Venture Capital Workflows
Investment Memo Drafting With Claude: A Practical Guide for VC Teams
The first draft is cheap. The decision underneath it still has to hold up.
Anthropic's Claude is increasingly part of how venture capital teams turn scattered call notes, reference checks, and data room summaries into an investment memo a partner can act on quickly. This guide walks through how VC teams draft, pressure-test, and version memos with Claude, including prompts for a devil's-advocate risks section and a decision record for follow-on rounds. You'll also find a comparison of Claude and ChatGPT for investment research, and the limits worth remembering before any memo reaches the investment committee.
Investment memo drafting with Claude means feeding it raw diligence material, call notes, reference call transcripts, founder decks, and data room summaries, and asking it to draft a structured memo: thesis, market, team, risks, and a recommendation. Claude does the first-draft synthesis and flags where evidence is thin, but a partner still owns the judgment calls, the devil's-advocate stress test, and the final investment decision. Used well, it compresses the hours spent assembling a memo, not the hours spent deciding whether to invest.
Published
Free lesson · no account neededYour AI Deal Desk: Where Claude Fits in Finance WorkPrivate Equity & Finance with Claude7 minStart the free lessonKey Takeaways
- Claude is best used to synthesize raw diligence material, call notes, reference calls, and data room summaries, into a structured memo draft, not to generate the investment decision itself.
- Evidence markers matter: a well-prompted memo flags which claims are sourced, which are inferred, and which rest on a single thin data point.
- A dedicated devil's-advocate risks section, written against the deal using the same evidence, catches the optimism that creeps into a first draft.
- A short decision record at the end of each memo, what you believed and why, becomes the baseline for judging a follow-on round later.
- Confidential founder, LP, and data room information needs the same care in an AI tool as it would in any other system that touches diligence data.
- Claude and ChatGPT are both capable for investment research; the real differences show up in long-document handling, adversarial framing, and connector ecosystems, not in one being categorically "better."
Claude vs ChatGPT for VC Investment Research
| Task | Claude | ChatGPT |
|---|---|---|
| Long data room documents | Handles long documents and multi-file context well, useful for stitching together a sprawling data room | Also handles long context, though very large document sets sometimes need more manual chunking |
| Reference call synthesis | Strong at structured synthesis and tends to follow instructions to mark uncertainty explicitly | Comparable synthesis quality; how clearly it marks uncertainty depends more on how the prompt is written |
| Devil's-advocate risk framing | Takes direction well on adversarial, skeptical framing without becoming evasive about the deal's weaknesses | Also capable of adversarial framing, though it can lean toward hedging unless pushed with a firmer prompt |
| Connectors to finance data sources | Plugin-based agent templates for financial services connect to market and company data providers for research workflows | Depends on the ChatGPT plan and whatever third-party plugins or browsing access is enabled |
| Writing in a firm's existing memo voice | Follows a supplied template and style guide closely across a long drafting session | Also follows a supplied template closely; both tools benefit from an explicit style example |
Copy and Paste
Prompts You Can Use Today
Draft a memo skeleton from raw call notes
You are a venture associate. Here are my raw notes from three calls with the founding team, plus two reference calls with former colleagues. Draft a structured investment memo skeleton with sections for thesis, market, team, traction, risks, and recommendation, and mark any section where the notes don't give you enough to write a confident paragraph.
Swap in your own call notes and section headings; keep the instruction to flag thin evidence, that's the part that matters most.
Turn reference call transcripts into a synthesized reference section
Here are transcripts from four reference calls about this founder. Summarize what each reference said about the founder's strengths and weaknesses, note anywhere references disagree with each other, and write a two-paragraph reference summary that an investment committee could read without needing to see the raw transcripts.
Paste real transcripts, not summaries you've already written, so Claude is working from the source material.
Build a devil's-advocate risks section
Act as a skeptical partner who wants to kill this deal. Read the attached memo draft and write a devil's-advocate risks section that argues against investing, using only the evidence already in the memo plus reasonable extrapolations, and flag any risk argument that would need more diligence before the partner meeting.
Run this on your own draft before anyone else sees it; it's far more useful as a private stress test than a polished section.
Flag thin evidence in a data room summary
Summarize this data room folder's financial model, cap table, and customer contracts into a one-page diligence summary, and explicitly mark any claim in the summary that rests on a single source, an unverified founder statement, or a document that looks incomplete.
Ask for the marking explicitly; without it, a summary reads with the same confidence whether a claim is well-sourced or not.
