How to review decision quality without creating more bureaucracy
Review decision quality by sampling consequential decisions and comparing the recorded reasoning with the information available at the time. Examine the outcome separately, then identify a specific improvement worth making. Routine choices do not need a formal memo merely to make the review process look thorough.
Outcome bias is the influence of a decision’s result on judgments about the quality of the decision itself. Baron and Hershey’s 1988 experiments examined this effect. Their work provides a reason to preserve the original information set, rather than treating a favorable result as proof of sound reasoning.
A useful review must still care about performance. Process quality, execution, and results are different questions that belong in the same conversation. Separating them helps the company choose a more accurate response.
The sampled decision review below is a proposed management practice. It is not a validated scoring instrument and does not guarantee better business outcomes.
Review learning value rather than paperwork volume
In this explicitly composite scene, a CEO opens a folder of decision templates at the kitchen counter while their partner clears dinner. The organization introduced them after a costly launch problem. Teams now complete forms for ordinary purchases, while a major commitment still moved through an informal conversation. The paperwork expanded without making the important reasoning easier to inspect.
The remedy begins with scope. Identify the decisions for which a mistaken assumption, unclear authority, or unmanaged commitment could create meaningful exposure. Give those decisions a proportionate record. Leave ordinary delegated work within its existing controls.
A review system should help the next decision, not create a second organization devoted to explaining the last one. If the review produces recommendations nobody owns, its apparent thoroughness offers little operating value.
Connect the practice to the CEO decision system. The initial decision record and the later review should use compatible definitions so teams can retrieve evidence without reconstructing the whole story.
The sampled decision review
1. Choose a small, defensible sample
Select decisions for their potential learning value. Consider consequence, uncertainty, novelty, and reversibility. Include some favorable outcomes as well as unfavorable ones. Reviewing only failures can miss unsupported choices that happened to work.
State the selection rule before looking for a convenient story. The sample does not need to represent every decision statistically unless the company intends to make that kind of inference. A targeted learning sample should be labeled accordingly.
Avoid selecting a decision solely because one leader has become unpopular. If a particular incident requires investigation, use the appropriate process. A learning review should not serve as a disguised disciplinary proceeding.
Choose a manageable review cadence and an owner who can ensure follow-through. The right frequency depends on the decisions and when useful outcome information becomes available. Do not invent a universal schedule.
For each selected decision, identify the question the review will answer. Examples include whether the evidence threshold was appropriate, whether commitments exceeded authority, or whether a review trigger was ignored. A clear question keeps the session bounded.
2. Reconstruct only the information available then
Collect the original decision statement, options, assumptions, approvals, and known constraints. Prefer records created before the commitment. Ask participants to distinguish remembered reasoning from documented reasoning.
Do not infer that an undocumented consideration was never discussed. Record the uncertainty. At the same time, do not accept a confident retrospective account as equivalent to a contemporaneous record.
Create a short timeline showing when material information arrived. Information learned after approval belongs in the execution and updating review. It cannot automatically establish that the original owner should have known it earlier.
Simon’s work on bounded rationality, discussed in the Stanford Encyclopedia entry, addresses decisions made under information and computational limits. A fair review examines the actual constraints rather than comparing the decision with an imaginary fully informed alternative.
Ask what additional information was obtainable at reasonable cost and within the available time. That question is more useful than listing everything the organization eventually learned. It also helps identify where a future evidence requirement would be worthwhile.
3. Examine the reasoning against explicit criteria
Use a short set of questions: was the decision defined, were feasible alternatives considered, were consequential assumptions visible, and did the authorized person approve the exposure? Adapt the criteria to the kind of decision.
Look for reasoning the record supports. A statement that the team “considered the risks” is less useful than a named risk, its evidence, and the response. Ask which assumptions carried the conclusion and whether those assumptions received appropriate attention.
Do not score every item merely because the form permits it. A precise number can imply more certainty than the evidence warrants. Descriptive judgments with concrete examples may support a better discussion.
