Insights · Guide
AI for the investment committee in private equity: what the IC actually needs
The investment committee is the highest-leverage decision surface in a private equity firm, and the surface where AI is applied most superficially. Here is the gap between the demo and the workflow.
Most 'AI for the IC' products are a chat window over the data room: a memo draft, a summarized CIM, an assistant that answers questions about the model. The IC meeting itself needs a decision, and the eighteen months of monitoring after it has little to do with summarization. What the IC actually needs is a record built from cited evidence, with the committee's conditions carried forward as monitored conditions instead of left in a meeting note.
What 'AI for the IC' usually means today
Walk into most AI-for-PE pitches and the demo looks the same: a chat window over the data room, a summarized CIM, a draft memo generated from a document corpus, an assistant that answers a question about the model. That is a useful productivity layer for the work before the IC meeting.
But the IC meeting itself does not need a chat window. It needs a decision. And the work that follows it, the quarterly reviews, the operator-update cycles where the team decides whether the case still holds, has little to do with summarization.
With $3.6T of unrealized value sitting on GP balance sheets and exit windows stretched, the cost of getting the post-IC decision wrong is higher than it has been.
What the IC actually needs from AI
The IC needs three things, and they are not what most vendors ship.
A record built from cited evidence
Not a draft of the memo, a memo whose figures trace to their source, cited to the page where a locator resolves.
Risk signals graded to the firm's own policy
Thresholds the firm sets, applied on each render, not a one-time summary.
Conditions carried forward
What the committee approved with conditions attached becomes a monitored condition at close, not a line in a meeting note.
Why building this is a structural problem
A chat window over the data room is LLM-first: it can summarize and answer questions, but it cannot guarantee that a figure in the answer traces back to a specific document and page unless the extraction underneath it is deterministic first. Most of the dominant AI-for-PE products were not built that way, and most operate inside the diligence window and stop at close.
That is not a prompt-engineering gap. It is an architectural choice about whether the memo is a typed record the firm can test later, or a document that gets filed.
A chat window over the data room
- Summarizes the CIM and answers questions about the model.
- Produces a memo draft as a document, not a typed record.
- Stops at close; the structured view does not carry into monitoring.
What the IC needs
- A memo whose figures trace to source, cited to the page where a locator resolves.
- Risk signals graded against the firm's own policy thresholds, re-graded on each render.
- Committee conditions recorded as monitored conditions that carry past close.
How Capital Refinery approaches the IC
Capital Refinery is not a chat window over the data room. It reads the documents the deal team already has and keeps the committee's decision connected to its evidence.
- Assembles memo sections from cited evidence across the spread; the narrative sections stay with the deal team.
- Records committee conditions as monitored conditions at close.
- Grades risk signals against the firm's own policy thresholds; move a threshold and the memo re-grades on the next render.
- Exports the memo as DOCX, with its evidence fingerprint.
Numbers are read by deterministic parsers. Where candidates conflict, a local model chooses among the named candidates; it does not generate the figures. Some written outputs, such as memo polish, are drafted by a hosted model from Anthropic, which receives deal details and summary figures.

See the structured record on a deal you already closed
Bring us a position from your portfolio. We will parse the diligence pack into a memo with its evidence cited, and show the layer your current AI-for-IC tool was not designed to produce.
The IC does not need a faster memo drafter. It needs a memo and a set of conditions that do not go stale.
Questions
What does 'AI for the IC' usually mean today?
Mostly a chat window over the data room: a summarized CIM, a draft memo generated from a document corpus, an assistant that answers model questions. It is useful before the IC meeting, and has little to do with what happens after it.
What does the investment committee actually need from AI?
A memo built from cited evidence, risk signals graded to the firm's own policy thresholds, and committee conditions that carry forward as monitored conditions instead of staying in a meeting note.
Why don't most AI-for-PE vendors ship a structured decision record?
Most are built LLM-first, which cannot guarantee a figure traces to a specific document and page, and most operate inside the diligence window and stop at close. Building the record requires a different, deterministic-first architecture.
Does Capital Refinery write the IC memo's narrative?
No. Capital Refinery assembles memo sections from cited evidence; the narrative sections stay with the deal team.
Does Capital Refinery use AI to decide what the committee approves?
No. Capital Refinery grades risk signals against the firm's own policy thresholds and records what the committee decides. The decision stays with the committee.