Write a decision record for a follow-on round
Write a decision record for our follow-on investment decision: what we believed at the time of the initial investment, what has changed since then based on the attached board updates, what our current thesis is, and what specific metric or milestone would make us reconsider before the next round.
Keep this short and specific. A vague decision record is no more useful later than no record at all.
Compare two companies in the same category
Compare these two competing companies in the same category using only the attached pitch decks and the public information I've pasted below, and produce a short comparison table covering market position, team background, and the single biggest open question for each company.
Good for sector mapping before a partner meeting, not as a substitute for direct diligence on either company.
Draft an investment thesis from a partner's voice memo notes
Here is a rough transcript of a partner's voice memo explaining why they're excited about this deal. Turn it into a one-paragraph investment thesis draft written in formal memo language, keeping every specific claim the partner made and flagging any claim that sounds like opinion rather than evidence.
This keeps the partner's original reasoning intact instead of letting Claude invent a tidier story than the one they actually told.
From Anthropic
What Anthropic offers for Venture Capital
Agents for financial services
Ten agent templates released in May 2026, including a market researcher and a meeting preparer, each available as a plugin in Claude Cowork and Claude Code, alongside new connectors such as Dun & Bradstreet, Guidepoint, IBISWorld, and Third Bridge.
· Source
In Practice
How Venture Capital Teams Use Claude
According to Anthropic's published customer story
Airtree
According to Anthropic's customer story, Airtree, a venture capital firm, uses Claude Cowork for market and competitor research and for board summaries, with research that used to take days now done in minutes.
Illustrative walkthrough
A seed-stage associate drafts a full memo from loose call notes
- The associate pastes raw notes from founder calls and two reference calls into Claude along with the firm's memo template.
- Claude returns a structured first draft, with sections marked wherever the notes don't support a confident paragraph.
- The associate fills the marked gaps with a quick follow-up call rather than letting the draft paper over them.
- A devil's-advocate pass against the finished draft surfaces a competitive risk the team hadn't discussed directly.
- The memo goes to partners with the risks section intact and a short decision record attached at the end.
Illustrative walkthrough
A growth-stage team pressure-tests a deal before the partner meeting
- The deal lead asks Claude to summarize a data room folder and flag any claim resting on a single, unverified source.
- Two flagged claims turn out to need a direct follow-up with the company before the memo can be finalized.
- The team runs the devil's-advocate prompt on the near-final draft and finds the competitive section is weaker than it reads.
- That section gets rewritten by hand using the specific evidence the devil's-advocate pass pointed to.
- The final memo includes a decision record stating the thesis and the specific milestone that would trigger a reconsideration.
Venture Capital and Claude: the Published Numbers
| Figure | Context | Source |
|---|---|---|
| two days to minutes | Market and competitor research time at Airtree, per Anthropic's customer story | Source |
| 30 minutes to two minutes | Board summary time at Airtree, per Anthropic's customer story | Source |
| 130+ portfolio companies | Size of Airtree's portfolio, per Anthropic's customer story | Source |
The Workflow
What Investment Memo Drafting With Claude Actually Looks Like
Investment memo drafting with Claude is less about one magic prompt and more about a sequence of passes over the same raw material.
From Notes to Structure
Claude turns scattered call notes and transcripts into a memo skeleton organized by thesis, market, team, traction, and risk, so a human writer isn't starting from a blank page.
Reference Call Synthesis
Multiple reference call transcripts get condensed into a short, readable summary that highlights agreement and disagreement across references instead of forcing a partner to re-read every call.
Data Room Gap-Flagging
Claude can scan a data room summary and mark which claims rest on a single document or an unverified founder statement, rather than presenting every line with equal confidence.
Devil's-Advocate Risks
A dedicated risks section argues against the deal using the same evidence the memo already has, so the optimism of a first draft doesn't quietly become the final word.
Decision Records
Each memo can end with a short record of what the team believed at the time of investment, which becomes the honest baseline for judging a follow-on round later.
Evidence, Not Confidence
Why Evidence Markers Matter in an AI-Drafted Memo
A language model writing prose has no natural reason to sound less confident about a thin data point than a well-documented one. Both come out as clean, declarative sentences unless you ask for something different. That's a real problem for an AI investment memo, because a memo's job isn't just to read well, it's to tell a partner exactly how much weight each claim can bear.
