconds.

Example bullet: Revenue fell 8% in Q3, driven mostly by the SMB segment; recommend auditing the SMB onboarding flow before the Q4 push.

10. What-If Scenario Modeling

Prompt:

Model three scenarios for next quarter: base (current trend continues), optimistic (churn drops 2 points), pessimistic (churn rises 2 points). Show the revenue impact of each in a table and state which assumption drives the biggest swing.

Sensitivity analysis in one prompt. The "biggest swing" line is the one your CFO actually wants.

11. The Formula Debugger

Prompt:

This Excel formula returns #N/A: =VLOOKUP(A2, Sheet2!A:C, 3, FALSE). List the 5 most common causes in order of likelihood, and give me the fix for each. Then rewrite it using XLOOKUP.

Rewrite:

=XLOOKUP(A2, Sheet2!A:A, Sheet2!C:C, "not found")

XLOOKUP exists in Excel 365 and Google Sheets; it removes the column-count fragility of VLOOKUP entirely.

12. The Data Dictionary for Handoff

Prompt:

Write a data dictionary for this table: column name, definition, type, example value, and known caveats. Format as Markdown. This is for a teammate who has never seen the data.

The most underrated prompt. A clean data dictionary saves hours of Slack threads and is the difference between a dataset that scales and one that dies with its creator.

Putting It Together

Here's the honest workflow: run prompt 1 once per dataset, prompts 2–3 whenever the data is dirty, and 4–10 when you actually need insight. Prompts 11–12 are maintenance — the boring work that makes everything else repeatable.

A few rules I've learned the hard way. Never paste raw customer PII into a prompt; anonymize first. Always re-run the numbers yourself for anything going into a decision — AI is a fast first draft, not a source of truth. And keep a note of which prompts produced which outputs, because next quarter someone will ask "where did this number come from."

The tools are not the bottleneck anymore. The bottleneck is asking a precise question. These twelve prompts are really twelve ways of asking better questions — and that skill compounds whether you're in Excel, Sheets, or writing SQL by hand.

If you want to go deeper on turning raw data into decisions, the asibiont.com blog has more hands-on guides, and the platform's AI-driven courses let you practice these patterns on real scenarios instead of toy datasets.