How to Write an Investment Committee Memo with AI

To write an investment committee memo with AI, give it your firm’s template and past memos, the deal’s documents and the model, then have it draft the factual sections with every figure cited while partners write the thesis, the recommendation and the judgment on risks. Kepler, the enterprise intelligence platform we build for investment firms, runs the memo as a saved playbook in house style, drafts it from the deal’s own documents, and computes every number in code with a link to the page or cell it came from, so when the committee asks “where’s that from?” the answer is one click away.

The IC memo is where months of work become one decision. It has to be accurate, consistent with the model, honest about risks and written the way the committee expects to read it. AI is good at the first three when it works from sources and computes in code. The fourth, the argument for the deal, belongs to the people who will defend it. This guide covers the structure, the division of labor, a step-by-step drafting process with a worked returns example, house style, citations and version control.

The structure of an IC memo

Section	What it answers	Drafted by	Typical sources
Recommendation and ask	What are we approving, at what price and structure?	Partner	Term sheet, model
Investment thesis	Why this business, why now, why us?	Partner, with AI pulling evidence	Diligence findings, market work
Company overview	What it does, for whom, how it makes money	AI	CIM, management presentation
Market and competition	Size, growth, share, competitors	AI, reviewed by the deal team	Market studies, expert calls, filings
Historical financials	Revenue, margins, cash flow by year	AI	Management accounts, audited statements
Adjusted EBITDA and QoE	What earnings really are	AI, with the deal team	QoE report, CIM bridge
Valuation	Entry multiple, comps, precedents	AI computes, deal team chooses	Model, public comps
Returns	IRR and MOIC by case	AI pulls from the model	Model
Financing	Debt terms, leverage covenants, equity	AI	Term sheets, commitment papers
Risks and mitigants	What could go wrong and what we do about it	Both	Diligence workstreams
Diligence findings and open items	What we learned, what’s left	AI	Workstream reports, Q&A logs
Value creation plan	What we change in the first 100 days and beyond	Deal team	Operating partner work

What AI should draft and what partners should write

A workable rule: AI drafts anything that is a fact, a figure or a summary of a source; people write anything that is a judgment or a commitment.

Good for AI: the company overview from the CIM and management presentation; historical financials and KPIs; the QoE summary and the bridge between the CIM’s adjusted EBITDA and the QoE’s; comps tables; returns tables from the model; financing terms; a first risk register built from every diligence workstream; the open-items list; a conflict list wherever sources disagree.

For partners: the recommendation; the thesis and why the firm should win the deal; which risks matter most and whether the mitigants are enough; the value creation plan; anything that commits the firm to a price, a structure or a person.

The line moves over time. A first risk register drafted by AI is useful because it’s complete: it lists every issue raised in every workstream with its source. Deciding which three issues the committee should spend its time on is still a partner’s call. Kepler drafts the complete register and leaves the ranking to the partner.

Step by step: drafting the memo with AI

1. Start from the template and the memos the committee liked. Section order, required tables, length limits and tone come from the firm’s own examples. Generic memo prose (“a leading provider with a strong market position”) is what committees learn to skip.

2. Load the deal record. The CIM, management presentation, QoE, the model, legal and commercial diligence reports, call notes, the Q&A log and the term sheet. The memo can only cite what the tool has read.

3. Draft the factual sections first. Overview, financials, QoE summary, financing and diligence findings. Every figure carries its source.

4. Pull returns from the model. Returns should come from the model’s cells, never recomputed in prose. A made-up example, gross and before fees:

Case	Exit EBITDA ($M)	Exit multiple	Exit EV ($M)	Net debt at exit ($M)	Equity value ($M)	MOIC	IRR (5 years)
Downside	50.0	9.0x	450	190	260	1.08x	1.6%
Base	62.0	11.0x	682	150	532	2.22x	17.3%
Upside	70.0	12.0x	840	130	710	2.96x	24.2%

Entry is 12.0x $40.0 million of adjusted EBITDA, a $480 million enterprise value funded with $240 million of equity. Each figure in the table should open to its cell in the model, and each model input to the document behind it.

