--- title: "How to Use Claude for Financial Due Diligence on a Deal | By The People Tech" description: "How a private equity deal team uses Claude for financial due diligence: one project per deal, a cited first pass on the data room, and the team's checks." url: https://bythepeople.tech/insights/claude-financial-due-diligence --- [Blog](/insights) Claude for due diligence Claude for financial due diligence # Use Claude on a deal as a first reader your team can _check_. To use Claude for financial due diligence on a deal, set up one Claude project per deal with the CIM, financials and data room files your compliance team has approved, ask for a structured first pass (a data pack, an add-back list, open questions), and require every figure to cite the page it came from. The deal team then traces the numbers that matter back to source. Claude drafts; the person whose name goes on the memo decides. By [Tim Miner](/about-us/tim-miner), founder and CEO · Published 27 October 2026 Financial due diligence is the buyer's test of a target's numbers: the quality of earnings, revenue, working capital, debt and the contracts behind them. Anthropic lists due diligence among Claude's main uses for private equity analysts, describing analysis of data room documents to extract commercial terms, financial metrics and risk factors. Anthropic now packages that work: a due diligence data pack skill, a private equity plugin with diligence checklists and investment committee memo drafting, and connectors to data rooms in Box, Egnyte and SS&C Intralinks. The tools are the easy part. What decides the result is how a deal team frames each request and checks each answer. Before the first file ## What needs to be in place _before_ Claude reads a deal's documents? Ask who at your firm has said which deal documents may go into which Claude plan, and whether that answer is written down. Start with where the documents may go. Use the firm's approved Claude plan, never a personal account, and have compliance say in writing what may be uploaded: the CIM, the management accounts, the quality of earnings report, customer contracts. Anthropic's financial services page states that by default your data is not used for training its generative models; your firm's own terms with the vendor still decide. Keep the rule to [one page the desk will read](/guides/ai-usage-policy-investment-firms). Then build one Claude project per deal, as Anthropic's own guidance suggests, so the deal's files and instructions stay together across conversations. Write project instructions that set the transaction, the review period, the reporting currency, the materiality threshold and your firm's memo template. Where the data room sits in Box, Egnyte or SS&C Intralinks, Anthropic lists a connector for each. Anthropic also notes that large files or many documents can exceed Claude's input limit, so load what the question needs. The first pass ## What diligence _work_ can Claude do on the deal documents? Claude reads and drafts; the deal team decides what any of it means. Anthropic's tutorials describe pulling commercial terms, financial metrics and risk factors out of data room documents and organising them for review. Its open-source private equity plugin packages several steps as skills, from the checklist to a first draft of the investment committee memo, and your firm can edit them. ### A data pack from the data room The due diligence data pack skill turns CIMs, offering memorandums and other diligence files into a structured Excel pack with historical and projected financials, customer lists and contract terms. Anthropic lists it as a research preview. ### A checklist by workstream The plugin's dd-checklist skill drafts diligence checklists by workstream. Start from your firm's own request list, and use the draft to show what that list is missing. ### Revenue quality The unit-economics skill covers ARR cohorts, LTV/CAC, net retention and revenue quality. Point it at the customer file in the data room, and ask for the top customers' share of revenue by year. ### Prep for the management meeting The dd-meeting-prep skill prepares for management presentations and expert calls. The useful output is a list of questions where the CIM and the financials disagree, each with the page on both sides. The request ## How should a deal team _word_ the request so the numbers come back right? Ask whether your analysts tell Claude which EBITDA, from which document, for which period, or simply ask for EBITDA. Most disappointing output is a specification failure. A request for “adjusted EBITDA” on a deal with a CIM, a quality of earnings report and monthly management accounts has three answers, and the model will pick one unless told. Name the document, the period and the definition. Ask for a table with a source column: file, page or tab, and the line as written. Tell it to write “not found” rather than estimate. Anthropic's own guidance is to decide what a good output looks like before you start, and to name the skill and the format the output needs in the same prompt. For the add-backs, that means one row per adjustment, the amount management claims, where it appears, and whether the audited accounts support it. The [accuracy guide](/guides/ai-accuracy-investment-firms) covers the other ways errors get in. The check ## How much of Claude's diligence output does the deal team need to _check_? Check it as you would a first draft from a new analyst. That is Anthropic's own advice for its finance skills: trace two or three of the figures back to the underlying source, and trace the numbers that matter before the output leaves your desk. On a deal, that means every figure that reaches the investment committee memo, the model or a price conversation. Claude for Excel tracks and explains its changes and lets you go straight to the cells it cites, which makes the tie-out quicker. Anthropic's tutorial still notes that complex Excel models may need manual review, and its financial services repository says its agents draft work for a qualified professional and do not make investment recommendations. Two things stay with people: whether to do the deal, and what the business is worth. The person whose name goes on the memo decides, with the source open. **Financial due diligence