Asset Management Playbook
IC Memos With Claude: Drafting Investment Committee Memos and Portfolio Reviews
The memo still says what you decided. Claude just stops you from retyping the holdings table to say it.
Investment committee memos eat an analyst's week every quarter: pull the holdings, reconcile the numbers against the performance system, write the narrative, and defend every sentence in the room. This guide covers how asset managers use Anthropic's Claude to draft IC memos, portfolio reviews, and performance commentary from existing holdings data and research notes, with every gap flagged and every figure checked against the system of record before a human signs off.
Using IC memos with Claude means turning holdings data, research notes, and prior commentary into a structured first draft of an investment committee memo, portfolio review, or performance write-up, then checking it against the firm's performance system before it goes to committee. Claude can organize the narrative, flag unsupported claims, and mark missing data points, but it does not have live access to your custody or performance platform, so every figure still needs a human reconciliation pass before the memo is final.
Published
Free lesson · no account neededYour AI Deal Desk: Where Claude Fits in Finance WorkPrivate Equity & Finance with Claude7 minStart the free lessonKey Takeaways
- IC memos with Claude work best as a drafting and structuring step, not a figures source: feed it holdings data and notes, and reconcile every number against your performance system before the memo goes to committee.
- Claude can turn scattered research notes and prior quarter commentary into a consistent memo structure, saving the analyst from rebuilding the same template from scratch each cycle.
- Marking gaps explicitly, rather than letting Claude guess at a missing attribution figure or benchmark number, is the single habit that keeps AI-assisted memos safe to circulate.
- Portfolio review AI tools and performance commentary AI drafts still need the same sign-off chain as a human-written memo: nothing changes about who is accountable for the numbers.
- Confidentiality rules for client holdings and non-public positions apply to AI drafting exactly as they apply to any other software touching that data.
Drafting an Investment Committee Memo: Three Approaches
| Approach | Time to first draft | Where accuracy risk sits | Best use |
|---|---|---|---|
| Fully manual | Longest: analyst builds narrative and tables from scratch each cycle | Lowest tool risk, but still prone to copy-paste errors between source and memo | Highly sensitive or unusual situations where every sentence needs direct authorship |
| AI draft, no verification | Fastest: a full draft in minutes from notes and holdings data | Highest: unverified figures can read as confident and still be wrong | Never appropriate for anything that reaches a committee |
| AI draft with reconciliation pass | Fast: draft in minutes, reconciliation against the performance system adds a fixed review step | Low: gaps are marked explicitly and checked before sign-off, so risk sits in the review step, not the writing step | The practical default for recurring IC memos and portfolio reviews |
Copy and Paste
Prompts You Can Use Today
Draft a first-pass IC memo from holdings and notes
You are helping me draft an investment committee memo for our quarterly review of the [strategy or fund name] portfolio. Here is the current holdings data and my research notes from this quarter: [paste holdings export and notes]. Write a structured first draft with sections for portfolio positioning, top contributors and detractors, changes since last review, and open risks. Mark clearly, in brackets, any figure you are inferring rather than finding explicitly stated in what I gave you, so I can confirm it against our performance system before this goes further.
Swap in your actual template sections if your committee expects a different structure.
Turn research notes into a consistent position write-up
I'm attaching my raw research notes on [company or position name], taken across several meetings and calls this quarter. Rewrite them as a position summary suitable for an investment committee memo: thesis, what has changed since initiation, the catalysts we're watching, and the risks to the thesis. Keep my actual conclusions and numbers intact, don't add new claims, and flag with [NEEDS CONFIRMATION] any place where you had to assume a number or date I didn't give you.
Works best when you paste the actual notes rather than summarizing them yourself first.
Draft performance commentary without overselling results
Using the attached performance attribution output for the quarter, draft the performance commentary section of our portfolio review. Explain what drove returns relative to the benchmark in plain language a committee member who isn't a quant can follow, attribute specific moves to specific positions or sectors where the data supports it, and avoid language that implies skill or repeatability unless the attribution data actually shows a consistent pattern across multiple periods.
