Decision Hygiene for Deal Teams: Less Noise and Bias in IC

Bias versus noise in deal decisions

An investment committee process can go wrong in two different ways, and the fixes differ.

Bias is error in a consistent direction. If every reviewer on a deal team overweights the first revenue figure they saw, the team's judgments lean the same way, and averaging more opinions does not cancel the lean.

Noise is scatter. The publisher's description of Noise by Daniel Kahneman, Olivier Sibony and Cass R. Sunstein defines it as "variability in judgments that should be identical," with examples such as two doctors in the same city giving different diagnoses to identical patients, or the same person deciding differently "depending on whether it is morning or afternoon" ^[1].

A 2016 Harvard Business Review article by Kahneman, Andrew M. Rosenfield, Linnea Gandhi and Tom Blaser opens with a business case. At a global financial services firm the authors worked with, a longtime customer accidentally submitted the same application file to two offices; the employees were supposed to follow the same guidelines, yet the offices "returned very different quotes," and the customer took the business to a competitor ^[2]. HBR's published description of the article explains how organizations can perform a "noise audit": members of a professional unit evaluate a common set of cases, and the degree to which their assessments vary provides the measure of noise ^[3].

Translate that to origination. If two associates read the same target's filings and one scores it a strong fit while the other passes, that is noise: the target did not change, the reader did. If both associates consistently rate founder-led businesses higher than the evidence supports, the firm has bias. A screening process should be designed to detect both, and the two require different tools:

Known biases that hit screening

Anchoring. Amos Tversky and Daniel Kahneman's 1974 paper in Science described three heuristics people use when judging under uncertainty: representativeness, availability, and "adjustment from an anchor, which is usually employed in numerical prediction when a relevant value is available." The abstract states that these heuristics "are highly economical and usually effective, but they lead to systematic and predictable errors" ^[4]. In screening, the relevant value is often a revenue estimate from a data vendor, an asking multiple in a teaser, or the first number in the associate's memo. Later numbers get read as adjustments to it.

Optimistic forecasts. Dan Lovallo and Daniel Kahneman's 2003 HBR article opens with Oxford Health Plans, which began building a claims and payments system in 1992; when the company disclosed on October 27, 1997 that its system and accounts were in disarray, its stock price "dropped 63%, destroying more than $3 billion in shareholder value in a single day" ^[5]. The article's summary attributes overly optimistic investment forecasts to "a combination of cognitive biases (including anchoring and competitor neglect) and organizational pressures" ^[5]. A deal team that has spent weeks on a target is under exactly that kind of pressure.

Confirmation and loss aversion. The 2011 HBR article by Kahneman, Lovallo and Sibony gives short working definitions: confirmation bias "leads people to ignore evidence that contradicts their preconceived notions," anchoring causes people "to weigh one piece of information too heavily," and loss aversion "makes them too cautious." The same opening adds that, in the authors' experience, awareness of biases "has done little to improve the quality of business decisions" ^[6].

That last point is the case for process over training. Telling a committee about anchoring does not stop anchoring. Changing the order in which information arrives can.

Where these show up in a typical origination funnel:

Independent assessments before discussion

The single most practical step is to collect judgments before people talk. Once a senior partner says "I like this one," every later score is partly a response to that statement.

A workable protocol:

1. Freeze the evidence packet. Every reviewer gets the same retained pages, filings and notes for each target. If one reviewer has extra information, disagreement may reflect unequal evidence rather than inconsistent judgment.
2. Score alone, in writing. Each reviewer scores each criterion on the shared screening rubric and writes one sentence citing the evidence behind each score. No overall score yet.
3. Submit before seeing others. Scores become visible only after everyone has submitted.
4. Compare by criterion, not by total. Look for criteria where reviewers differ by two or more points. Those are either ambiguous rubric language or genuinely borderline evidence.
5. Discuss the gaps, then rescore. Keep the original scores. The record of disagreement is how you improve the rubric.
6. Track agreement over time. Measure inter-rater reliability on a recurring sample. This is the noise audit idea from the 2016 HBR article, applied to screening: a common set of cases, several reviewers, and a measure of how much they vary ^[2].

Illustrative example (fictional). Three reviewers score three fictional targets on a 1 to 5 scale before any discussion.

Company C's recurring revenue scores span 2 to 5. That spread is the signal. In the fictional discussion, R2 counted a one-time project list as recurring work, while R3 required a stated service contract. The fix is not a vote; it is a rubric definition of "recurring" that names the evidence required.

If the committee had discussed Company C first and heard R2's 5 aloud, the spread would probably never have surfaced.

Structure the rubric

Independent scoring only helps if the rubric gives reviewers the same target to aim at. Several drafting rules reduce noise directly:

• Define each criterion by observable evidence. "Strong recurring revenue" invites opinion. "The company publishes service agreements, maintenance plans or subscription pricing on its own site or in filings" can be checked.
• Anchor every scale point. Write what a 1, 3 and 5 look like for each criterion, with a short example. Unanchored scales are where within-person noise lives.
• Name the allowed sources. Company-published pages, filings and registry records behave differently from vendor estimates. State which count.
• Score criteria before any overall judgment. An overall impression formed first becomes an anchor for the components. Compute or record the overall view last.
• Separate "no" from "unknown." A missing fact is not a negative fact. Give "unknown" its own value so it is not silently scored as a 1 or a 3.
• Version the rubric. When a definition changes after a disagreement, record which version governed each score.