The fix is simple and worth building into every drafting prompt: ask Claude to tag each claim by source, this came from a call note, this came from the founder's deck, this is an inference I'm making from two data points, and to flag anywhere it's filling a gap rather than reporting something it was told. That one instruction changes an AI investment memo from a tidy narrative into something closer to a working document.
This matters more as the memo moves further from the person who did the diligence. A partner reading a draft three steps removed from the original call has no way to know which paragraph is load-bearing and which one is a reasonable guess, unless the memo tells them. Treat any unmarked claim in a Claude investment thesis draft as unverified until someone checks it against the source.
The Drafting Pass
A Workflow for Drafting a Memo From Raw Diligence Material
Most VC diligence starts as a mess: loose notes from founder calls, a handful of reference call transcripts, a data room with uneven document quality, and a deck that's already three versions out of date. That's the actual starting material for investment memo drafting with Claude, not a clean brief, and it's worth feeding it in roughly that raw state rather than pre-summarizing it yourself first.
The first pass is structural. Give Claude the raw notes and a memo template, if your firm has one, and ask for a full draft organized by section, with explicit markers anywhere the notes don't support a confident paragraph. This draft is disposable in the sense that no sentence in it should be treated as final, but it's genuinely useful as a shape to react to.
The second pass is adversarial. Feed that draft back in and ask for a devil's-advocate risks section that argues against the deal. This catches the thing a first draft almost never catches on its own: an overly tidy narrative where every data point happens to support the thesis.
The third pass is the decision record, a short closing section that states what the team believes right now and what would change their mind. It's the cheapest insurance a memo can carry, because it turns a future follow-on debate from a guessing game about what everyone thought a year ago into a quick re-read of what they actually wrote down.
Prompting for Skepticism
The Devil's-Advocate Risks Section: How to Prompt It Well
Ask Claude to write a risks section without more direction and you'll often get a list of generic, deal-agnostic risks: market timing, competition, execution. Those aren't wrong, but they're not useful either, because they'd apply to almost any company in the sector. A devil's-advocate risks section only earns its place in the memo if it's specific to this deal and this evidence.
The better prompt assigns a role and a constraint: act as a partner who wants to kill this deal, and argue using only the evidence already in the memo plus reasonable extrapolations from it. The role framing pushes past the model's natural tendency to be agreeable, and the evidence constraint keeps the risks section grounded instead of drifting into speculation that has nothing to do with this specific company.
It's worth running this pass privately before anyone outside the deal team sees the draft. The value isn't in producing a polished adversarial section for the memo, though that can be useful too, it's in surfacing the one or two arguments against the deal that the team hadn't fully reckoned with before walking into the partner meeting.
The Long Game
Decision Records for Follow-On Rounds
A follow-on decision is really a comparison between what you believed at the time of the initial check and what's true now. Most firms don't actually have a clean record of the first half of that comparison. Memories soften, the original thesis gets remembered as whatever turned out to be right, and the follow-on debate ends up re-litigating the first investment instead of evaluating the company's actual progress.
A decision record closes that gap. Ask Claude to draft one at the end of the initial memo: what we believe right now, why, and what specific metric or milestone would make us reconsider. It only takes a paragraph, but it has to be written at the time of the decision, not reconstructed later, or it loses the thing that makes it useful.
At follow-on time, feed that original decision record back in alongside the company's recent updates and ask for a comparison: what's changed, what's held up, and what the current thesis should be. This is a much more useful version of an AI investment memo than one written from scratch, because it's explicitly measuring the company against the firm's own prior stated beliefs rather than against a generic growth-stage checklist.
The Broader Stack
Where Claude Fits in the Broader VC Research Stack
Memo drafting is one piece of a research workflow that increasingly runs through Claude end to end.
Market and Competitor Research
According to Anthropic's customer story about Airtree, a venture capital firm has used Claude Cowork to compress market and competitor research that used to take much longer into a far quicker task, the same kind of research that feeds the market section of a memo.
Board Summaries
The same customer story describes using Claude Cowork to speed up board summary writing, a nearby workflow to memo drafting: both are about turning raw material into something a partner or board member can read quickly.
Agent Templates for Finance Work
Anthropic has released agent templates for financial services, including a market researcher and a meeting preparer, available in Claude Cowork and Claude Code, extending the same idea into connected research workflows.
Where to Draw the Line
The Limits: Confidentiality, Hallucination Risk, and Partner Judgment
A data room contains some of the most sensitive material a founder will ever hand over: unreleased financials, cap tables, customer contracts, sometimes personal information about the team. Before pasting any of it into any AI tool, know your firm's own data handling policy and the tool's actual terms, rather than assuming convenience implies permission. This isn't a Claude-specific concern, it applies to any system that touches diligence material, but it's the first thing to get right before any of the workflow above is worth using.
