Workplace Communication
How to Keep a Decision Log People Can Actually Use
Keep decisions findable with clear owners, context, alternatives and status, plus a worked three-month review that preserves the earlier rationale.
Keep a decision log by recording the question, decision owner, context, alternatives, rationale, consequences and review conditions for each significant choice. Give the record a stable identifier, state whether it is proposed or accepted, and make the currently applicable decision easy to find. When circumstances change, link the replacement to the earlier record instead of rewriting history. Keep decision approval separate from implementation and observed results.
This LandOffer.ai guide includes a completed fictional record and its review three months later. The people, documents, approvals and events are invented teaching inputs, not actual company records or a tested collaboration workflow. The example shows how to maintain a usable history; it makes no claim about measured efficiency or improved delivery.

Decide which choices deserve a durable record
A log is useful when someone will need to understand a consequential choice after the original conversation ends. Examples include accepting a scope reduction, selecting an integration boundary, assigning an ongoing review responsibility, or changing a release process. Routine coordination can stay in the task record unless it changes an important constraint or commitment.
Start with the question a future colleague will ask: what are we doing, why, who authorized it, and does it still apply? A date followed by “use the simpler approach” answers none of those questions. A transcript creates the opposite problem: the reader must reconstruct the decision from everything everybody said.
Michael Nygard's original 2011 architecture-decision-record proposal describes short records containing context, decision, status and consequences, with superseded decisions retained. Its stated scope is architecturally significant choices. Applying a similar structure to a team's operational decisions is an editorial adaptation here, not a claim that every meeting requires an ADR.
Agree on the decision threshold and recording owner with the team. Avoid making one diligent colleague responsible for reconstructing choices they did not attend. The person recording a decision can request clarification; they cannot create approval by writing someone's name beside a proposal.
Make status and ownership mean something specific
Distinguish the person organizing the decision from the person authorized to make it. Name contributors where their input matters, and identify people who need the result. Confirm those responsibilities with the people involved rather than treating a copied invitation as their acceptance.
Atlassian's decision-documentation template distinguishes driver, approver, contributors and informed participants. Its public guidance also calls for context, options and action owners. These are useful prompts for a record, not evidence that the template itself resolves authority or prevents repeated debate.
Use status words consistently. Proposed means a choice is under consideration. Accepted means the authorized decision was made for the stated scope. Superseded means a linked later decision replaces it. Rejected means that option was considered and declined, with its reason preserved. Agree on any additional labels your team needs.
Track execution separately. An accepted plan can have an outstanding implementation task. A completed implementation can still lack evidence of the hoped-for benefit. If those states share one green “done” label, a reader may mistake permission to proceed for proof that the change already works.
A completed record with a rejected alternative
The fictional developer-documentation team maintains two guides: Aster SDK and Birch CLI. Priya owns documentation publishing policy. Rowan coordinates the pilot; Miles and Noel provide the agreed backup technical review for Aster and Birch respectively. Existing primary maintainers retain technical approval of guide content. The record concerns a publishing workflow, not an architecture or employment decision.
On January 12, 2026, Priya accepts the following record after confirming the named participants' roles. It is complete as a decision record; its future pilot outputs are not treated as already published.
DOC-014 — Require recorded technical approval for the two-guide pilot
Decision date and owner: January 12, 2026. Priya approves the publishing rule; Rowan maintains this record and coordinates the agreed review route. The primary maintainers approve technical content, with Miles and Noel available as the named backups within confirmed review capacity.
Context and scope: the team is preparing a limited pilot of revised Aster SDK and Birch CLI guides. A proposed publication slot is not evidence that a maintainer has approved the instructions. The team needs an explicit approval boundary without designing a new automated approval system for this pilot.
Options considered: publish when an author marks a draft ready; require a recorded technical approval for each guide; or postpone both guides until a new approval system exists. Author readiness alone is rejected because it does not establish technical approval. Waiting for a new system is rejected for this bounded pilot because no such implementation is scoped or funded.
Accepted choice and reason: use the existing review workspace and require a named technical approval for each guide's exact draft version before publication. Missing approval holds that guide. A slot may be moved; the content gate is not waived merely to keep the date. This fits the limited scope and available review capacity, while leaving a manual coordination burden.
Consequences and limits: Rowan must connect the draft version, approval and publishing task. Review availability can delay a guide. An approval applies to the reviewed version, not every later edit. This choice does not establish that the instructions were tested, authorize a new system, or prove that manual review is more effective than automation.
Follow-through and review: Rowan prepares the pilot checklist before the January 26 slot. Priya reviews the arrangement on April 13, or earlier if confirmed review coverage changes. Technical-content concerns follow the existing maintainer route immediately. Rowan brings changed guide scope or missing review coverage to Priya for a new scope decision; neither condition authorizes publication.
The record tells a later colleague why the team chose a manual gate and what cost it accepted. It also prevents a reader from treating the automation alternative as approved work. No automation implementation was selected or funded by this decision.
