Claude for Excel: What It Does Well, Where It Stops, and What to Use for Shared Numbers
For one person working through one workbook, Claude in Excel is a strong assistant. The limits show up when the same number has to match across people, across months and across a client deck.
Claude for Excel is useful for one person in one workbook: explaining a formula, drafting a summary, finding the odd outlier. Where it stops is the moment a number has to be the same for everyone who asks, next month and in the client deck. That is a question about definitions and sharing, not about how clever the assistant is.
What it does well
- Working through a workbook you already have. Ask what a formula does, where a total comes from, or what changed between two tabs.
- Drafting. A first version of a summary or a note on a variance is quicker to edit than to write from nothing.
- One-off analysis. A question you will ask once, of data that will not change, is well suited to a chat.
Where it stops
Plan and usage. The add-in needs a paid Claude plan, and Excel work counts against the same usage limits as the rest of the account. A long modelling session can use an allowance faster than a short chat. If you are planning for a team, that cost sits on the plan, not on the add-in.
Chat history. A conversation is not a saved report. Anthropic’s documentation notes that after recent Microsoft 365 updates on Windows, some users have lost chat history and uploaded skills in the add-in, and that Anthropic is working on it with Microsoft. Check the current status on Anthropic’s site before you depend on a conversation to be there next week.
Definitions. The assistant reads your workbook and your question. It does not know that your team calls a sale "booked" when the invoice goes out, or that one client’s refund belongs to the month of the original order. Those choices decide the number, and they live in people’s heads unless they are written down.
A quick test you can run today
Pick one total you have already checked by hand, such as last month’s net sales. Ask the add-in for it, then ask again with different wording, and once more the next day. The figures should match your hand check and each other. Where they do not, the difference is usually a definition: a date field, a refund treated two ways, or a column name the assistant had to guess.
Illustrative case: a net sales figure of $84,200 came back as $84,200 on the first ask, $86,050 on a reworded ask, and $84,200 the next day. The gap was $1,850 of refunds that one phrasing included and the other did not. Nothing in the workbook said which was right.
When one chat is not enough
If the number goes to a client, a board or a colleague who will not ask the assistant, it needs a home outside the chat. Three options, in rising order of how much they fix:
- Write the definition down. One sentence per metric, kept beside the workbook.
- Save the report. A workbook with fixed formulas gives the same answer each time, as long as the data is the same.
- Use a tool that holds the definition. A governed layer keeps each metric defined once and answers every question from that definition.
For the first two, the pipeline analysis guide shows how to keep definitions consistent in a spreadsheet. For the third, Quiriz holds each definition in Company Context, so the same question gives the same figure whether it is asked in Excel, in Sheets or in a shared chat. The Quiriz Excel add-in reads the same definitions.
When Claude for Excel is enough
It is enough when the number stays in one workbook, one person checks it, and nobody else needs to reproduce it. It is not enough when the figure is reported, compared across months or handed to someone who cannot ask follow-up questions. Most reporting is the second kind, which is why the definition matters more than the assistant.
The same number, asked in any tool
Try a question on sample company data with no signup, then connect your data and see how definitions carry across tools.
Try Quiriz free →