Build a portfolio review from a prior quarter's version
Here is last quarter's portfolio review memo and this quarter's updated holdings and performance data. Update the memo to reflect what has actually changed: new positions, closed positions, material weight changes, and any shift in sector or factor exposure. Keep the same section structure as last quarter so the committee can compare easily, and list every number you changed from the prior version in a short changelog at the end.
The changelog is what makes the reviewer's job faster, not the update itself.
Pressure-test a memo's claims before circulating it
Review this draft investment committee memo as if you were a skeptical committee member. List every factual claim or figure in the memo that isn't directly supported by the data I provided earlier in this conversation, every place the reasoning jumps from evidence to conclusion without showing the step, and any sentence that reads more confident than the underlying data supports. Don't rewrite the memo, just produce the list.
From Anthropic
What Anthropic offers for Asset Management
Agents for financial services
Ten agent templates released in May 2026, including an earnings reviewer, market researcher, valuation reviewer, and statement auditor, shipped as plugins in Claude Cowork and Claude Code, with Excel, PowerPoint, and Word add-ins generally available.
· Source
In Practice
How Asset Management Teams Use Claude
According to Anthropic's published customer story
NBIM (Norges Bank Investment Management)
Anthropic's customer story reports that NBIM, which manages Norway's sovereign wealth fund, saw weekly time savings per employee on Claude-assisted analytical and operational tasks and broad adoption across departments within two months.
According to Anthropic's published customer story
Walleye Capital
In Anthropic's May 2026 financial services post, Walleye Capital's chief executive and chief investment officer said all of the firm's employees use Claude Code.
Illustrative walkthrough
A portfolio manager heading into quarterly review season
- Pulls the current holdings export and performance attribution from the firm's system of record.
- Gathers research notes and call summaries accumulated since the last review.
- Asks Claude to draft an updated portfolio review using last quarter's memo as the template, with every inferred figure marked for confirmation.
- Reviews the changelog Claude produces, checking each flagged figure against the performance system directly.
- Finalizes the memo and presents it to committee, owning every number in the room.
Illustrative walkthrough
An analyst building a new position write-up for IC
- Compiles raw notes from several management calls and internal discussions on the position.
- Asks Claude to turn the notes into a structured thesis, catalyst, and risk summary without introducing new claims.
- Asks a second pass to list every claim that isn't directly supported by the notes provided.
- Resolves each flagged claim by either sourcing it properly or removing it from the draft.
- Submits the memo for committee review with the underlying notes attached for traceability.
Asset Management and Claude: the Published Numbers
| Figure | Context | Source |
|---|---|---|
| 20% time saved weekly per employee | NBIM, on Claude-assisted analytical and operational tasks, per Anthropic's customer story | Source |
| 600+ active users across all departments within the first two months | NBIM's rollout, per Anthropic's customer story | Source |
| 100% of Walleye employees use Claude Code | Stated by Walleye Capital's CEO and CIO in Anthropic's May 2026 financial services post | Source |
Capabilities
What an AI Investment Memo Generator Can Actually Do for an IC Memo
Treat Claude as the analyst who writes the first draft and formats the deck, not the one who owns the numbers.
Structure the narrative
Claude can take a holdings export and a pile of research notes and organize them into the sections a committee expects: positioning, attribution, changes since last review, and risks, without the analyst rebuilding the template every quarter.
Flag what's missing
When asked directly, Claude will mark a number it inferred rather than found explicitly in the source material, which is the difference between a memo you can trust and one you have to re-derive from scratch.
Draft performance commentary
Given an attribution output, Claude can explain in plain language what drove returns relative to a benchmark, which is usually the section that takes longest to write well under deadline.
Speed up the revision cycle
Updating last quarter's memo with this quarter's data is mostly mechanical work; Claude can do the update pass and list exactly what changed, so a human reviewer checks a changelog instead of rereading the whole document.
Turn meeting notes into position write-ups
Research notes taken across several calls over a quarter rarely read as a coherent thesis; Claude can turn them into the kind of position summary a committee memo needs, while preserving the analyst's actual conclusions.
Workflow
From Holdings Data to a First Draft
The workflow that works in practice starts with the same two inputs every quarter: a holdings or performance export from the system of record, and whatever research notes, call summaries, or prior commentary already exist for the positions in question. Paste both into the conversation, describe the memo structure your committee expects, and ask Claude to produce a first draft that follows it.
