How Investment Committees Decide (and Where the Process Goes Wrong)

Every organization that deploys capital has some version of a committee that makes the final call. A fund has its partners, an angel network has an investment committee drawn from its members, an accelerator has a selection panel. Enormous effort goes into what feeds that committee: sourcing, screening, diligence, the memo. Comparatively little goes into the design of the decision itself, which is treated as a procedural formality rather than as something that shapes outcomes.

That is a mistake, and a well-documented one. The order in which people speak, whether opinions are collected before or after discussion, who holds a veto, and how the vote is counted all change which companies get funded. Two committees with identical members, looking at identical companies, will produce different portfolios depending on how the meeting is run.

The Voting Rule Is a Strategy Choice

Start with the mechanics, because they vary far more than most operators assume, and each variation embeds a different appetite for risk. A survey conducted by academics from Boston College, MIT Sloan, and Imperial College London gathered data from 35 of the 55 largest US venture firms and found that voting rules shift systematically by stage: 90 percent of respondents use some version of a champion rule for seed investments, where a single partner with conviction can carry a deal, and a third of those use a no-veto champion rule, where one partner can proceed alone provided nobody else blocks it. At Series A, only about 20 percent still use a champion rule, rising to roughly half once the no-veto variant is included.¹

The logic behind that shift is sound. At seed, the companies that turn out to matter often look strange, and a rule requiring broad agreement filters out the non-obvious bets that produce outsized returns. Requiring conviction from one person rather than comfort from everyone is a deliberate choice to tolerate more variance in exchange for access to outliers. At later stages, with more evidence available and larger checks at stake, the calculus reverses and consensus becomes the more sensible protection.

Smaller organizations often default to unanimity without recognizing what they are choosing. Unanimous-vote partnerships kill more deals to friction, which does protect against weak picks, but the same friction filters out contrarian winners.² Practice across the industry runs the full range: some committees require a majority, others effectively require no strong objections, others allow a single respected member to carry a deal even when colleagues are lukewarm, and some allow any member to veto.³ None of these is correct in the abstract. What matters is that the rule is chosen on purpose, with its tradeoff understood, rather than inherited from whoever set up the group.

What Goes Wrong Once Discussion Starts

The voting rule determines how the decision is counted. The discussion determines what gets counted, and this is where most of the damage happens.

Several failure patterns are consistent enough to be predictable. Anchoring means the first piece of information encountered carries disproportionate weight, so an opening valuation or a market-size figure quietly frames every judgment that follows. Similarity bias means reviewers favor founders whose background, education, and experience resemble their own, whether or not those traits predict anything. And information overload cuts in a direction that surprises people: research finds that the more information an investor receives, the more confident they become in their decision, which is not the same as being more accurate. Some reviewers keep requesting additional data indefinitely, hoping the data will make the decision for them.⁴

Then there is the group effect, which is the one specific to committees rather than individuals. Once the first person states a view, especially a senior person, subsequent opinions cluster around it. This is not weakness of character; it is a well-replicated finding about how people behave in groups, going back to the conformity experiments of the 1950s. Committee members become reluctant to voice disagreement even when they have solid grounds for it, and the group converges on an apparent unanimity that never existed as independent judgment. The result looks like agreement and functions like an information cascade.

The practical consequence is that a committee can be full of capable, experienced people and still produce a decision no better than the opinion of whoever spoke first, plus some noise. All the independent expertise around the table gets spent on agreeing with the opening position rather than testing it.

The Noise Problem Underneath the Bias Problem

There is a second, less discussed problem, and it is not bias. When the same set of cases was given to fifty insurance underwriters at the same firm, each asked to put a value on them, the firm expected variation of roughly 10 percent between assessments. What the audit found was 55 percent.⁵ Not one underwriter being wrong in a consistent direction, which is bias, but wide, unpredictable disagreement among professionals doing the same job with the same information. Daniel Kahneman and his co-authors gave this unwanted variability a name, noise, and made an argument that lands uncomfortably for anyone who prizes individual judgment: judgment is not the place to express your individuality.

Investment assessment is unusually exposed to this. The inputs are ambiguous, the outcomes take years to arrive, and feedback is so delayed and so confounded that reviewers rarely learn how well calibrated they are. A network reviewing the same company through two different sets of members can easily produce two incompatible verdicts, and nothing in the process would surface that inconsistency, because the second verdict never gets made.

The Fix Is Structural, Not a Matter of Trying Harder

Knowing about anchoring does not stop you from anchoring, and telling a committee to be more objective produces nothing. What does work is changing the sequence of the meeting, which is a design decision rather than a discipline problem.

