Prompts for Excel Copilot: Analysis You Can Check

Excel Copilot (the AI assistant in Microsoft Excel) can analyze tables, write formulas, and summarize data — but the quality of what it produces depends almost entirely on how clearly you ask. A vague prompt like "analyze this data" returns a generic summary. A specific prompt that names the columns, defines the question, and sets the output format returns something you can actually check and use. Tools like fenbs let you save the prompt recipes that worked and the approved analysis assumptions behind them, so every spreadsheet session builds on tested patterns rather than starting over.

This page covers prompts for four common Excel Copilot tasks: exploring a table to understand its shape and content, analyzing data to answer a specific business question, explaining or checking a formula, and summarizing results. For how to set up Copilot in Excel and what it can and cannot access, see how to use Copilot in Excel. For prompt patterns that apply across Microsoft Copilot products more broadly, see Microsoft Copilot prompts.

What makes an Excel Copilot prompt effective

Excel Copilot works within your spreadsheet context — it can see the data in your current workbook without you pasting it into the chat. This means your prompts can reference column names and table names directly rather than describing the data. The more specific you are about which columns, which time range, and which question you are asking, the more targeted the output.

Always verify the output. Excel Copilot can write formulas that look correct but contain an error in cell references, a misapplied function, or a filter that excludes rows you intended to include. Check every formula against a small test case before applying it to the full dataset. Check every summary statistic against a manual spot-check on a few rows.

Save the prompt patterns that produce reliable results. A prompt that consistently gives you a well-structured output for a recurring analysis task is worth keeping. After a few uses, the prompt stops being something you type from scratch and becomes a repeatable recipe.

Prompts for exploring a table

Before analyzing data, it is worth understanding what you are working with. These prompts ask Copilot to describe the structure and surface any quality issues before you proceed.

Describe this table for me:
— How many rows and columns does it have?
— What does each column appear to contain? (data type and a few example values)
— Are there any columns with blank or missing values? Which ones, and how many blanks?
— Are there any obvious duplicates in the [column name] column?

Do not analyze the data yet — just describe what is there.

Before I run any analysis on this table, check for common data quality issues:
1. Columns where the format is inconsistent (e.g., dates in different formats, text and numbers mixed)
2. Columns with unexpected outliers (values far outside the typical range for that column)
3. Rows where a required field is blank

List each issue you find. For each one, give a count of the affected rows. Do not fix anything — only describe the issues.

Prompts for data analysis

Analysis prompts should state a specific business question rather than asking Copilot to "analyze" without a target. "What is the average revenue per customer by region?" is a question Copilot can answer precisely. "Tell me about the revenue" is not.

Using the data in [table name or sheet name], answer this question: [specific business question]

Rules:
— Use only the data in this workbook.
— Show the calculation or formula you used for each figure.
— If a figure required an assumption (e.g., which rows to include or exclude), state the assumption explicitly.
— If the data does not allow you to answer the question fully, say what is missing.

Compare [segment A] to [segment B] on [metric]. Use the data in [table name].

Show: the value for each segment, the difference in absolute terms, and the difference as a percentage. Explain in one sentence what each figure means.

State any filter or assumption you applied to isolate each segment.

Comparison prompts often reveal that the two segments are not defined clearly enough in the data. If Copilot cannot isolate one segment from the other, that tells you something useful about the structure of your data before you build anything on top of it.

Prompts for explaining and checking formulas

One of the most reliable uses of Excel Copilot is having it explain a formula in plain language before you copy it to a larger range. A formula that looks correct can contain a reference error that only shows up after you apply it to a thousand rows.

Explain this formula to me in plain English: [paste formula]

Step through each function and what it does. Explain what each cell reference or named range refers to. Note any condition where the formula might return an unexpected result — for example, blank cells, zeros, or negative numbers.

I want to check this formula: [paste formula]

Using these test inputs: [describe a few rows with known values and expected outputs]

Calculate the expected output for each test row and compare it to what the formula produces. For any row where the output is unexpected, explain why.

Prompts for summarizing results

A summary prompt is the last step, not the first. Ask Copilot to summarize after you have checked the analysis, not before. A summary of unverified analysis is a polished presentation of errors.

Summarize the analysis results from this session in a format suitable for sharing with [audience — e.g., the sales team, the finance committee].

Include: the question we answered, the key findings in plain language (no jargon), the assumptions the analysis relied on, and one sentence on what this means for next steps.

Do not include the formulas or raw numbers unless they are essential to understanding the finding. Keep the summary under 200 words.

The instruction to name the assumptions in the summary is important. A finding presented without its assumptions can be acted on in ways the analysis does not support. Including the assumptions gives the reader the information to judge whether the finding applies to their decision.

How to do this with fenbs

1. Create a free account and add a project for your recurring spreadsheet analysis.
2. Save each reliable analysis prompt as a Skill with a short title, the table structure it requires, and any assumptions it makes.
3. Add approved analysis assumptions — segment definitions, inclusion rules, comparison periods — as Decisions in the project.
4. At the start of each analysis session, open the project in fenbs and review the current assumptions before starting. This prevents running an analysis against last quarter's definitions when this quarter's have changed.
5. Connect your AI app over MCP from Integrations if you want an assistant to read the project context — including current assumptions — before you start a session.

fenbs does not connect to Excel or read spreadsheet data. It keeps the prompt patterns and approved assumptions that make your Excel Copilot sessions repeatable and traceable. A well-labeled prompt library also speeds onboarding: a new analyst can find the right starting prompt and see what assumptions apply rather than rewriting everything from scratch. For broader prompt patterns across Microsoft Copilot products, see Microsoft Copilot prompts and copilot-vs-chatgpt for how they compare.