# Doc Decision Memo

> Writes a Decision Memo with the question in one line, options with evidence, a recommendation, a reversibility call, who decides by when and a signature line. Use for "run doc-decision-memo", "write the decision memo", "get sign-off that sticks", "we keep reopening this decision", "lay out the options for the VP", "one-way or two-way door", "who is the approver here", "decision doc for leadership", part of the Claude Docs for Product Managers Pack by 90n.

- Skill: `polar-bear-org/doc-decision-memo` (Agent Skill)
- Install (CLI): `npx get-skill add polar-bear-org/doc-decision-memo`
- Raw SKILL.md: https://api.skillmd.com/api/skills/polar-bear-org/doc-decision-memo/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: polar-bear-org (https://skillmd.com/u/polar-bear-org)
- Updated: 2026-11-14
- Page: https://skillmd.com/skills/polar-bear-org/doc-decision-memo

---


# Decision Memo

## When To Use
You get sign-off, then people change their minds and the roadmap lasts a week. The memo answers one question: which option do we take, who alone approves it, and what new information would it take to reopen it?

## When Not To Use
If one request is fighting for a slot in current work, Trade-off Memo is faster. If the decision is already made and you only need it on record, add it to the Decision Log; this memo makes a decision, it does not store one.

## Inputs
- The question as people are currently arguing it, and any thread or notes where it came up
- The options on the table and the evidence for each (data, research, cost, risk)
- Who is involved: who drives, who approves, who has a view, who needs to know
If you have none of this, I start from the question in your words and two options plus doing nothing, and mark the output as a first draft.

## Approach
DACI, as the Atlassian Team Playbook sets it out (https://www.atlassian.com/team-playbook/plays/daci): one Driver, one Approver, Contributors with a voice but no vote, Informed told after. Reversibility comes from Amazon's 2015 shareholder letter (https://www.aboutamazon.com/about-us/shareholder-letters): one-way doors deserve slow care, two-way doors deserve speed. The failure this prevents is the decision with three approvers, where whoever spoke last wins and the memo gets rewritten every Monday.

## Workflow
1. Ask at most three questions: who reads it and who is the single Approver, by when the decision is needed, and any length budget. Skip them if a Doc Brief is pasted. Two Approvers is a finding: name it and ask who holds the call.
2. Write the question in one line, answerable by choosing an option. "What should we do about pricing" fails; "do we move the export feature to the paid plan this quarter" passes.
3. Set out 2 to 4 options, always including do nothing. For each: what it means, evidence with its source, cost, main risk. Missing evidence stays an open question, never a guess.
4. Call reversibility per option: one-way door (hard to undo, so name what slowing down buys) or two-way door (cheap to undo, so decide now and watch a signal). Say why in one line.
5. Recommend one option and tie it to the evidence. Then write the revisit trigger: the named fact or threshold (user sets) that would justify reopening. Without new information, the decision stands.
6. Fill the DACI table and the signature line: Approver name, date, option chosen. Leave the decision blank until the Approver signs; Claude never records a decision that was not made.
7. Draft in Claude Docs (beta) and share it inside your organisation; for a formal review on a .docx, Claude for Word puts edits in tracked changes. If neither is on your plan, I give the same memo as plain chat output.

## Output Format
```markdown
# Decision Memo
**Question:** [one line] | **Recommendation:** [option] | **Decide by:** [date]
## Options
| Option | What it means | Evidence and source | Cost | Main risk | Door |
|---|---|---|---|---|---|
| Do nothing | [placeholder] | [source] | [placeholder] | [placeholder] | [one-way / two-way] |
| [Option B] | [placeholder] | [source] | [placeholder] | [placeholder] | [one-way / two-way] |
## Why this option
[Two to four sentences tied to the evidence above.]
## What would reopen it
[Named trigger or threshold, and who watches it]
## Roles
| Driver | Approver (one) | Contributors | Informed |
|---|---|---|---|
| [name, role] | [name, role] | [names, roles] | [teams] |
## Decision
Approver: [name] | Option chosen: [blank until signed] | Date: [date] | Logged in the decision log by [Driver] on [date].
```

## Done When
- The question is one line and each option answers it
- Do nothing is an option, with its own cost and risk
- There is exactly one Approver, by name
- Every option has a door call and every claim a source or an open question

## Quality Bar
- Options are compared on the same columns; no option gets a longer, warmer write-up
- The revisit trigger is specific enough that a change of mind needs new information
- Roles are named for accountability only; no judgment of anyone's past calls
- The decision field stays blank until the Approver signs
- Claude lays out the options; the named Approver signs, and Claude never records a decision that was not made.

## Next
Run doc-pr-faq (Working Backwards PR/FAQ) to test a big bet against the customer before building.

## About the makers

This pack is made by 90n ([90n.ai](https://90n.ai)): AI-native ways of working and tools for Product, Marketing and HR teams, built by ex-McKinsey founders. We map where time and money leak, redesign the work around AI and stay until your people run it themselves, and we run our own company on Claude. If your team has outgrown the self-serve version, message Pauline (linkedin.com/in/paulinebertry).

