Record a decision, not a meeting transcript
A meeting note captures a discussion. A decision log isolates the commitment that came out of it. A roadmap shows intended work; a log explains why that work, rather than another option, was chosen.
Not every small edit needs an entry. Record choices that change scope, user promises, priorities, constraints, or a previously accepted direction. Give each entry an owner and a status so a suggestion is not mistaken for a commitment.
Product decision history can include proposed, accepted, deferred, rejected, and superseded entries. The history should make change understandable, not prevent people from changing their minds.
A reusable decision-log entry
Decision ID / title:
Status: Proposed | Accepted | Deferred | Rejected | Superseded
Owner / reviewers / decision date:
Context: What question requires a choice now?
Evidence: Which sources support or contradict the proposal?
Options: What alternatives, including doing nothing, were considered?
Decision: What exactly is accepted or declined?
Rationale: Why does this option fit the current goal and constraints?
Consequences: What trade-offs and follow-up work are accepted?
Scope: Who or what is affected? What is explicitly excluded?
Revisit condition: What evidence or event would reopen this decision?
Links: Goal, requirements, evidence, and any previous decision.
Review date / later outcome:Prefer specific rationale over “the team agreed.” “We chose manual approval because exceptions remain poorly understood” explains a trade-off. Consensus alone does not explain it.
Example: defer automatic refunds
Fictional entry D-18: a service team wants to automate refunds for cancelled bookings.
- Context: manual review is slow, but the team has not classified the reasons for exceptions.
- Options: retain the current process, add a structured request form, or approve refunds automatically.
- Decision: accept a structured request form with manual approval; defer automatic approval.
- Rationale: collecting the missing information can improve review while preserving judgment for unclassified exceptions.
- Consequences: some delay remains; the reviewer needs a clear queue and ownership.
- Revisit condition: review a representative set of completed requests and agree on which exceptions can be specified reliably.
No measured savings are claimed. The example demonstrates how an honest unresolved uncertainty can justify a bounded first step.
Decision lineage explains how direction changes
Suppose the team later learns that one clearly defined cancellation category can be handled consistently. New entry D-24 may authorize a limited automation for that category while leaving other requests under manual review.
D-24 should reference D-18 and the new evidence, identify the exact part superseded, and point to the affected requirements. Do not rewrite D-18 to make it appear that the team always intended the new policy. That would destroy the explanation for earlier behavior.
A current product state might now say “limited automation accepted.” The log retains the path to that position. This is decision lineage: the relationship between earlier choices, later evidence, and revised commitments.
Keep a log usable during delivery
- End a decision discussion by reading back the precise commitment and its exclusions.
- Store one entry in a known location and link relevant evidence rather than copying entire conversations.
- When preparing a PRD, reference the accepted entry and version.
- If implementation reveals a conflicting assumption, reopen the decision through its owner.
- After outcome review, append what was learned and whether the commitment still holds.
For AI-assisted work, require proposed entries to remain proposed until the owner accepts them. An assistant can draft the wording, but it should not silently convert an options brief into an approved decision.
AWS describes an architectural decision-record process for technical decisions. The product log here applies a related recordkeeping idea to product choices; it is not an architectural specification. See PM Harness for where decisions connect to research and verification.