How to Build a Decision Journal That Improves Decisions

How to Build a Decision Journal

A decision journal preserves what you knew before the outcome. Its value is not the act of writing. It is the ability to compare your later story with a dated record of evidence, uncertainty, alternatives, authority, and expected consequences.

The journal should be proportionate, searchable, and hard to rewrite.

flowchart LR
    A["Decision record"] --> B["Evidence and alternatives"]
    B --> C["Probabilities and risks"]
    C --> D["Choice and stop rule"]
    D --> E["Outcome appended later"]
    E --> F["Calibration and rule update"]

1. Create a stable identity

Give the record a decision ID, title, owner, date, deadline, and status. Link the project, source material, approvals, and affected systems.

Write one sentence that states the decision. If the sentence contains several choices, split the record.

2. Define why the decision exists

Describe the problem, desired outcome, constraints, and cost of delay. Name the people affected and the person with authority to decide.

Record whether the choice is reversible. Reversible decisions can often use smaller experiments. Irreversible or high-consequence decisions need stronger evidence and review.

3. List real alternatives

Include doing nothing, delaying, reducing scope, testing, buying, building, delegating, and stopping where they are genuinely available.

Do not add imaginary alternatives after the outcome. Explain why any plausible option was unavailable.

4. Preserve evidence and unknowns

Link the sources used. Separate observed facts, estimates, interpretations, assumptions, and missing information.

Record evidence you chose not to gather and why. Time and cost limits can be reasonable, but they should remain visible.

The outcome-bias studies indexed by PubMed show why a pre-outcome record matters. Results can change how people judge an unchanged process.

5. State probabilities

Define each forecast as an event with a resolution date and rule. Replace vague language with a probability or range when the evidence supports it.

Do not pretend to know more than you do. A rough 60 percent estimate with named assumptions can be more honest than 63.4 percent with no basis.

Preserve the estimate. If evidence changes, append a new probability, date, source, and reason rather than overwriting the old one.

6. Record base rates and inside views

The outside view asks what happened in comparable cases. The inside view asks what is distinctive here.

Record both. A base rate without context can mislead, and a detailed project story can make every case look unique.

7. Name harm, authority, and controls

List foreseeable failure modes, affected people, legal or policy boundaries, privacy and security concerns, conflicts, and required approvals.

Assign preventive and detective controls. Name an owner and response.

A decision journal is not a way to authorize yourself. High-stakes decisions still require the appropriate legal, financial, medical, employment, safety, scientific, or regulatory process.

8. Capture dissent

Ask reviewers to write independent concerns before discussion. Preserve minority views and the evidence that would change each position.

Annie Duke's book profile emphasizes truth-seeking groups. The useful implementation is a record that makes disagreement safe and specific.

9. Run a premortem

Assume the decision failed. Generate causes independently, then convert material causes into indicators, controls, owners, responses, and exit rules.

Gary Klein's premortem method page describes the exercise as beginning after a team has been briefed on a plan. The related prospective-hindsight research belongs in the evidence record.

Do not let vivid failure stories replace probability or evidence. A premortem expands the risk set; it does not rank every risk correctly.

10. Write the choice

Record the selected option, reasons, decision maker, approval, date, implementation owner, monitoring plan, and next review.

State what would change the decision. A useful stop rule names a signal, threshold, owner, response, and deadline.

11. Append the outcome

After the result, do not edit the original entry. Add the observed outcome, date, benefits, harms, costs, surprises, and data quality.

Review the process before judging it with the outcome. Then compare the result with the predicted range and update future rules.

Keep the journal usable

Use a short template for low-risk choices and a fuller record for costly, irreversible, regulated, safety-sensitive, or people-affecting decisions.

Protect confidential or personal information. Set access, retention, correction, and deletion rules appropriate to the organization.

The journal succeeds when it changes a later decision. It should reveal calibration, repeated blind spots, weak controls, and assumptions that deserve better evidence.

Review the journal itself. If entries are never reopened, fields are routinely empty, or teams write them after commitment, simplify the template and repair the operating habit. A smaller contemporaneous record is more useful than a perfect form completed from memory.

AI assisted with research organization, structure, drafting, and validation. Dalton Anderson remains the attributed author and final editorial authority. The transcript and linked public sources control factual claims. Publication remains unauthorized.