5. Reconcile the sources that disagree. In a typical deal, the CIM shows adjusted EBITDA of $44.1 million, the QoE $40.6 million, and the model uses $40.0 million on the firm’s own definition. The memo should show all three, explain the differences line by line, and say which figure the price rests on. A memo that silently picks one leaves the deal team defending a number the committee never saw challenged. Kepler builds this reconciliation in code, one row per difference, each citing the CIM page, QoE schedule or model cell behind it.

6. Write the risk section with evidence. Each risk gets a source, an estimate of impact where possible, and a mitigant. “Customer concentration” becomes “the top customer is 14% of revenue, with a contract that renews in 2028 (master agreement, section 12.1); mitigant: a three-year extension is a condition of signing.”

7. Partners write the thesis and the recommendation. With the facts drafted and cited, the partner’s time goes to the argument.

8. Check before it goes out. Click through the figures the recommendation depends on. Ask the tool to list every figure that changed since the last draft. Our guide on how to check AI-generated numbers covers the spot checks.

Citations: answering “where’s that from?”

Committees ask where numbers come from, and the honest answer is often three clicks deep: a figure in the memo, a cell in the model, a line in the QoE. Every figure in the memo should link to its source, and every derived figure should show its formula and inputs. That matters most for the numbers the price depends on: adjusted EBITDA, the entry multiple, net debt and the exit assumptions. It also matters later: when a deal underperforms, the memo is the record of what the firm knew and when.

On Kepler, every number in the memo is pulled from a source or computed in code over cited inputs, and every citation is checked against its source while the draft is written.

House style as a template and a playbook

A template sets the format. A playbook sets the procedure: which sections to draft from which sources, which definitions to use, the firm’s risk categories, the tables every memo must have, and the length of each section. Write it down once:

• Section order and length limits
• Required tables: sources and uses, returns by case, EBITDA reconciliation, comps
• The firm’s definitions of adjusted EBITDA, net debt and net revenue retention
• Risk categories the committee expects, in its order
• Tone rules: no adjectives the evidence doesn’t support, numbers before adjectives
• What the preliminary memo includes and what waits for the final

Partner edits are the most valuable input. When a partner rewrites a section the same way twice, that’s a rule for the playbook. On Kepler, corrections become labels that improve the firm’s own models privately, so the next deal’s draft needs fewer edits.

Version control as diligence moves

Numbers change between the preliminary and the final memo, and committees want to know which and why.

Figure	Preliminary memo	Final memo	Why it changed
Adjusted EBITDA	$41.8M	$40.0M	QoE removed a pricing add-back
Net revenue retention	108%	105%	Rebuilt from the customer revenue file
Entry multiple	12.5x	12.0x	Price revised after QoE
Base case IRR	19.0%	17.3%	Lower entry EBITDA, lower multiple

Keep each draft, record each change with its new source, and include a short “changes since the preliminary memo” table in the final. Building that table automatically requires a tool that knows which document each figure came from, which is how Kepler produces it.

Where AI-drafted memos go wrong

• Numbers the model wrote. Rounded, transposed or from the wrong period, and out of step with the model.
• Stale figures. A number from the preliminary memo survives into the final after the QoE changed it.
• Generic prose. Descriptions that would fit any company in the sector.
• The CIM taken at face value. The banker’s adjusted EBITDA reported as the company’s. See how to review a CIM with AI.
• Silent conflicts. One of several disagreeing sources chosen without saying so.
• Confidentiality. Deal documents under NDA in a tool without deal walls or logs.

Several tools aim at the memo. Anthropic’s Claude for Financial Services page lists IC memos among its uses for asset management teams, Hebbia’s investing page lists compiling comps and positioning for IC memos, and V7’s private equity page lists a data-room-to-IC-memo workflow. For a comparison, see the best AI tools for private equity due diligence. For more on why models invent figures, read AI hallucinations in finance.

