The AI-Ready Investment Committee

1. The IC Is Where AI Meets Judgment

Every other part of the firm can experiment with AI quietly. The investment committee cannot. It is the place where a recommendation becomes a decision, and where the firm's judgment is exercised on the record.

So when AI starts touching the memo before the committee reads it, and it already is at most firms, the committee is the part of the firm that has to change how it works. Not because AI is dangerous, but because the thing the committee was built to review, a document assembled by a person, is now partly assembled by a tool that is fluent, fast, and occasionally confidently wrong.

The good news is that the committee already has the right instinct: trust nothing you have not checked. AI does not require a new philosophy. It requires applying the old one to a new kind of input, deliberately, before a well-formatted error walks through the door unchallenged.

2. The New Risk: Confident, Wrong, and Well-Formatted

The old bad memo announced itself. Thin analysis, clumsy prose, gaps you could see. The reviewer's discomfort was a useful signal, and the committee learned to read it.

An AI-assisted memo removes that signal. It is clean, structured, and confident on every page, including the pages where it is wrong. The model does not hedge when it is uncertain. It writes a fabricated figure in the same calm voice it uses for a sourced one, and a reasonable-sounding comparable that does not exist reads exactly like one that does.

This is the specific danger for an IC. Polish is a form of social proof, and a polished document quietly lowers the reader's guard at the precise moment it should be raised. The error is not that AI makes mistakes. Every analyst makes mistakes. The error is that AI makes them in a register that discourages checking.

So the committee's job is to decouple two things that used to travel together: how good a memo looks and how much it has been verified. Under AI they are no longer correlated, and treating them as if they are is the single most expensive habit an IC can keep.

3. What the Committee Should Disclose

The committee should know whether AI was used in the work in front of it, and where. Not to police it, to read it correctly.

This is light. It is not a forensic log of every prompt. It is a line in the memo or a norm in the room: was AI used in producing this, and on which parts. A model that drafted the industry overview is a very different input from a model that produced the revenue build, and the committee reads each with the right level of scrutiny only if it knows which is which.

The goal is not suspicion, it is calibration. A reviewer who knows the comparables were assembled by a tool checks the comparables. A reviewer who assumes a person did it by hand may not. Disclosure simply points the committee's existing skepticism at the right paragraphs.

Make it a habit, not an interrogation. The firms that get this right treat AI use as normal and disclosed, the way you would note that a figure came from a third-party data provider rather than the company. It is context, and context is what lets a committee do its job. The mechanics of standardizing this in the pack itself are covered in the IC memo and board pack guide.

4. What to Verify, Always

Three things get verified every time, regardless of how good the memo looks: the numbers, the sources, and the load-bearing claims.

The numbers. Any figure the decision rests on traces back to a primary source the committee can see. A model can carry a number forward, transpose it, or invent one that fits the narrative. The market size, the growth rate, the multiple, the covenant headroom: these are checked against the document they came from, not the memo that quotes them.

The sources. A citation is not evidence that the source exists. Models will produce a plausible reference to a report, a precedent, or a comparable that is not real, or that is real but says something different. If a claim leans on a source, someone opens the source.

The load-bearing claims. Every memo has two or three assertions the entire thesis hangs on. Those get scrutinized hardest, because that is where a confident, wrong sentence does the most damage. The fluent paragraph that sounds obviously true is exactly the one to slow down on.

This is not new work for a serious committee. It is the work it always did, made non-negotiable. The shift is removing the unspoken exception, "the memo looks thorough, so we can trust this part," because that exception is precisely the gap an AI error fits through.

5. The Four IC Standards

An AI-ready committee can be summed up in four standards. None is technical. All of them are about how the room treats the work.

Write these on one page and they become the committee's operating rule for AI. They are simple on purpose. A standard the room cannot remember under time pressure, on the morning of a live decision, is not a standard.

6. The Memo Looks Better. Hold the Bar Higher

The most dangerous sentence in an AI-ready firm is "the packs are so much better now." They are. That is the problem.

