Decide what a small marketing team must document using consequence, repetition, dispute and reversibility tests, with templates and cost scenarios.
Short answer: answer: Document choices that are consequential, repeated, disputed or hard to reverse; keep low-risk, one-off craft choices informal. Use the smallest record that lets another person recover context and act: owner, decision or procedure, reason, evidence, date and review trigger. If nobody can find it in two minutes or it has no owner, it is not useful documentation.
Small teams often swing between oral memory and a giant wiki. The first loses decisions when someone is absent; the second becomes a second job and soon contradicts reality.
The right boundary is not “important versus unimportant.” It is whether future error, repetition, disagreement or reversal costs more than capture and upkeep.
The Documentation Consequence Filter
The Documentation Consequence Filter scores four conditions: Consequence, Repetition, Dispute and Reversibility. CDM’s position is: a team should deliberately leave some low-risk craft choices undocumented; mandatory capture of everything makes consequential records harder to find and maintain.
Test 1: consequence
What happens if a new teammate guesses wrong? Document claims, consent, rights, spend authority, customer handoffs, publication credentials, corrections and contractual commitments. A font-spacing preference may stay in the design file unless it recurs as a brand rule.
Test 2: repetition
Will the team perform this at least three times in a quarter, or will several people do it? Repeated work benefits from a checklist or standard. One unusual research exploration can remain in the project record.
Test 3: dispute
Could reasonable people later disagree about what was decided, who owned it or which version applied? Capture decision, options, rationale, actor and date. The more stakeholders or money involved, the less reliable conversational memory becomes.
Test 4: reversibility
Can the team undo the choice cheaply and completely? A temporary headline test is reversible; licensing creator likeness for synthetic use or changing pricing architecture may not be. Hard-to-reverse choices need stronger evidence, review and retention.
Apply the filter
Use a 0–2 score for each test: 0 low, 1 moderate, 2 high.
| Total | Capture level | Example |
|---|---|---|
| 0–2 | Informal | one-off crop, internal working title |
| 3–4 | Task note | reversible channel choice, small test |
| 5–6 | Decision/checklist | audience rule, campaign measurement plan |
| 7–8 | Controlled record | claim approval, rights, consent, public correction |
The score routes attention; it does not override law, contract or policy. Any mandatory record goes directly to the required level.
Four documentation objects
Decision record
Use for a choice with alternatives. Minimum: Decision ID, question, options, choice, rationale, evidence, owner, date, consequence and review trigger. AWS’s architectural decision record guidance is a technical analogue: significant choices capture context and consequences, and a later record supersedes rather than silently rewriting an accepted decision.
Standard or checklist
Use for repeated work where sequence or acceptance matters. Minimum: purpose, trigger, owner, steps, exception, evidence produced and last test. Keep it near the work and update it when practice changes.
Register
Use for many governed objects: claims, evidence, creator rights, consent, campaigns or tools. Minimum fields depend on the object, but stable IDs, owner, status, dates and relationships matter.
Incident/correction record
Use when reality departs from intended control. Record what happened, affected scope, containment, evidence, decision, correction and prevention. Avoid blame narratives and vague lessons.
What can stay informal
- early brainstorming with no commitment;
- reversible aesthetic choices inside an approved brief;
- one-off exploratory queries whose result is not reused;
- routine conversation already reflected in the controlling task;
- personal craft notes that create no shared dependency.
Informal does not mean secret. The work product may still be visible; it simply does not need a separate governed record.
Two-minute retrieval test
Give a teammate who did not make the choice one prompt: “Why did we limit paid creator use to 30 days?” They must find the current answer, evidence, owner and superseded decision within two minutes. If they cannot, fix naming, search, links or authoritative location.
Test three records quarterly: a current decision, a repeated procedure and an old correction. Also test a departed owner. Documentation that exists only in a private drive or channel fails.
Cost model
Documentation has capture and maintenance cost. Illustrative team:
- 10 task notes weekly × 3 minutes = 30 minutes;
- 3 decision records × 12 minutes = 36;
- 1 checklist update × 20 minutes = 20;
- weekly retrieval/expiry review = 20.
Total is 106 minutes. At a $70 loaded rate, modeled weekly cost is about $124. If the practice avoids one 3-hour reconstruction, modeled avoided capacity is $210; net modeled difference is $86. These are scenario calculations, not a time study or savings claim.
Maintenance matters. Fifty pages reviewed for 10 minutes quarterly consume 8.3 hours per quarter. Archive or remove a page whose owner, user or decision no longer exists. Do not preserve stale instructions because writing them was expensive.
Complete examples
Document: “Use customer testimonials in paid social.” It scores Consequence 2, Repetition 2, Dispute 1 and Reversibility 1 = 6. Create a decision record plus rights register entries, disclosure requirements and review date.
