Some workbooks are too important to guess at, such as a financial model, a forecast or a pricing calculator someone else built.
Claude for Excel is an add-in that brings Claude into your workbook to explain it, update assumptions, debug errors and build or edit models. It cites the cells behind its answers and highlights the changes it makes, so you can check every step.
Where to find Claude in Excel
Claude is an add-in, so you install it once:
- On your own: install Claude by Anthropic for Excel from the Microsoft Marketplace, or open Excel, go to Home, select Add-ins and search for Claude.
- At work: your IT team may need to deploy the add-in and assign it to you first.
Once installed, Claude appears on the right of the ribbon. Open it and sign in with your Claude account.
What your plan includes
- Paid Claude plans: Claude for Excel has been generally available since May 2026 on Pro, Max, Team and Enterprise plans. It is not included in the Free plan.
- Where it runs: Excel on the web, Excel for Windows with a Microsoft 365 subscription and Excel for Mac. The iPad and Android apps are not supported.
- Connectors and skills: the connectors and skills you set up in Claude are also available in Excel, so Claude can bring in context from other tools.
- Team and Enterprise: admins control deployment and which connectors are available.
When Claude for Excel earns its place
- A financial model you need to understand, including how one assumption flows through to the result.
- Scenario work, such as changing growth or pricing assumptions and seeing the impact.
- Debugging broken references, circular logic or errors you cannot trace.
- Analysis you must defend, where cell-level citations let you show your work.
How to analyze a workbook with Claude, step by step
The examples below use a public job market dataset with roles, salaries, skills and locations. The same prompts work on any dataset, from a sales export to survey results.
Step 1: Check how Claude reads your data. Open the add-in and ask:
List every column and explain what each appears to represent. List five random samples of the data for each column.
If the headers, samples or meanings look wrong, fix them now, before Claude analyzes anything.
Step 2: Ask what the data can answer.
What are the five most interesting questions this dataset is sufficient to answer?
Step 3: Drill into a decision-relevant question.
Which job titles offer the highest salary ranges, after normalizing by required experience?
Claude explains its reasoning with cell-level citations. Click back to the cells it referenced and check that the assumptions match what you would tell a stakeholder.
Step 4: Turn the finding into a chart. Claude can create and edit native Excel charts, so the result stays inside the file:
Create a chart showing the top skills associated with high-paying roles.
Step 5: Review the changes. Claude highlights what it edited. Check every highlighted change before you save or share the workbook.
What should you watch out for?
- Some Excel features are out of reach, such as macros and VBA.
- Final numbers need a human check. Anthropic advises against using it for audit-critical calculations or client deliverables without review.
- Untrusted spreadsheets carry a risk. Hidden instructions in cells or formulas can try to steer Claude, so use it on trusted internal files.
A CODO prompt to start with
Once you trust how the AI reads your data, use this prompt, built with our CODO framework, to turn the analysis into findings your audience needs. Replace the bracketed parts. Your own definitions matter most, because the AI otherwise picks its own.
Character: You are a senior data analyst who specializes in [sales data] and knows how to frame findings for [a leadership team]. Objective: I am the [Head of Sales] at [company] in [industry]. Analyze the [attached / open] dataset, surface the three most important patterns and flag anything concerning, so [my leadership team] can decide [where to focus next quarter]. Do: - Before you analyze, list each column, what it appears to represent and any data quality issues you notice. - Use my definitions: [a high-value customer has annual revenue above $50,000]. - Lead with the "so what", not the method. - Show the calculation or the cells behind every number you report. - Suggest one chart that would make the main finding clear. Don't: - Don't change or delete my original data. Put any new analysis in a separate sheet or file. - Don't present estimates or assumptions as facts. Label them. Output: Three key findings of one paragraph each, followed by a 3 to 5 sentence summary for [my leadership team] that I can paste into an email or slide.
Want to build something reusable? Build a data visualization generator with Claude Artifacts shows how to turn CSV files into filterable charts.
What will your IT and legal teams ask?
- Is our data used for training? Anthropic does not train its models on Team and Enterprise data by default.
- How is it deployed? Admins can deploy the add-in from the Microsoft Marketplace and assign it to specific people.
- What can Claude see? The workbook you have open, plus any connectors your admin allows.
- What about prompt injection? Anthropic warns about hidden instructions in untrusted files, so set a policy for which files people may open with Claude.
Claude for Excel or another tool?
Claude for Excel is the strongest choice for deep work on models, especially when you need to trace every number back to its cell.
If your company has Microsoft 365 Copilot licenses, Copilot in Excel is already in your workbook. If ChatGPT is your main assistant, ChatGPT for Excel offers a similar add-in. For quick exploration and interactive charts outside Excel, try Claude. Our overview of AI for data analysis compares all eight options.
Key takeaways
- Install the add-in once, or ask IT to deploy it.
- Check how Claude reads the workbook before you ask for analysis.
- Use the cell citations to verify every important number.
- Review every highlighted change before you save.
- Stick to trusted files to avoid hidden instructions.