The useful instruction isn't 'write me an IC memo.' It's the instruction that tells Claude what to do when it doesn't have a number: mark it, don't guess it. Every example prompt in this guide includes that instruction for a reason. A draft that quietly fills a gap with a plausible-sounding figure is more dangerous than a draft with an obvious hole in it, because the hole gets caught and the smooth guess doesn't.
Anthropic's May 2026 release of agent templates for financial services, including a dedicated earnings reviewer and a valuation reviewer shipped as plugins alongside Excel, PowerPoint, and Word add-ins, points at the same underlying idea: a model that can draft and format is most useful when it is also built to flag what it can't verify, not just produce fluent prose.
None of this replaces the analyst's judgment about which positions matter or what the real risk in a thesis is. It replaces the hour spent reformatting last quarter's memo and the hour spent staring at a blank performance commentary section.
Verification
Where the Numbers Actually Live
An IC memo lives or dies on its numbers, and Claude does not have a live connection to your custody platform, your performance attribution system, or your portfolio accounting system. Anything it reports as a return, a weight, a contribution figure, or a risk statistic came from what you pasted into the conversation, not from checking your actual books.
That means the reconciliation step is not optional and does not shrink just because a draft arrived faster. Before a memo goes anywhere near a committee, every figure in it needs to be checked against the performance system it's meant to describe. The gap-marking habit from the previous section makes this step faster, because the reviewer is checking a short list of flagged items plus a spot-check of the rest, not rereading every sentence with the same suspicion.
Firms that have adopted this kind of workflow at scale describe it in terms of time saved rather than headcount replaced. In Anthropic's customer story about NBIM, which manages Norway's sovereign wealth fund, the firm reports weekly time savings per employee on Claude-assisted analytical and operational tasks, with broad adoption across departments within two months of rollout. That is a description of faster work, not of removing the review step; the memo still needs a human who owns the final numbers.
Limits
What Still Needs a Human Before the Memo Goes to Committee
Final figure sign-off
Every number in the memo gets checked against the performance system by a person with the authority to stand behind it; a flagged gap and an unflagged assumption carry the same risk if nobody closes the loop.
Judgment calls on thesis and risk
Claude can summarize a thesis clearly, but deciding whether that thesis still holds, or whether a risk has become material enough to escalate, is the analyst's and portfolio manager's call.
Confidentiality boundaries
Client holdings, non-public positions, and material non-public information follow the same handling rules inside an AI conversation that they follow everywhere else in the firm; what tool is drafting the memo doesn't change what's allowed to go into it.
The room, not the draft
A memo that reads well is not the same as a memo that survives committee questioning; the analyst who presents it still needs to understand every claim in it well enough to defend it live.
Portfolio Reviews
Portfolio Review Season Without the All-Nighter
Quarterly portfolio review season is where this adds up fastest, because the work is repetitive in a way IC memos for a single new position usually aren't. Every strategy needs the same sections updated with the same quarter's data, and most of the actual thinking happened already, in the notes, in team discussions, and in the prior quarter's memo.
A workable pattern: feed Claude last quarter's finished review alongside this quarter's holdings and performance data, and ask it to update the document section by section, producing a short changelog of what moved. The reviewer's job shifts from rereading an entire memo to checking a changelog against the performance system and reading closely only where something changed materially.
Portfolio review AI tools built specifically for this exist as well, often wired directly into a firm's data systems, which closes part of the reconciliation gap described above. Claude's role in a review built this way is still the drafting and formatting layer; the connection to live data, where it exists, is a separate piece of infrastructure the firm has to build or buy, not something a chat conversation provides on its own.
Performance Commentary
Performance Commentary That Doesn't Oversell
Performance commentary is the section of a memo most likely to drift into overconfidence, because attribution output is naturally dense and the instinct when drafting under deadline is to simplify it into a cleaner story than the data supports. Asking Claude to draft performance commentary from an attribution output is useful specifically because the instruction can include the discipline that's easy to skip under time pressure: attribute a move to a position or sector only where the data actually supports it, and avoid language implying skill or repeatability unless a consistent pattern shows up across multiple periods, not just the one being written about.
