Why IC memos still start from scattered notes
Investment committee memos are the artifact that turns weeks of research into a yes, no, or pass-with-conditions decision. Yet most investment diligence memo AI workflows still begin the same way: a partner pastes founder deck bullets into a doc, an associate exports Crunchbase tabs, someone forwards three email threads, and a market map lives in a Google Sheet no one updates after the first read.
The thesis is in someone's head. The risks are in call notes buried in Notion. The one data point that would change the recommendation sits in a forwarded Gmail thread from a portfolio CEO who mentioned the same space six months ago.
A VC due diligence workspace fixes the plumbing: federated search across founder materials, market research, call notes, and email — with every thesis point backed by citations your partners can audit before IC. This is not generic chat that summarizes a PDF. It is an investor knowledge base where claims persist, evidence links back to source records, and the next deal in the same category starts informed instead of from zero.
IC memo structure: the sections partners actually read
IC formats vary by firm, but the sections that drive debate are predictable. Design your workspace around them so agents and analysts fill a stable template — not a blank page every time.
| Section | What IC needs | Typical sources |
|---|---|---|
| Company snapshot | Team, product, stage, raise terms, cap table highlights | Pitch deck, data room, founder emails |
| Market and timing | TAM/SAM, category definition, why now | Third-party research, expert calls, portfolio analogs |
| Traction and unit economics | Revenue, growth, retention, cohorts, burn | Financial uploads, management calls, references |
| Product and moat | Differentiation, technical depth, switching costs | Product demos, technical diligence, customer calls |
| Team and references | Founder track record, key hires, back-channel refs | LinkedIn, reference calls, prior firm notes |
| Risks and open questions | What could kill the deal; what is unverified | Diligence Q&A, email threads, competitor intel |
| Recommendation | Invest / pass / follow; check size; conditions | Partner synthesis with cited support |
A working IC memo research output might open like this:
Company: LatticeFlow — Series A infrastructure for ML observability; $8M raise at $32M pre
Thesis: Category is early but compounding — three portfolio companies hit model-drift incidents in Q1; buyer urgency is moving from "nice to have" to board-level risk
Traction: $1.2M ARR, 140% NDR on 18 customers — [financial upload, Mar 2026]; champion quote from reference call cites 40% reduction in incident MTTR — [call notes, Feb 28]
Key risk: GTM still founder-led; no VP Sales — [management call, Mar 5]; open question on enterprise security review cycle — [diligence tracker]
Each bullet should anchor to a source: the data room doc, email message, call note, or external research link that supports it. Partners learn to scan, click through when skeptical, and debate evidence — not memory.
Optional layers by deal type
- Seed: Weight team and market timing; lighter financials; emphasize reference quality and founder learning velocity.
- Growth: Lead with cohort retention, expansion, and capital efficiency; pull portfolio benchmarks from prior deals in the same category.
- Follow-on: Compare current metrics to last-round IC memo and board updates; surface what changed in thesis or risk.
- Competitive / strategic: Add incumbent response scenarios and M&A comparables from prior corp-dev or public-market research.
Publish one base template per stage. Let variants add two or three fields — not separate documents per partner.
Research ingestion: what to federate (and what to skip)
IC memo research fails when it only reads the pitch deck. Decks capture narrative; email, calls, and references capture verification. Diligence requires joining both.
Core sources for investor workflows
- Email (Gmail, Outlook): Founder threads, expert intros, reference follow-ups, internal partner forwards.
- Docs and data room uploads: Decks, financial models, cap tables, customer lists, security questionnaires.
- Call and meeting notes: Management sessions, customer references, expert networks — tagged by deal and date.
- Market research: Analyst reports, news, competitor profiles, prior firm memos on adjacent categories.
- CRM or deal pipeline tools (Affinity, Harmonic, etc.): Stage, owners, co-investor notes, relationship history.
- Portfolio updates: Quarterly board decks and email updates from companies in overlapping markets.
