All workflows
Research

Deal and investment memo generation

Structured investment memo from 20-50 source documents - CIM, financials, management decks, diligence notes - with citation-backed sections.

Investment memo generation on a DealCloud + PitchBook + Box stack: $5.45/run, 58% hidden in revision re-injection and connector pulls. Firedog tracks cost per revision across PE deal workflows.

Model 32%Prompt 10%Orchestration 10%Retrieval 4%History 32%Connectors 12%
ComponentWhy it is unattributedcalls$ / run
Model · 32%Long-form synthesis (Claude Opus 4)The model ingests a stitched document context and produces a 4,000-8,000 token structured memo with section-level citations. Opus 4 is used for hallucination risk tolerance reasons.tokens in: 85,000 · tokens out: 6,500Visible on the invoice.3$1.76
Prompt · 10%IC memo template and section schemaA 12,000-token prompt encodes section headings, financial model hooks, risk taxonomy, and 4 full few-shot examples. Sent on every call without caching.tokens in: 36,000 · tokens out: 0Visible on the invoice.3$0.54
Orchestration · 10%Document ingestion and section routing agentsAn ingestion agent chunks and classifies source documents; a routing agent dispatches chunks to section-specific sub-calls; a merge agent assembles the output.tokens in: 22,000 · tokens out: 3,000Document ingestion and merge agents are instrumented as 'data pipeline' steps in engineering logs, not as LLM spend in the finance system.9$0.56
Retrieval · 4%Source document chunk retrievalEach section call retrieves the top-12 relevant chunks from the source document index. Re-embedding runs on every revision cycle even when source docs have not changed.tokens in: 42,000 · tokens out: 0Re-embedding unchanged source docs on revision runs is a known inefficiency; the cost is buried in the embedding vendor line rather than the deal P&L.18$0.20
History · 32%Prior memo draft re-injection on revisionsRevision runs prepend the full prior memo draft to the model call context so the model can diff its own output. A 5,000-token draft becomes 25,000 tokens re-billed across 5 revisions.tokens in: 48,000 · tokens out: 0Revision-cycle context is framed as 'draft history' by the application; it appears nowhere in the per-run cost breakdown visible to finance or risk.3$1.73
Connectors · 12%DealCloud, PitchBook, Box, and Salesforce connectorsThe run pulls source documents from the Box data room, comps from PitchBook, and sourcing context from Salesforce, then writes the completed memo and IC status back to the DealCloud deal record. Each external pull is normalized into memo-section inputs by an LLM mapping step, and revision runs re-pull unless cached.tokens in: 22,000 · tokens out: 1,600The PitchBook pull is on the data-vendor contract, the Box data-room reads on the storage tier, and the DealCloud write-back on the deal-CRM plan; none is attributed to the deal's memo P&L.10$0.66
An investment memo costs ~$5.45 to generate; revision cycles mean 58% of the $24.5k/yr total is prior-draft re-injection plus DealCloud/PitchBook/Box connector pulls repeated on every edit nobody is tracking.
$5.45
cost per run
$25k
per year
58%
unattributed today
$16.35
per user, per day
Scenario: Mid-market PE fund, 3 active deals. 6 active users, 3 runs/day, 250 working days. 35 documents/run. Avg doc: 12,000 tokens.

Default model: claude-opus-4

Trigger: New deal entry to IC pipeline; typically 1-4 memos per analyst per month. Revision cycles add 2-5 additional runs per memo.

Who runs it: PE and VC investment teams, M&A desks at companies. Memo generation is a high-value, low-frequency workflow with high tolerance for cost but zero tolerance for hallucination. Memos are triggered from the DealCloud IC pipeline, built on Box data-room documents and PitchBook comps with Salesforce sourcing context, so every run and revision round-trips through that stack.

What Firedog shows

  • Per-run cost by memo section, showing which revision run triggered the largest context spike.
  • How much of each run's token spend is the prior-draft re-injection versus new synthesis.
  • Dollar cost of enabling prompt caching on the IC memo template across 250 working days.
  • How much of each memo run is Box data-room pulls, PitchBook comps, and DealCloud write-back, and how revision runs re-pull unchanged sources.

Decisions this informs

  • Store prior drafts as structured section diffs rather than full-text re-injection to cap revision overhead.
  • Cache the 12,000-token IC memo template prefix; at 3 calls/run and $15/M input this saves roughly $0.54 per run.
  • Flag revision runs for incremental embedding updates; re-embedding unchanged source docs on every revision is the largest single waste line.
  • Cache Box data-room and PitchBook pulls for the life of a deal so revision runs don't re-fetch and re-normalize unchanged source material.

Connected systems

  • DealCloud · Deal / CRM · Intapp (DealCloud)
    Memo generation is triggered by a new deal entering the IC pipeline in DealCloud; the finished memo and IC status are written back to that deal record.
  • PitchBook · Market data · Morningstar (PitchBook)
    Comparable transactions, multiples, and market data are pulled from PitchBook into the memo's market section.
  • Box · Document storage
    The CIM, financials, management decks, and diligence notes are read from the deal's secured Box data room.
  • Salesforce · CRM
    Relationship and sourcing context is pulled from Salesforce to populate the deal-origination section.

Understand every AI dollar.

Book a demo to see how Firedog attributes every AI request across your organization, and what your leadership team can do with the data.