This is also where the earnings reviewer and valuation reviewer templates Anthropic shipped in May 2026 are aimed: reviewing a single quarter's numbers against a company's own history and against a sector, rather than writing from a blank page. Walleye Capital's chief executive and chief investment officer told Anthropic that the firm's employees all use Claude Code, which is a data point about adoption depth at one hedge fund, not a claim about what any specific workflow produces; the discipline in the prompt still has to come from the person writing it.
The test for performance commentary isn't whether it reads smoothly. It's whether a skeptical committee member, reading the same attribution data independently, would reach the same conclusions the commentary states as fact.
Compliance
Confidentiality and Professional Responsibility
Nothing about using Claude to draft a memo changes the rules that already govern what information can leave the firm's systems. Client holdings, position-level detail on illiquid strategies, and anything that qualifies as material non-public information are subject to the same confidentiality and data-handling policies whether they're being typed into an email, pasted into a spreadsheet, or pasted into an AI conversation. If a firm's policy requires anonymizing position names before sharing research externally, that requirement doesn't loosen because the recipient is a model instead of a person.
The professional responsibility question is separate from the confidentiality question. A portfolio manager who presents a Claude-drafted memo to committee is making the same representations about its accuracy as if they'd written every word themselves. 'The AI wrote that part' is not a defense for an unverified figure that turns out wrong, in the same way that 'the intern wrote that part' never has been. The discipline this guide describes, marking gaps and reconciling every figure against the performance system, exists because that accountability doesn't move.
Ready to Build Faster, More Reliable Memos?
Reading about drafting IC memos with Claude and actually being fast and careful with it are different skills. Nightschool AI's free curriculum teaches the hands-on habits, like gap-marking and prompt structure, that make AI-assisted drafting safe to put in front of a committee.
Frequently Asked Questions
Can AI generate an investment memo?
AI can generate a full first draft of an investment memo, including structure, narrative, and a summarized thesis, when it's given holdings data, research notes, and clear instructions to mark any figure it's inferring rather than confirming. It cannot verify those figures against your performance or custody systems on its own, so the draft still needs a human reconciliation pass before it goes to committee.
Can AI generate an investment memo without introducing errors?
Give it your actual holdings data and notes rather than asking it to write from general knowledge, and explicitly instruct it to flag any number it had to infer. Reconcile every flagged figure, and a sample of the rest, against your performance system before the memo circulates.
Can AI generate an investment memo without a portfolio management system?
Yes, but the two aren't the same thing. An AI investment memo generator like Claude drafts and formats text from data you provide; it doesn't connect to your custody platform, track positions live, or calculate performance attribution on its own. Those remain separate systems that supply the inputs for the draft.
Can AI write performance commentary that's accurate?
AI can draft performance commentary that's accurate if it's working from a real attribution output and instructed not to imply skill or repeatability beyond what the data shows across multiple periods. The accuracy depends entirely on the quality of the data it's given and the discipline built into the prompt, not on the tool alone.
Can AI generate an investment memo for smaller asset managers, or is this only for large funds?
The workflow scales down well: a small team benefits from the same time savings on repetitive drafting that a large fund does, often more, because there's no dedicated team to absorb the reformatting work every quarter. The reconciliation discipline matters at every size.
What happens if Claude gets a number wrong in a memo?
The same thing that happens if any draft contains a wrong number: whoever signs off on the memo is responsible for catching it before it reaches committee. That's why gap-marking and reconciliation against the performance system are treated as mandatory steps, not optional polish.
Can I paste confidential holdings data into Claude?
Only under the same confidentiality and data-handling rules that already govern where that data can go, set by your firm's compliance policies. The tool doesn't change what's permitted; it's still your firm's existing rules about client data and material non-public information that decide what can be shared and how.
Can AI generate an investment memo the same way Anthropic's finance-specific agents do?
Anthropic's May 2026 release added purpose-built agent templates for tasks like earnings review and valuation review, shipped as plugins with spreadsheet and document add-ins. Using Claude directly for IC memo drafting, as described in this guide, is a more general conversational workflow; the dedicated agents are narrower tools aimed at specific recurring review tasks.
Claude in Asset Management, by role
Draft Your Next Memo With Confidence
The committee doesn't care how the draft got written. They care whether the numbers hold up.