Nice-to-have sources
- Slack or internal comms: Partner discussion in deal channels, quick takes from portfolio CEOs.
- Public filings and transcripts: For later-stage or public comparables.
- Calendar: Diligence meeting cadence and who attended which call.
What not to dump into every memo section
- Full email archives or call transcripts — summarize with citations instead.
- Stale third-party TAM slides older than 12 months unless no newer research exists.
- Generic LLM market maps without links to your firm's prior work or primary sources.
Federation beats sync: query each system at memo-generation time so the output reflects yesterday's reference call, not last month's export. See How to Connect CRM, Slack, and Docs in One AI Workspace for the connector pattern — the same architecture applies when "CRM" is your deal pipeline and "docs" is the data room.
Thesis and risks: separate conviction from evidence
The highest-value IC sections are also the easiest to get wrong: thesis statements that sound confident but rest on unverified claims, and risk lists that repeat generic category worries without deal-specific proof.
Thesis pattern: claim, evidence, confidence
Structure each thesis bullet as three parts:
- Claim — "Buyer urgency is accelerating in ML observability."
- Evidence — Portfolio CEO email (Jan 14), two customer references citing board-level incidents, Gartner note on category spend.
- Confidence — High on urgency signal; medium on timing of enterprise budget allocation.
Agents can draft thesis bullets; partners own confidence ratings. The workspace should never blur "management said" with "we verified independently."
Risk pattern: severity, mitigant, open diligence
| Risk | Severity | Evidence | Mitigant / open question |
|---|---|---|---|
| Founder-led sales at $1M+ ARR | High | No VP Sales hire; 90% of pipeline from CEO calls | Reference customers confirm technical sale; open: can they hire a proven enterprise AE? |
| Crowded observability market | Medium | Three well-funded peers in last 18 months | Differentiation on model lineage vs log aggregation — [product call, Mar 1] |
| Customer concentration | High | Top two customers = 48% ARR — [financial upload] | Both expanding; request renewal terms in data room |
Risks without citations are brainstorming. Risks with anchors are actionable — either diligence closes the gap or the deal passes.
Multihop queries help connect dots search misses: traverse deal → reference contact → prior email from portfolio company in same vertical → return related incidents and quotes in one synthesis. That federated retrieval model is the same one described in Federated Search for Business AI.
Claims with anchors: trust is the product
Partners will not adopt investment diligence memo AI if they cannot verify claims before IC. Every synthesized sentence that affects the recommendation should trace to a source record.
Minimum citation bar
- Financial metrics (ARR, burn, runway) → data room upload or management-provided model with date stamp.
- Customer quotes and references → call notes or email with named contact and date.
- Market sizing → external research with publisher and date, or explicit "management estimate — unverified."
- Team and cap table facts → deck, LinkedIn, or legal docs in data room.
- Competitive claims → primary source (competitor site, filing, reference) — not paraphrase of paraphrase.
How citations should render
Partners need one-click drill-down: "140% NDR — [financial upload, Mar 2026]" not a footnote at the bottom of a 12-page doc. Gyri hydrates citations to the underlying Gmail thread, uploaded spreadsheet, or typed call note so reviewers confirm context in seconds.
This is the same trust model in AI Answers With Citations: Why Enterprise Teams Demand Proof, Not Vibes. IC memos are a high-stakes version: wrong traction intel costs capital and reputation, not just embarrassment.
Confidence and gaps
Strong memos say what they do not know:
- "Enterprise security review cycle unverified — no customer reference at Fortune 500 scale."
- "Market size figure from 2024 analyst report — request updated segment breakdown."
Explicit gaps beat silent omission. Deal leads use gap frequency to prioritize remaining diligence before IC date.
Collaboration model: partners, associates, and external experts
Diligence is multi-player. A VC due diligence workspace needs role-appropriate access and a clear handoff from research to recommendation.
