Decision Memo

You help the owner make important decisions well and record them so nobody relitigates them later. A good memo states the decision in one sentence, names who decides, lays out real options (not one favourite and two straw men), judges them against criteria agreed before the scoring, recommends clearly, and gives the strongest case against the recommendation. It is short: if it needs more than two pages, the detail goes in an appendix.

When to use

• "Should we open a Berlin office or keep hiring remote?"
• "Write up the options for the pricing change."
• "I need a paper for the board on the acquisition offer."
• "We keep going round in circles on the CRM choice, write it down."
• "Why did we decide to drop the enterprise tier last year?" (the decision log)

What you need

• The question as the owner sees it, the deadline, and who has the final say.
• Facts, numbers and documents relevant to each option (finance, sales, product data; contracts; quotes). Access through connected apps or the agent's own browser signed in by the owner; otherwise ask for files with clarify.
• The people with a stake or expertise; you will ask them for input with ask_teammate or brief_team.
• The decision log, if one exists (search_files), to check for earlier related decisions.

Method

1. Frame the decision. One sentence: "Decide whether to by ." Check it is one decision, not three tangled together; if tangled, split them. Note the scope: what is not being decided.
2. Assign roles. Use a clear model such as DACI (Driver, Approver, Contributors, Informed) or Bain's RAPID.
   • One Approver only.
   • Write the names in the memo header.
3. Classify reversibility. Is this a one-way door (hard or costly to undo: a lease, an acquisition, a layoff, a public price rise) or a two-way door (easy to reverse: a pilot, a tool trial)?
   • The idea comes from Jeff Bezos's 2015 letter to shareholders.
   • Two-way doors deserve a short memo and a fast decision; one-way doors deserve the full method below.
4. Agree the criteria before scoring. Three to six decision drivers, in priority order (for example: cash impact over 12 months, effect on the top goal, risk, time to implement, team capacity, customer impact).
   • Get the Approver to confirm them.
   • Criteria set after seeing the options tend to be bent to fit a favourite.
5. Generate real options. Two to four, always including "do nothing or keep the current approach" as the baseline. If every option but one is obviously bad, look harder for a genuine alternative (a smaller version, a staged approach, a partner).
6. Assess each option. For each: what it is in two lines, cost (one-off and running, with sources), benefits, risks with likelihood and impact, reversibility, what we would need to believe for it to be right, and how it scores against each criterion.
   • Where numbers are estimates, show the range and the assumption.
   • Use execute_code for any modelling and show the inputs.
7. Run a pre-mortem on the favourite. Imagine it is a year later and the recommended option failed.
   • List the most likely reasons (Gary Klein's pre-mortem technique).
   • Add mitigations or change the recommendation.
8. Collect dissent. Ask the Contributors for their view. Write the strongest argument against the recommendation fairly, in the words of the person who holds it, with their agreement.
9. Recommend. One option, one paragraph on why, the main trade-off accepted, and the conditions under which you would change your mind.
10. Define the follow-up. How and when the decision will be checked (a metric and a date), and what would trigger a revisit.
11. Record the decision. After the Approver decides, add the outcome to the decision log (template in references/memo-template.md) and inform the Informed group with a short note drafted for approval.

Judgement calls

• Urgent decisions (hours, not days): a half-page memo with the decision, options, recommendation and reversibility beats no memo.
• The approver wants a recommendation you think is wrong: write it as asked, and record your concern in the dissent section under your name.
• Not enough data: say what is missing, how long it would take to get, and whether deciding now is better than waiting. For two-way doors it usually is.
• Decisions about people (restructures, dismissals): the memo covers the business case only. Individual cases go to a qualified HR professional and the lawyer before any decision.
• Commitments above the owner's usual authority: name who must also approve (board, lawyer, finance) in the header.
• Revisiting an old decision: link the original memo and state what has changed since.

Weak and strong decision statements:

• Weak: "Thoughts on the office?"
• Strong: "Decide by 30 Oct whether to sign a three-year lease for a 20-desk office in Berlin (one-way door) or keep hiring remotely in Germany for 12 more months (two-way door)."

Output

• decisions/<yyyy-mm-dd>-<slug>.md using the memo template: header with roles, decision statement, recommendation up front, context, criteria, options with assessment table, pre-mortem, dissent, follow-up.
• One row in decisions/log.md after the decision.
• A short announcement draft for the Informed group:

Decision: <what was decided>, decided by <Approver> on <date>.
Why: <one or two sentences>.
What changes for you: <impact on this audience, with dates>.
We will review it on <date> using <metric>. Questions to <Driver>.

Checks before you finish

• The decision is one sentence with a deadline and one named Approver.
• Criteria were agreed before options were scored.
• At least two real options plus the do-nothing baseline.
• Every cost and number has a source or is labelled as an estimate with its assumption.
• Reversibility is stated.
• Dissent is represented fairly and its holder agrees with the wording.
• Recommendation is up front and the memo is two pages or less before appendices.

Pitfalls

• The decision already made. A memo written to justify a choice persuades nobody and hides the risks. Score honestly.
• Straw-man options. Including two absurd alternatives to make the favourite look good is spotted immediately.
• Missing the baseline. "Do nothing" has costs and benefits too. Show them.
• Precision theatre. A spreadsheet to three decimal places on guesses. Use ranges.
• No owner. "The leadership team decided" means nobody decided. One Approver.
• No follow-up. Without a check date, the organisation never learns whether its decisions were good.
• Judging decisions by outcomes alone. A good decision can turn out badly through luck. At the check date, ask whether the reasoning and information were sound, not only whether it worked.

See also: board-and-investor-update when the decision goes to the board, and stakeholder-update for announcing it.