Start Claude for Excel on a quarter you have already closed.

Claude for Excel is generally available on the Pro, Max, Team and Enterprise plans, and its use is subject to the same usage limits as the Claude account it signs in with. For a finance team handling LP data, start on the firm's Team or Enterprise plan, so the add-in sits under your organization's settings rather than personal accounts.

An admin deploys it through the Microsoft 365 Admin Center: allow Office Store access, open Settings, Integrated apps, Add-ins, search for Claude for Microsoft 365, and assign it to specific users or groups. Assign it to the finance pilot group first. Where Office Store access is switched off, the documentation gives a custom manifest instead.

If your firm routes AI traffic through Amazon Bedrock, Google Cloud Vertex AI, Azure AI Foundry or an LLM gateway, the add-in can be deployed without individual Claude accounts, keeping prompts and responses within your trust boundary. One trade-off: working across Excel, Word, PowerPoint and Outlook is not supported on that path.

Start with the data handling section. Inputs and outputs are deleted from Anthropic's backend within 30 days, with stated exceptions, and chat history is stored locally in the user's browser. The add-in does not inherit custom data retention settings your organization has set. Its activity is not included in Enterprise audit logs, although for Enterprise organizations with the Compliance API enabled, add-in sessions are included in that API, in public beta.

Then the prompt injection warning. Anthropic says to use the add-in only with trusted spreadsheets, because downloaded templates, vendor files and data imports can carry hidden instructions that try to extract data or modify records. When Claude proposes a risky operation, it asks you to confirm first. Write down which files may be opened with the add-in, and put that list in your AI usage policy.

Start with the reconciliation, and start by reading. Take a quarter that has been closed and tied out, make a copy, and ask Claude to explain the file: how a capital account balance was derived, where a break between the administrator's figures and your workpapers comes from, which cells drive the management fee line. Answers carry cell-level citations you can click, so a reviewer checks the reasoning against the cell itself.

After the numbers are tied out. Anthropic says Claude for Excel is not recommended for final client deliverables without human review, or for audit-critical calculations without verification, and an LP statement is the kind of file both limits describe. The useful work sits around the figures: explaining what moved in a capital account, checking that a table in the letter agrees with the workbook, and drafting commentary for the person who signs it.

Claude for Excel shares context with Claude for Word, PowerPoint and Outlook, so one conversation can pull figures from a workbook into a Word memo, and Outlook drafts wait for someone to click send. On Team and Enterprise plans the cross-app setting defaults to off. Set the firm's formatting conventions once in the add-in's Instructions field, and save a finished process as a skill the team can reuse.

IT deploys the add-in and owns the controls. The workflow belongs to the people who run the close, because the steps of a reconciliation often live in one or two heads, and the written version is rarely how it runs. Write the process down first: the files, the order, the tolerances, who signs. Then decide which steps Claude reads and drafts, and which stay with a person.

That is the work a forward deployed engineer does with a fund's finance team: discovery on the process as it actually operates, a design, then a build your own people write and run. Our homepage describes one quarterly roll forward case this way: they review and explain what moved in hours, not days.