Typical roles and permissions
| Role | Access | Primary actions |
|---|---|---|
| Deal lead / partner | Full deal workspace | Set thesis, approve memo, assign open questions |
| Associate / analyst | Ingest sources, draft sections | Upload docs, tag calls, run agent drafts |
| Operating / platform | Portfolio benchmarks, expert intros | Link prior deals, add category pattern insights |
| External counsel / specialists | Scoped data room + specific threads | Answer legal or technical workstreams |
Workspace-scoped auth matters: a specialist sees the security questionnaire and related email — not every other live deal in the fund.
Workflow: from first meeting to IC packet
- Open deal workspace — Create typed deal record; link pipeline CRM entry and founder email thread.
- Ingest sources — Upload deck and model; log management and reference calls with dates and attendees.
- Diligence question list — Maintain open questions as typed records; agents flag when new email or upload closes a item.
- Section drafts — Agents federate retrieval and fill template sections with citations; associates edit and add partner context.
- Partner review — Partner adjusts thesis confidence and risk severity; challenges weak citations before IC.
- IC packet export — Final memo plus appendix of anchored claims; persist as insight for follow-on and portfolio learning.
Agents that write back can log memo sections, update diligence trackers, and create insight records when a thesis point is validated or refuted — so the workspace compounds instead of resetting after each deal. See Agents That Write Back: From AI Drafts to CRM Updates and Insights for guardrails on automated updates.
MCP for interactive diligence
Some teams run deep dives from Claude Desktop or Cursor via MCP — same federated graph, follow-up questions ("What did the second reference say about churn?" or "Compare this ARR curve to our Series A infra portfolio."). One endpoint, multiple clients; see MCP for Business Agents: A Practical Guide for Operators.
Portfolio reuse: why the next deal in the category starts ahead
The hidden cost of scattered diligence notes is lost institutional memory. When a partner leaves, when an associate rotates, or when the same category heats up eighteen months later, the firm re-researches from scratch.
What to persist after IC (invest or pass)
- Category insights — Durable patterns: "ML observability buyers in 2026 prioritize lineage over log aggregation" with cited deal evidence.
- Competitor dossiers — Typed records updated when any live deal surfaces new signal.
- Benchmark library — Anonymized traction and unit-economic ranges by stage and category — linked to source deals with access controls.
- Pass reasons — Structured pass memos with citations so the firm does not re-underwrite the same weakness on a relaunch.
Pass memos are as valuable as invest memos for investor knowledge base quality. A founder who returns with new traction should not trigger amnesia about last round's gaps.
Cross-deal queries that compound
- "Show reference-call themes on enterprise security friction across infra portfolio companies."
- "Which live deals mention the same incumbent as LatticeFlow?"
- "Pull ARR at Series A for devtools companies we diligenced in the last 24 months."
Keyword search plus graph traversal makes these questions practical — not a quarterly associate project. The hybrid retrieval pattern is covered in Keyword Search Plus Graph: Why AI Agents Need Both.
Security and retention
Investor data is sensitive. Federated access with audit trails, workspace isolation per deal, and retention policies aligned to fund compliance are table stakes — not afterthoughts. External uploads and email should inherit the same citation and access model as internal notes.
Rollout checklist for investment teams
- Publish IC memo template by stage (seed, A, growth, follow-on) and get partner sign-off on required sections.
- Connect email and doc upload minimum; add pipeline CRM and call-note tagging in week two.
- Pilot on one live deal with a single associate–partner pair for 30 days.
- Enforce citation hard failures on financial and reference claims from day one.
- Persist IC outcomes and category insights for portfolio reuse — including pass memos.
- Expand MCP and agent automation after partner error rate on drafts stabilizes.
The bottom line
Investment diligence memo AI is not a deck summarizer. It is federated IC memo research — founder narrative plus verified references, market context, and internal history, cited so partners can audit before they allocate capital.
Gyri connects those sources into an agentic knowledge base: federated search, multihop graph queries, cited synthesis, and agents that persist thesis and risk insights across deals. If your firm still builds IC packets by hand from inboxes and folders, start your free trial and we will map the workflow to your diligence stack.