How to Configure Claude for Investment Banking Research Tasks

You cannot train Claude to be an investment banker. Investment banking involves regulated advice, client relationships, professional licensing (Series 79 and related FINRA qualifications), fiduciary duties and judgment that no language model holds. What you can do is configure Claude as a research assistant that handles the information-gathering and drafting tasks in your workflow, with a clear review step for everything it produces.

The distinction matters practically. The useful question is not "can Claude be my investment banker" but "which parts of investment banking research can Claude help me do faster and more thoroughly." The answer is broader than most practitioners expect. Tools like fenbs retain the non-confidential instructions and project decisions that make Claude consistently useful across sessions.

What investment banking research tasks Claude handles well

Research tasks that involve reading and summarizing large volumes of text are where Claude adds the most value. Reading a 10-K to identify key risk factors, summarizing an industry report, explaining a financial instrument's structure, and drafting sections of a pitch book from a brief are all tasks Claude can accelerate.

• Reading and summarizing public filings: 10-K, 10-Q, 8-K, proxy statements.
• Explaining the structure and mechanics of financial instruments and transactions.
• Drafting sections of pitch books, information memoranda or client briefings from your outline.
• Summarizing industry or market reports for a specific thesis or question.
• Comparing public company financials across a specified set of metrics.
• Structuring a research note from your own analysis and findings.

The common pattern: you give Claude the source material and a specific question; it produces a structured output; you verify every material claim against the source before it goes anywhere. Claude is the research assistant; you are the analyst.

What Claude should not do in this work

Claude does not have access to live market data, proprietary financial databases or non-public information. It cannot verify its own outputs against Bloomberg or FactSet. Every number, date or claim it produces from its training data needs to be checked against a primary source before use in any client-facing or compliance-sensitive document.

Do not put material non-public information, deal terms, client identities or confidential transaction details into Claude. Even in a context window, treating an AI tool as a secure environment for MNPI is a compliance risk. Describe transaction structures in general terms where you need assistance; bring the specifics back in yourself after the draft.

Do not ask Claude to produce investment recommendations, price targets or ratings. These are regulated outputs. Claude can help you structure an argument you have already made; it cannot make the argument for you on a regulated matter.

Setting up Claude with verified examples

The most effective configuration is a brief that tells Claude: the type of document it is helping with, the audience, the standard of evidence required and the review step that will follow. Start with a format example from a previous piece of work you are happy with — a section of a research note or pitch book that represents the output quality you want.

Give Claude that example and say: "This is the format and level of precision I need. When you summarize a filing, follow this structure: [sections]. When you draft a section of a pitch book, follow this approach: [approach]. Flag any claim that comes from your training data rather than the source material I give you."

That flagging instruction is important. Claude can often distinguish between information you gave it and information it retrieved from training. Asking it to flag training-data claims gives you a fast audit path: claims it flags need primary source verification; claims from the materials you gave it need verification against those materials.

Writing review rules that make output reliable

Review rules tell Claude what to do before producing output, not just what format to use. Useful rules for investment banking research: "Always cite the source document and section for any financial figure." "Do not produce a claim about a company's future performance." "If the source material does not answer the question, say so rather than filling the gap."

The last rule is especially important. Claude will try to be helpful even when it lacks the information to be accurate. Explicitly asking it to say "I do not have source material for this claim" produces more useful output than letting it produce a confident-sounding answer you then spend time disproving.

Organizing research prompts for reuse

Investment banking research involves recurring task types: reading a 10-K to answer a specific question, comparing multiples across a peer group, summarizing a management discussion section. For each of these, a well-crafted prompt that you have tested and refined is more valuable than starting from scratch each time.

Build a library of tested prompts for your most common tasks. Include the format you want, the level of precision required and the review rules. When a new analyst joins the team, a shared prompt library reduces the ramp-up time for AI-assisted research.

For the broader question of how to use AI in financial analysis, see what is Claude AI. For configuring Claude across other professional tasks, how to train Claude AI covers the general approach.

Compliance and confidentiality considerations

Before using any AI tool in an investment banking workflow, check your firm's AI use policy. Most firms have one, and it will specify what information may be used with external AI tools and what review and approval process applies. Your compliance and legal teams are the right contacts, not a general-purpose guide.

FINRA and the SEC have published guidance on AI use in regulated activities. The SEC's Division of Examinations has flagged AI use as an examination priority. Review the current guidance from your regulators and keep it current; this area is changing quickly.

How to do this with fenbs

1. Create a free account and set up a project for your research work.
2. Add your non-confidential research instructions as a context note: format examples, review rules, the standard of evidence you require.
3. Save tested prompts for recurring task types in Skills and prompts — 10-K reading, peer group analysis, pitch book section drafting.
4. Record project-level decisions in Decisions and rules: which analytical framework applies, which sources are approved for a specific thesis, which claims require primary source verification.
5. Connect Claude from Integrations and ask it to read the project context at the start of each research session.

fenbs stores non-confidential instructions and project context. Do not record deal terms, client names, MNPI or any information subject to a confidentiality agreement in fenbs. The research assistant configuration lives in fenbs; the confidential deal work lives in your firm's secure systems.