The rejected options are specific enough to reconsider if circumstances change. They are not caricatures designed to make the accepted choice look inevitable. A future team could have a funded approval system, different guide scope or different review capacity; the rationale would then need another decision.
Give the log a usable index and a clear home
Store records where the relevant team can actually access them through its approved workspace. Use a descriptive title, stable identifier, owner, decision date, status and replacement link in the index. Put the question and rationale in the record itself so someone joining the project can understand the choice before following its supporting links. Link discussions instead of making a chat search the only way to recover the reasoning.
AWS's ADR process describes a decision log as a collection whose headlines help readers understand context, while individual records carry the detail. Its owner and review process are guidance for architecture records. For an operational log, agree on an appropriately small process rather than copying every review step.
Prefer a stable link over attaching a fresh copy to every update. If copies are necessary, label their date and point back to the current location. Keep confidential personnel, customer or security material in its authorized location; a team-readable index does not justify opening sensitive supporting records to everybody.
Test the index with a concrete question: “Can Birch publish now, and who can answer a technical approval question?” The reader should find the applicable publishing rule, current scope and named owner without comparing several nearly identical files. This is a practical review question, not a measured usability result from an actual team.
Three months later: preserve the reason and replace the scope
At the fictional April 13 review, Rowan brings three example pilot packet records. The January 26 Aster packet and February 23 Birch packet contain recorded approvals for their reviewed versions. Neither packet approves a later draft or establishes that the guide was actually published. The March 23 Birch packet remains held because a technical instruction still needs correction. These invented packet states do not demonstrate real publication, tested instructions or customer benefit.
A separate April 10 planning change ends Noel's confirmed backup review allocation for the next pilot period. Miles confirms capacity for Aster only. Rowan does not infer Birch coverage from the fact that a previous Birch packet had an approval. The changed capacity reaches Priya through DOC-014's review condition.
Priya considers three choices: continue two guides with an unconfirmed backup; narrow the next pilot to Aster; or pause both while arranging broader coverage. She rejects the first because it represents coverage the team has not secured. She does not choose the third because Aster retains confirmed support and can remain bounded. Priya accepts DOC-015 on April 13.
DOC-015 — Limit the next pilot period to Aster SDK
Owner, date and status: Priya, April 13, 2026; accepted for the next pilot period. This supersedes DOC-014's two-guide scope. Rowan coordinates the change; Miles confirms Aster backup review. No Birch replacement reviewer is approved.
Decision and reason: proceed with Aster only, retaining the exact-version technical approval gate. Pause Birch pilot publication until Priya approves a supported scope with confirmed review coverage. Existing published material is outside this decision's change scope; this record does not order its removal.
Retained constraints: the March 23 Birch draft remains held for its unresolved correction. Narrowing the pilot does not resolve that technical issue. Author readiness, previous approval of another version and an open publication slot remain insufficient to authorize publication.
Consequences and next actions: Rowan will update the pilot checklist, scope notice and publishing task by April 16. Priya will check those updates before the next slot. These actions are outstanding when this record closes. No actual publication, removal or implementation is claimed.
Revisit condition: return to Priya when a named Birch reviewer has confirmed capacity and the maintainer has resolved the draft's technical concern. Those inputs permit a new scope review; they do not automatically restart Birch or reverse this decision.

Rowan adds a dated supersession notice and DOC-015 link to DOC-014's lifecycle information, leaving its January rationale intact. The index now reads:
| Record | Current lifecycle status | What a reader should do |
|---|---|---|
| DOC-014: two-guide approval pilot | Superseded by DOC-015 on April 13 | Read for the original alternatives and approval-gate rationale; follow DOC-015 for current pilot scope |
| DOC-015: Aster-only next period | Accepted; checklist and task updates outstanding | Use the narrower scope; confirm the separate publishing requirements before any publication |
This is a linked change history, not a retroactive claim that the team always intended to exclude Birch. The original choice made sense under a stated capacity assumption. The later choice records why that assumption no longer supports the same scope.
Review changed evidence without rewriting the past
A scheduled review should answer whether the decision still fits its context. It need not produce a replacement every time. Record the reviewed evidence, the owner's conclusion and the next relevant trigger. If the choice remains valid, a dated review note can say so without pretending a new decision occurred.
Microsoft's ADR guidance recommends an append-only history, linked superseding records, and explicit options, trade-offs, confidence and status. It also distinguishes a decision record from a design guide. Borrow that discipline while keeping your log proportionate to the task.
Correct factual mistakes transparently. A wrong link or misspelled name can be corrected with a visible note; changing the accepted scope requires its authorized decision. Do not use “cleanup” to erase a rejected alternative, hide an earlier constraint or make a forecast look like an observed result.
Before the next handover, ask a colleague to locate the current decision, explain its reason and identify the remaining actions. Record any missing link, unclear status or unconfirmed owner, then ask the responsible decision owner to resolve it by an agreed review date. For more practical workplace guides, visit LandOffer.ai.