on a deal: what Claude drafts and what the deal team keeps** | | What Claude drafts | What the deal team keeps | | ------------------- | --------------------------------------------------------------------------------------- | ----------------------------------------------------------------------- | | Financials | A data pack in Excel from the CIM and data room files, each figure with its source page | Tracing the figures that reach the memo or the model back to the source | | Add-backs | One row per adjustment, the amount management claims and where it appears | Whether the audited accounts support each one | | Revenue quality | Cohorts, net retention and the top customers' share of revenue by year | What the concentration means for the deal | | Management meeting | Questions where the CIM and the financials disagree, with the page on both sides | Which questions to put to management | | Investment decision | A first draft of the investment committee memo in the firm's template | Whether to do the deal and what the business is worth | The deal team ## How does a _whole_ deal team learn to use Claude this way? A deal team uses Claude well when each person has done the work once on a document they know, and seen where the answer went wrong. One associate who has worked it out alone does not carry the rest. Anthropic's skills and plugins are starting points; your firm's checklist, definitions and memo template are what make them fit. That is what our [Prompt Engineering Workshop](/workshops) is for: ninety minutes to two hours, one deal team, on its own CIMs and data room files or on mock files written for its work. It covers your document and process, what not to delegate, how to specify a whole task and how much checking the result needs. We have delivered over 150 workshops at more than 100 financial services firms. The [retainer](/retainer) and a [forward deployed engineer](/forward-deployed-engineer) come after it, if the firm wants more. Most firms start with the workshop and decide from there. Tell us the deal documents your team is working and where Claude stalls. If we are the wrong firm for it, we will say so. Claude can do the first read of a data room; the judgment on the deal stays with the people who sign it. ## Questions on Claude for financial due diligence ### Can we upload a CIM and data room files to Claude? Only under your firm's approved plan, and once compliance has said what may go in. Anthropic states that by default your data is not used for training its generative models, but your firm's own terms with the vendor decide. A workshop can run on mock files written for the team, so no live deal data needs to leave the firm to learn the method. ### Which Claude plan do the financial services skills and plugins need? Anthropic's six financial services skills, including the due diligence data pack, are in research preview for Claude for Financial Services users who join a waitlist. Its agent plugins for Claude Cowork and Claude Code are available on all paid plans, and on Team and Enterprise plans an admin adds the Financial Services marketplace for the whole organization. ### Can Claude read the whole data room at once? Claude can process multiple documents in one request, and Anthropic's tutorial describes it working across hundreds of data room documents. The same tutorial notes that large files or many documents can exceed Claude's input limit. Load the files the question needs, or use a data room connector such as Box, Egnyte or SS&C Intralinks where your firm has one, and ask for the source of every figure. ### Will Claude tell us whether to do the deal? No. Anthropic's financial services repository states that its agents draft analyst work product for review by a qualified professional and do not make investment recommendations. Claude can assemble the data pack, the questions and a first memo draft. Whether to invest, and what the business is worth, stays with the people whose names go on the memo. ### How do we know Claude is working on our deals? Measure it on your own volume. Pick one diligence task, such as building the data pack for a new deal, and record the time and the errors found at tie-out, before and after. For adoption, count how many people on the deal team use it well on their own work, by month, rather than how many licences are active. ## Sources * [Anthropic, “Claude for Financial Services”, updated 10 April 2026](https://www.anthropic.com/news/claude-for-financial-services) * [Anthropic, “Advancing Claude for Financial Services”](https://www.anthropic.com/news/advancing-claude-for-financial-services) * [Anthropic, “Agents for financial services”](https://www.anthropic.com/news/finance-agents) * [Claude Academy, “Six skills for financial service professionals”](https://academy.claude.com/tutorials/claude-for-financial-services-skills) * [Claude Academy, “Claude for financial services overview”](https://academy.claude.com/tutorials/claude-for-financial-services-overview) * [Claude Academy, “Install financial services plugins for Claude”](https://academy.claude.com/tutorials/install-financial-services-plugins-for-cowork) * [GitHub, anthropics/financial-services, “Claude for Financial Services”](https://github.com/anthropics/financial-services) ## Read next * [Claude Workshops for Private Equity Teams](/insights/claude-workshop-private-equity). What a Claude workshop for a private equity deal team covers on CIMs, IC memos and portfolio reports, and how to judge one. * [How to make AI more accurate on investment documents](/guides/ai-accuracy-investment-firms). Where the errors come from when a model reads a CIM or a credit agreement, and the five things that cut them. * [AI for Private Equity Firms in New York](/industries/private-equity). Deal screening, CIM review, IC memos and portfolio monitoring, taught and built on your own documents. * [AI usage policy for an investment firm](/guides/ai-usage-policy-investment-firms). The one page the desk reads: what may go in, what may not, and who decides the edge cases. ## Tell us what you are seeing. If your documents or your experience say otherwise, we would like to hear it. [Book 30 minutes with Tim](https://calendly.com/timminer-1/chat-with-tim) [hello@bythepeople.tech](mailto:hello@bythepeople.tech)