The second limit is simpler to state and easy to forget in practice: a language model can write a confident, well-structured paragraph about something that isn't true, especially when the underlying notes are ambiguous or incomplete. Evidence markers reduce this risk but don't eliminate it. Every claim in a memo that's going in front of an investment committee needs a human who actually did the diligence to confirm it against the source, not just skim it for tone.
The decision itself is the clearest limit of all. Claude can produce a structured, well-sourced first draft, a sharp devil's-advocate section, and a clean decision record, but the judgment call, whether this company, this team, and this market add up to a check worth writing, sits with the partners, along with the fiduciary responsibility that comes with it. Investment memo drafting with Claude speeds up the paperwork around a decision. It was never meant to make the decision.
Reading About Claude and Being Good at It Are Different Things
A workflow like this reads simply on a page and takes real practice to run well, especially the parts about prompting for skepticism instead of agreement. The free Nightschool AI curriculum is hands-on from the first lesson, built around exactly this kind of practical, high-stakes work.
Frequently Asked Questions
How do VCs use AI to write investment memos?
VCs mainly use AI, including Claude, to turn raw diligence material, call notes, reference call transcripts, and data room summaries, into a structured first draft organized by thesis, market, team, risks, and recommendation. The AI does the synthesis and can flag thin evidence; the partner still does the judgment work, the adversarial stress test, and makes the final call.
Is Claude or ChatGPT better for investment research?
Neither is categorically better; the differences show up in specific tasks. Claude tends to handle long data room documents and adversarial, devil's-advocate framing well, and Anthropic offers finance-specific agent templates and connectors for it. ChatGPT is also a capable research assistant. Most teams get better results from clear prompting than from picking one tool over the other.
How do VCs use AI to write an investment memo from raw diligence notes?
Investment memo drafting with Claude means feeding it raw diligence inputs, call notes, reference calls, decks, and data room summaries, and asking it to produce a structured memo draft with explicit markers for thin or unverified evidence. It's a drafting aid, not a decision-making one: the investment call still belongs to the partners.
How do venture capital firms use AI to write investment memos today?
Firms use AI like Claude to produce a complete first draft of an investment memo from the diligence material they give it, but the draft shouldn't be the last step. A memo that goes to an investment committee needs a human who did the diligence to verify every claim against its source, run a genuine devil's-advocate check, and make the actual recommendation.
How do you prompt Claude for a devil's-advocate risks section?
Assign a clear adversarial role, a partner trying to kill the deal, and constrain it to the evidence already in the memo plus reasonable extrapolations. Without that framing, Claude tends to produce generic, sector-wide risks instead of risks specific to this deal and this evidence.
Is Claude better than ChatGPT for investment research in venture capital?
Neither is categorically better; both can produce a strong first draft that still needs human verification of every sourced claim, a genuine adversarial read, and sign-off from whoever did the diligence. Claude tends to handle long data room documents and devil's-advocate framing well, and with either tool the output depends heavily on prompt quality. Either way, the AI investment memo workflow speeds up assembly, not the review.
What's a decision record and why do VCs need one for follow-on rounds?
A decision record is a short note, written at the time of the initial investment, stating what the team believed and what would make them reconsider. At follow-on time, it gives the team an honest baseline to compare the company's actual progress against, instead of relying on memory or revisionist hindsight.
Is it safe to put confidential founder data into Claude?
Before pasting data room material into any AI tool, confirm your firm's data handling policy and the tool's actual terms for how that data is used and retained. This is true of any system that touches diligence material, not just Claude, and it should be settled before the workflow is used on real deals.
What's a VC memo template AI can actually fill in reliably?
A template with clear, consistent section headings, thesis, market, team, traction, risks, recommendation, works well because it gives the AI an explicit structure to populate and gives a reader a consistent place to look for each kind of claim. Vague or freeform templates produce much less consistent drafts.
How do VCs use AI to synthesize reference calls for an investment memo?
Feed Claude the raw reference call transcripts and ask for a short synthesis that highlights where references agree and disagree, rather than a flattering summary. The useful output is the disagreement, not the consensus, since that's usually what a partner needs flagged before a decision.
Claude in Venture Capital, by role
Memo Drafting Is a Skill, Not a Button
The prompt is the easy part. Knowing what to distrust in the answer is the real skill.