Better-looking packs create a quiet pressure to relax. The analysis arrives cleaner, the formatting is sharper, the prose is tighter, and the committee starts to feel the work has already been done to a higher standard. Some of it has. The thinking has not, and the thinking is the part the committee is there to test.

So the rule is counterintuitive: as the memo gets more polished, the committee's scrutiny holds or rises, it does not fall. Treat fluency as a reason to check, not a reason to trust. The cleaner the comparable, the more it deserves a look, because clean and wrong is the failure mode AI introduced.

This is a cultural choice as much as a procedural one. A committee that lets polish stand in for proof will make worse decisions with better-looking documents, which is the exact opposite of what the firm bought the tools for.

7. Where AI Genuinely Helps the IC

None of this is an argument against AI in the committee process. Used well, it makes the IC better in three concrete ways.

Faster, fuller packs. The assembly that used to eat an analyst's week, the overview, the formatting, the first-pass financial summary, comes together faster, which means the committee can see more complete material earlier and spend its time on the decision rather than waiting on the document.

A tireless devil's advocate. Ask a model to argue the other side and it never gets tired, political, or attached to the deal. It will surface the bear case, list what would have to be true, and name the risks the team fell in love past. The committee still judges. The blind spots get smaller.

Consistency across deals. A model applies the same checklist to the tenth memo as the first. It does not cut corners on a Friday or a deal everyone already likes. That consistency is genuinely useful to a committee trying to compare opportunities on the same terms.

The pattern is the same as everywhere else AI works at a firm. It does the assembly and the stress-testing. The judgment stays human, and the committee that uses AI to sharpen the debate, not shorten it, gets the upside without the risk. The automation side of producing those packs is covered in the IC memo automation solution.

8. Questions a Good IC Now Asks

A few questions belong in the room now that did not need to be there before. They are short, and they change behavior upstream the moment people know they are coming.

Where did this number come from. Asked about any figure the decision rests on, and answered with a source, not a memory. Has anyone opened the sources. A direct question that quietly ends the era of uninspected citations. Which parts did AI help produce. The disclosure question, asked plainly. What does the bear case say. Best asked of the team and, increasingly, of the model that argued the other side.

The point of these questions is not to catch anyone. It is that the questions shape the work before it arrives. An analyst who knows the committee will ask who opened the sources, opens the sources. The room's standard becomes the firm's standard, which is how a committee has always set the tone for the work beneath it.

The same instinct sits behind the broader leadership conversation in the partner AI briefing: the people at the top set the bar by what they ask for.

9. Training the Committee

An IC does not need a course. It needs ninety minutes and a shared understanding, focused on decisions, not tools.

The committee does not have to learn to prompt, build, or operate anything. The members need to know one thing well: how AI-assisted work fails, so they can review it correctly. Shown a handful of real examples, a confident fabricated comparable, a transposed figure, a citation to a report that does not exist, an experienced committee recalibrates fast. They have spent careers spotting work that does not hold up. This is the same skill, pointed at a new source of error.

Keep it short and keep it concrete. One session, real artifacts from the firm's own deals where possible, and the four standards on a single page. That is enough to change how the room reads a pack, which is the entire goal. The deeper, hands-on training belongs with the analysts producing the work, not the committee reviewing it, and that split is the heart of change management at an investment firm.

This is exactly the scope of a Partner and IC Review: a focused session for the people who review the work, not the people who build it, built around how AI-assisted output fails and what to demand of it.

10. Where to Start

Adopt the four standards at the next meeting. Disclosure, source-checking, author accountability, judgment unchanged. Put them on one page and read them into the room. That alone changes how the committee treats the next pack.

Then add the questions: where did this number come from, has anyone opened the sources, which parts did AI help produce, what does the bear case say. The questions do the upstream work, shaping the memos before they reach the table.

If you want the committee aligned in a single focused session, that is a Partner and IC Review, built around how AI-assisted work fails and the standards that keep the decision honest. We can run the IC process with you over time as an AI Operating Partner, and where it helps, automate the pack itself with IC memo automation, so the committee gets fuller material and spends its time on the decision. The firm-wide version of all of this sits inside an AI strategy and roadmap.