AI Prompts for Analysing Data and Fixing Spreadsheets

Before any analysis, audit this file: missing values by column, duplicates, inconsistent formats, outliers, and anything suggesting the data was collected differently over time. Tell me what I cannot trust.

Why it works: Almost every wrong conclusion comes from skipping this step. Run it before every analysis.

Clean this dataset: standardise formats, handle missing values, fix inconsistent categories. For every change, log what you changed and how many rows were affected. Do not drop rows without telling me.

Why it works: Silent row-dropping is the most common way an analysis quietly stops representing reality.

Analyse this file. Do not summarise it. Tell me the three things a manager would be wrong about if they only read the totals, with the chart that shows each.

Why it works: Aggregates conceal segment differences, and segment differences are usually the finding.

Explain what this spreadsheet formula does, step by step from the innermost function outward, and tell me what happens when it hits an empty cell or a text value.

[paste formula]

Why it works: The edge case behaviour is where inherited spreadsheets break silently.

Schema:
[paste table definitions]

Write a query that [requirement]. Explain the join logic. Then tell me how the result would be wrong if [likely data issue] exists.

Why it works: Asking how it could be wrong catches the duplicate-row join problem before it reaches a report.

Here is my conclusion and the data behind it: [paste]. Argue against it. What alternative explanation fits the same data? What would I need to check to rule it out?

Why it works: Cheap peer review. This prompt has saved more presentations than any other on this page.

I want to show [message] to [audience] using this data. Suggest three chart types, explain what each emphasises and hides, and recommend one. Then generate it.

Why it works: Charts are arguments. Choosing deliberately rather than defaulting to a bar chart changes what people take away.

Here are my assumptions and calculation: [paste]. Which single assumption, if wrong by 20 percent, changes the conclusion most? Show the sensitivity.

Why it works: Sensitivity analysis is what separates a forecast from a guess with decimal places.

Write a one-page brief for [audience] from this analysis. Structure: the finding in one sentence, three supporting points with numbers, what we should do, what we still do not know. No methodology section.

Why it works: Ending with what you do not know builds credibility and pre-empts the question you would otherwise be asked.