The core principle is to collect independent judgments before allowing discussion. The procedure sometimes called estimate-talk-estimate captures it: each member privately records an assessment first, then members explain and defend those assessments, then each records a revised assessment, and the second round is aggregated into the decision. Because everyone starts from an independent position, the information cascade never forms, while the discussion step still lets the group learn from each other and pulls outlier views toward the middle.⁶

Two further principles matter. Break the assessment into specific dimensions and score each one separately rather than forming a single overall impression, keeping each judgment as fact-based and as independent of the others as possible. And delay intuition rather than banning it: the goal is not to remove judgment from the process but to prevent it from arriving before the evidence has been examined. Aggregating judgments that are both independent and diverse is described by the authors of that work as the easiest, cheapest, and most broadly applicable of these measures.⁷ Which means the highest-return change available to most committees is also close to free: ask people to write down their assessment before anyone speaks.

Why Networks Have an Advantage Here, and Usually Waste It

An angel network or an accelerator selection panel has a structural advantage over a small fund partnership: more reviewers, from more varied backgrounds, than four partners who have worked together for a decade and increasingly think alike. Diversity of judgment is the raw material that makes aggregation work, and a network has more of it available than almost any fund.

That advantage evaporates the moment the process collects opinions in a live discussion. Twenty reviewers who anchor on the first confident opinion produce less usable signal than four partners who each wrote down an independent view first. Worse, the larger the group, the stronger the pressure toward conformity and the more reluctant a quieter member is to be the sole dissenting voice. A network that runs its committee as an open conversation is taking its main advantage and converting it into a liability.

There is also a member-engagement dimension specific to networks. Reviewers who submit a considered independent assessment and then see how it compared with the group are doing something more substantial than attending a meeting, and they tend to stay engaged. Reviewers whose role is to listen while two senior members debate learn quickly that their presence is decorative.

The Practical Version

None of this requires a rebuild. Four changes cover most of the available improvement. Collect written assessments before the meeting, not during it, with reviewers scoring specific dimensions rather than giving an overall verdict. Withhold those assessments from the group until they are all in, so nobody's submission is influenced by seeing someone else's. Choose the voting rule deliberately, matched to stage and check size, and write it down so it stops being renegotiated case by case. And keep a record of who assessed what and how it turned out, since without that record nobody ever finds out which reviewers are well calibrated and which are guessing confidently.

The last one takes years to pay off and is the one most organizations skip, which is why so few committees have any evidence about their own accuracy. It is also the only way the process improves rather than simply repeating.

Where This Connects to Pynn

The reason this problem persists is rarely that operators disagree with it. It is that running a structured, independence-preserving assessment process by hand, across a distributed group of reviewers, on every company in the pipeline, is more administrative work than a small team can sustain. So the process reverts to a meeting and a conversation, which is the version that requires no infrastructure.

That is the gap a system of record should close. Pynn assesses incoming companies against a network's stated thesis and holds deal flow in one place, which means the evidence a reviewer needs arrives structured rather than as a deck to skim, and the assessment of a company can be captured and stored per reviewer rather than reconstructed from memory after a discussion. The principle worth holding onto, whatever tooling a network uses, is that independent judgment has to be captured before the group talks. Once the conversation starts, the information it would have contained is gone.

Sources

1. The VC Factory, Venture Capital Investment Committees: Best Practices From Elite VC Firms (citing the Boston College, MIT Sloan and Imperial College London survey of 35 of the 55 largest US VC firms). https://thevcfactory.com/investment-committees/

2. Auryn, Investment Committee (VC Glossary). https://www.auryn.vc/glossary/investment-committee

3. Khoroshylova, Inside the Black Box: Understanding Organizational Decision Making Processes in VC Funds, Strategic Change (Wiley), 2026. https://onlinelibrary.wiley.com/doi/10.1002/jsc.70074

4. Antler, The Elephant in the Room: The Role of Unconscious Bias in Venture Capital Decision Making. https://www.antler.co/blog/the-elephant-in-the-room-the-role-of-unconscious-bias-in-venture-capital-decision-making

5. George Roussos, Quality Judgment and Decision Hygiene (summarizing the noise audit reported in Kahneman, Sibony and Sunstein, Noise: A Flaw in Human Judgment). https://www.linkedin.com/pulse/quality-judgment-decision-hygiene-george-roussos

6. Shortform, Decision Hygiene: Principles and Best Practices (summarizing Noise: A Flaw in Human Judgment). https://www.shortform.com/blog/decision-hygiene/

7. Behavioral Scientist, A Conversation with Daniel Kahneman About Noise. https://behavioralscientist.org/a-conversation-with-daniel-kahneman-about-noise/