Task note: “Test two CTA labels for one landing page.” It scores 1, 1, 1, 1 = 4. Record hypothesis, variants, metric and stop rule in the experiment task; no separate policy page.
Informal: “Crop the approved image two pixels higher.” It scores 0, 0, 0, 0. The final asset version is enough.
Controlled record: “Correct a false durability claim.” It scores 2 in every dimension. Preserve the claim family, affected uses, correction, audience remedy and approval.
Documentation workflow
- Apply the filter when a decision or repeat pattern appears.
- Create the smallest appropriate object in the authoritative system.
- Name one owner and review trigger.
- Link it from the work rather than copy it.
- Mark a replacement as superseding the former record.
- Run retrieval and stale-content checks.
Automation may generate an ID, remind owners and flag broken links. It should not invent rationale, decide that a choice is consequential or update a procedure from observed behavior without human approval.
Security, privacy and retention boundary
More documentation creates more discoverable and sensitive data. Restrict customer, personnel, contract and legal records; follow the applicable retention schedule; avoid pasting secrets into general workspaces. GOV.UK security guidance emphasizes identifying information assets and risks, while risk owners decide acceptable controls. Small size does not eliminate information governance.
Related guides
Frequently asked questions
Should every marketing decision have a record?
No. Record decisions whose consequence, repetition, dispute potential or reversibility justify capture. A one-off reversible visual adjustment can remain in the asset history; a new claim, audience exclusion, spend rule or rights interpretation should not. The filter protects attention for consequential context.
The caveat is mandatory records: law, contract, finance, privacy or platform policy may require documentation regardless of the score, so route those directly to the governed system. Publish examples for each capture level and review disagreements quarterly; inconsistent scoring can reveal an unclear policy or different risk assumptions.
How long should a decision record be?
Usually one screen: question, options, choice, rationale, evidence, owner, date, consequences and review trigger. Complexity belongs in linked evidence, not pasted transcripts. AWS guidance for architectural decisions emphasizes context and consequences; marketing can use the same economy. A public correction or regulated claim may need a fuller file and specialist sign-off.
Brevity is valuable only if a future reader can still understand why the choice was reasonable then. Test one record with a teammate who was absent; missing questions should improve the template, while background that never changes action should be removed.
Where should small-team documentation live?
In the system closest to the governed object and accessible to the people who need it. Customer lifecycle decisions may live in CRM; claims and rights in registers; repeated procedures in the work/knowledge system.
Use stable IDs and links instead of copying. One universal wiki is not necessary, but an index of authoritative locations is. Sensitive legal, HR or security records require separate access even if that makes search less convenient. Assign an owner for the index, test links and access with non-admin accounts, and export critical records under the organization’s recovery policy.
How do we stop documentation becoming stale?
Give each controlled record an owner and event or date trigger, then surface overdue items in a regular review. When a new decision replaces an old one, mark the old record Superseded and link both. Sample retrieval quarterly and archive material with no current user or obligation.
A fixed annual review is insufficient for volatile pricing or platform guidance; high-change information needs shorter or event-driven checks. Do not update the date without reviewing substance; record what changed, what was confirmed and which evidence supports the current version.
Can AI write our documentation automatically?
It can draft from cited inputs, format a decision and suggest missing fields, but an accountable person must verify the actual choice, rationale, evidence and access. Transcripts can misidentify consensus and models can invent reasons. Never let AI copy confidential conversations into an unapproved system.
A low-risk task summary may be auto-drafted and confirmed; consequential decisions should show the human owner and source links explicitly. Preserve enough prompt or transformation context to reproduce material outputs, and provide a manual route when the model is unavailable or uncertain.
How do we know documentation is worth the time?
Track capture minutes, retrieval success, repeated questions, reconstruction hours, errors from stale guidance and onboarding use. Run the two-minute retrieval test rather than surveying whether people “like the wiki.” Compare cost with observed avoided work cautiously; an avoided incident is difficult to prove. If a document is never found, used or required, simplify, relocate or archive it.
Do not use page count as a maturity metric. Segment governed records from optional knowledge notes, because one may be valuable through rare risk prevention while the other should justify recurring maintenance through frequent use.
Next decision: Build a modern content supply chain
Related reading: Design the contemporary marketing operating system · Create a practical experimentation system · Correct one false claim across channels
Sources and research notes
- AWS Prescriptive Guidance, Architectural decision record process.
- UK Government, Architectural Decision Record Framework.
- GOV.UK Service Manual, Securing information in your service.
- GitLab Handbook, Team decision-making and transparency and DRIs.
Research checked 26 September 2026. CDM conducted no original failure-cost study or retrieval benchmark. Scores, thresholds, capture times, rates and modeled savings are illustrative scenario calculations. Apply legal, contractual, security, privacy and records-retention requirements before using the filter, and validate retrieval with the actual team.
This article is editorial guidance. Apply the principles in proportion to your market, evidence, and responsibilities.