Our page on screening criteria covers building and weighting the scored dimensions.

Reference classes and the outside view

Lovallo and Kahneman's remedy for optimistic forecasts is what the HBR summary calls the "outside view," also known as reference-class forecasting. It "completely ignores the details of the project at hand"; managers instead examine the experiences of a class of similar projects, lay out a rough distribution of outcomes for that class, and then position the current project in that distribution ^[5].

For a deal team, the reference class should come from the firm's own records, not from a market statistic quoted in a pitch book. Prompts to add to the IC memo template:

• What is our class? For example, all add-on targets in this sector the firm took to LOI in the past several years, or all founder-owned targets above a size band the team has screened.
• What happened to that class? Of those, how many closed, how many broke in diligence, and for what stated reasons? Use the firm's own deal log.
• Where does this target sit, and why? The memo author must argue for a position relative to the class, citing specific evidence, before presenting the inside-view case.
• What would move it? List the evidence that would push the target above or below the class median.

Two cautions. A small firm's history may hold too few comparable deals to form a meaningful class; say so rather than inventing a base rate. And the outside view can itself become an anchor, so present it as the first estimate and require inside-view adjustments to be justified in writing.

Pre-mortems

Gary Klein's 2007 HBR article describes the technique: "In a premortem, team members assume that the project they are planning has just failed... and then generate plausible reasons for its demise." The article's framing is about dissent: those with reservations "may speak freely at the outset, so that the project can be improved rather than autopsied," and making it safe for knowledgeable dissenters to speak up improves a project's chances of success ^[7].

An IC version, run before the final vote:

1. Set the scene. The chair states: "It is two years after close. This investment has failed badly."
2. Write alone. Each member writes, silently and independently, the most plausible reasons for that failure. Several minutes is enough.
3. Collect in rotation. Go around the room, one reason per person per round, starting with the most junior member and ending with the deal lead.
4. Sort. Group reasons into those the team has evidence on, those it could test before signing, and those it cannot test.
5. Assign. Each testable reason gets an owner and a diligence item. Untestable reasons go into the decision log as accepted risks.

Illustrative example (fictional). A committee runs a premortem on a fictional regional services platform. Reasons include: the founder's informal customer relationships do not transfer; the largest customer's contract renewal was assumed rather than confirmed; the add-on pipeline is thinner than the thesis needs. The second reason is testable before signing, so it becomes a confirmatory diligence item with a named owner. The first is partly testable through reference calls. The third is logged as a known risk with a review date.

Memo and meeting mechanics

Much of an investment committee process's decision hygiene is sequencing and record-keeping. The 2011 HBR article makes the general case: with important decisions, executives should conduct a careful review "not only of the content of recommendations but of the recommendation process," and the authors offer a 12-question checklist for doing so ^[6]. Practical mechanics that follow from that idea:

• Circulate scores and memo in writing first. Members read and score before the meeting.
• Speaking order. Junior members first, deal lead and senior partners last. The aim is to hear independent views before the most influential one.
• Devil's advocate, with limits. A rotating assigned skeptic can surface objections, but an assigned role is easy to discount as theater. Pair it with written dissent: anyone who disagrees with the outcome records why, in one paragraph, in the decision log.
• Separate the evidence from the recommendation. The memo should state what is known (with sources), what is unknown, and what is inferred, in that order. See our evidence standards for how we separate those categories.
• Decision log. Record the decision, the vote, the key assumptions, the premortem risks, the dissent and a date to revisit. Without a log, the reference class for future deals does not exist.

When to trust intuition. Experienced partners often have strong instincts about targets. Kahneman and Gary Klein's 2009 paper in American Psychologist addresses when that is warranted. Its abstract concludes that "evaluating the likely quality of an intuitive judgment requires an assessment of the predictability of the environment in which the judgment is made and of the individual's opportunity to learn the regularities of that environment," and that "subjective experience is not a reliable indicator of judgment accuracy" ^[8].

Applied to deal teams, that suggests two questions before deferring to a partner's gut call:

Confidence alone is not on the list, because the paper's abstract says subjective experience does not indicate accuracy.

A one-page decision-hygiene checklist

Print this and attach it to the IC memo template.

Before screening

• Rubric criteria defined by observable evidence, with anchored scale points
• Allowed sources named per criterion; "unknown" is a separate value
• Rubric version recorded

During screening

• Every reviewer scores from the same frozen evidence packet
• Scores submitted independently, before anyone sees others' scores
• Criteria scored before any overall judgment
• Disagreements of two or more points flagged by criterion; originals preserved

Memo

• Reference class from the firm's own deal log, or a statement that none is large enough
• Target positioned against the class before inside-view adjustments
• Known, unknown and inferred facts separated, each with sources

IC meeting

• Written scores circulated before the meeting
• Junior members speak first, deal lead last
• Premortem run, with reasons sorted into tested, testable and untestable
• Written dissent recorded

After

• Decision log entry: decision, assumptions, risks, dissent, review date
• Recurring agreement check on a sample of screened targets, with results tracked over time

Limits. None of the sources cited here studies private equity screening or fund returns, and this article makes no claim that these practices improve any firm's returns. What these practices can show directly is narrower and measurable: whether reviewers agree more often, whether disagreements trace to fixable rubric language, and whether the committee's records support learning from past decisions.

Docket's Scout agent answers a client's screening criteria as typed values from a fixed list, each quoting a retained page verbatim.