Lerner and Tetlock’s 1999 accountability review shows that expectations to justify judgments have different effects depending on their design. Explanation requirements should encourage examination of evidence, rather than reward people for defending a decision the audience already favors.
Let participants identify a weakness without treating admission as an automatic performance failure. The review still requires accountability, but it should not make polished self-defense the easiest way to succeed.
4. Separate the original choice from execution and updating
A reasonable decision can be implemented poorly. A weak original choice can be rescued through strong execution. Identify what the review is evaluating before deciding what to change.
Examine whether the team acted within the approved scope, responded to new information, and used the agreed escalation route. A missed trigger is different from a trigger that was never defined. Each suggests a different correction.
Use decision reopening rules to evaluate changes after approval. Ask whether new evidence warranted reconsideration and whether the company could act before exposure increased. Do not assume that consistency always meant good judgment or that every change showed indecision.
Then review the result using the outcome measures established for the decision. If the company changed its definition of success later, preserve both definitions and explain the change.
This sequence allows a clear conclusion: the initial choice was defensible, execution failed in a specific way, and the response requires an operating change. Another review may reach a different conclusion. The method should permit both without forcing a single culprit.
5. Choose one or two changes with observable effects
Translate the findings into a small number of actions. Name the owner, due date, and evidence of completion. Avoid a long list of general aspirations such as improving communication or becoming more data driven.
A proposed correction might require confirmation of a specific dependency before the next major commitment. Another might clarify who can authorize a scope increase. These are examples, not universal requirements for every business decision.
Check the burden the correction creates. If a new form takes extra effort but does not change any decision, simplify it or remove it. A control should address the failure mechanism the review identified.
Use decision metric gaming as a check on the proposed measure. Counting completed memos can show activity without showing whether important assumptions became visible. Choose evidence connected to the behavior the company needs.
Where the finding concerns regulated conduct, employment, safety, or legal obligations, involve the relevant qualified adviser. A management review does not replace an investigation, required reporting, or professional advice about a specific situation.
6. Verify that the change reached the next decision
At the next relevant commitment, inspect whether the correction was used and whether it helped answer the intended question. Completion of a training session is different from evidence that the decision process changed.
Ask the people doing the work whether the requirement is understandable and practical. A well-intended correction can create a bottleneck if its owner lacks access to the necessary information or authority.
Keep the verification short. Compare the new decision record with the specific change requested. If the control did not help, revise it rather than adding another layer automatically.
The proposed closing script is: “This was the reasoning available then. This is what execution and later evidence changed. We will make this adjustment and inspect its use in the next relevant decision.” These are suggested words, not a research quotation.
Retain the review where the next decision owner can find it. A useful lesson stored in a private presentation may not inform the work it was intended to improve. Use appropriate access controls when the record contains sensitive information.
Keep a short decision record
For consequential choices, a compact record can contain the decision, owner, date, feasible alternatives, key assumptions, accepted exposure, and reopening condition. Attach existing evidence rather than copying it into several documents.
Do not require every field for every delegated action. The record should scale with the decision and fit the company’s existing governance. A standard purchase under an approved policy need not receive the same treatment as a major acquisition.
Separate facts from forecasts. A signed agreement is a fact about a commitment. Its expected commercial benefit remains a forecast until the relevant outcome is observed. This distinction helps later reviewers assess what the original record established.
If the original record is thin, complete the review with that limitation visible. Do not invent a detailed history. Use the gap to decide which future decisions merit better documentation.
Judge the review by what it changes
A review has practical value when it clarifies authority, improves a consequential assumption, or changes how the company responds to new evidence. Its value does not increase automatically with the number of pages produced.
Before scheduling another session, ask whether the previous actions were completed and whether the selected decisions still offer useful learning. Retire a review ritual that no longer serves a defined purpose.
The CEO’s role is to protect a fair comparison between what was knowable, what was chosen, what happened, and what should change. That comparison can fit into a focused meeting when the company preserves the right evidence from the start.