How Kepler does this

The memo is the last document in a deal and the one that depends on all the others, so Kepler drafts it from the same record the deal team has worked in since the CIM arrived.

Long before a target shows up, our engineers build your firm’s ontology from your past memos, models, CRM notes and portfolio reporting. When a new deal starts, its company, management, competitors, lenders and advisers resolve to records the firm already holds, and the memo can cite what the firm learned on earlier deals in the sector.

By the time drafting starts, extraction models we train have read the CIM, the QoE, workstream reports, call notes and every data-room file, with each figure kept beside its page. Your template, section rules, definitions and risk categories run as a saved playbook. Kepler pulls figures from their sources and computes the rest in code (the EBITDA reconciliation, returns by case, comps, sources and uses), so each number in the draft opens to its formula and inputs. Where the CIM, the QoE and the model disagree, the draft shows each value with its source and flags the conflict for the deal team to settle. The memo fills in Word, returns and comps in Excel, and the committee deck in PowerPoint.

As diligence moves, Kepler’s agents re-read each new upload overnight and flag any memo figure it changes. They work inside the deal team’s wall with no more access than the person who launched them, and a seller’s document can’t tell them to do anything. Kepler redrafts and builds the “changes since the preliminary memo” table, one row per figure with its new source. Partner edits become labels that improve your firm’s own models privately.

On committee day, a partner can click any figure, in the app or from inside Claude or ChatGPT, and see where it came from, and every read and query is logged (security). Our engineers built this layer before at Palantir and work alongside your deal team on site while the first memos run.

Try it on the next deal:

• “Draft the final IC memo in our format and list every figure that changed since the preliminary memo, with the new source.”
• “Reconcile adjusted EBITDA across the CIM, QoE and our model, line by line.”
• “Build the risk register from every workstream report, with the source for each risk.”

Book a demo to see Kepler draft a memo from a data room in your house format, or start on app.kepler.ai.

Frequently asked questions

Can AI write an investment committee memo?

Yes, most of it. Kepler, which we build, drafts the memo from the deal's own documents as a saved playbook in house style: the company overview, market, historical financials, quality of earnings summary, returns tables, financing terms, diligence findings and a first risk register, with every number computed in code and traced to its page or cell. Partners write the thesis, the recommendation and the judgment on risks.

What sections go in an investment committee memo?

A typical private equity IC memo opens with the recommendation and the ask, then covers the investment thesis, company overview, market and competition, historical financials, adjusted EBITDA and quality of earnings, the entry price and comps, returns under base, upside and downside cases, financing structure, key risks and mitigants, diligence findings and open items, and the value creation plan, with supporting detail in appendices.

How long should an IC memo be?

It depends on the stage and the firm. A first-round memo that asks for approval to spend on diligence is often a few pages. A final memo for a binding bid is usually much longer, with the detail in appendices. Most committees prefer the main body to stand on its own, so a partner can read the recommendation, thesis, returns and risks without opening an appendix.

How do you make sure the numbers in an AI-drafted memo are right?

Require a source for every figure: a page in the CIM or QoE, a cell in the model, a line in the management accounts. Derived figures should show their formula and inputs. Where sources disagree, the memo should show each value and say which one the price uses. Then spot-check the figures the recommendation depends on before the memo goes to committee.

How do you keep an AI-drafted memo in house style?

Give the tool your template and two or three past memos the committee liked, plus written rules: section order, required tables, length limits, the firm's risk categories, its definitions and its tone. Saved once as a playbook, the rules apply to every deal. Partner edits on each draft should feed back into the rules so the next memo needs fewer.

What is the difference between a preliminary and a final IC memo?

The preliminary memo comes early, usually after the CIM and first management meeting, and asks for approval to bid and spend on diligence. The final memo comes before a binding offer, after quality of earnings, legal and commercial work, and asks for approval of the price, structure and financing. Figures often change between the two, so the final memo should show what changed and why.