Decision Memos

This article is short on theory and long on the artefact. We define what the memo is and isn't, walk through the eleven sections of the template, and end with a full Bean & Basket sample memo that pulls evidence from every prior Part.

The Executive Question

What is the one-page document that turns analysis into action — and what does it look like when it's ready to sign?

The memo is the synthesis. Every other artefact feeds it; nothing replaces it.

What the Memo Is

A one-page document that names:

• The decision being recommended.
• The context that prompted it.
• The evidence supporting it, drawn from every relevant evidence language.
• The counterfactual — what happens if we don't act.
• The uncertainty — what could be wrong.
• The recommendation — the action, the threshold, the owner.
• The next test — what we will measure to know whether the action worked.
• The open questions — what this memo did not answer.

That's it. Eleven sections in the template below; one page when each section is a short paragraph. The discipline is brutal: brevity forces clarity.

What the Memo Isn't

A memo is not:

• A research report. (That's a different document; it can be appended.)
• A model card. (That's an artefact; the memo references it.)
• A slide deck. (That's a presentation; the memo is the source.)
• A status update. (That's a recurring email; the memo is the artefact you sign.)
• A justification. (The memo precedes the decision; a justification follows it.)

The memo is the page the sponsor reads before approving the action. Everything else lives downstream or upstream of it.

The Template

Three sections deserve a second look:

The counterfactual section. This is the §5.2 question made operational. What happens if we don't act? If the team cannot answer this in a paragraph, the memo isn't ready. Naming the counterfactual is what separates a real decision from a one-sided pitch.

The uncertainty section. This is where assumptions live. Sensitivity to costs, to model accuracy, to the holdout, to the unknown unknowns. The temptation is to skip it because it complicates the recommendation. The discipline is to write it because it forces the team to know what could be wrong.

The next test. Every action should produce learning. A memo without a follow-up test commits the firm to a one-way decision; a memo with one converts the action into the next iteration of the learning loop. Section §17.3 builds on this.

A Worked Memo: Bean & Basket Retention

The strongest way to show the memo is to write one. Below is a full Bean & Basket memo from July 2026 that pulls evidence from every Part of the book — descriptive, visual, causal, predictive, AI, and operational.

A short tour of where each section pulls its evidence:

• Descriptive evidence comes from the Part I/II monitoring layer — the weekly sentiment chart from §13.4 and the churn-rate metric from §2.4.
• Causal evidence comes from §7.1's difference-in-differences logic applied to the 2025 pilot.
• Predictive evidence comes from §10.1's churn model and §10.2's threshold–profit curve.
• AI / unstructured evidence comes from §14.3's embedding clustering and §14.4's construct measurement.
• Counterfactual construction uses the §5.2 framing and the §5.3 holdout design.
• Next test and monitoring are the §16.4 governance card applied to this specific action.

Every claim in the memo traces to an artefact. Every artefact traces to a chapter. The memo is the apex of the artefact family from §0.2.

How the Memo Gets Written

The honest version of the writing process:

1. Start with the decision. The recommendation goes at the top. If the team can't write it as a single sentence, they aren't ready to write the memo.
2. Work backwards to the evidence. What would a sponsor need to know to believe the recommendation? Each section answers one of those needs.
3. Run the counterfactual gauntlet. Ask explicitly: what if we do nothing? What if we do the next-best alternative? What if the assumptions are wrong?
4. Pull the artefacts. Each evidence section references a card, a chart, a model, or a dashboard the team has already produced. The memo is the cover sheet, not new analysis.
5. Trim. A two-page memo is rarely better than a one-page memo. The cuts are usually in the evidence sections (which want to be exhaustive) and in the recommendation (which wants to hedge).
6. Read it aloud. If a sentence sounds like a hedge, it is. Rewrite or delete.

A memo that survives this process is one the sponsor can sign in a five-minute read.

Communication Patterns

Five habits that separate memos that get read from memos that get filed:

• Lead with the recommendation. Not with the methodology. Not with the data. The sponsor's first question is "what should we do?"; the memo answers it in the first paragraph.
• One chart, not three. Pick the single chart that does the most work. If a second chart is essential, the memo is doing two things; split it.
• Name the threshold for changing the recommendation. Under what evidence would we reverse course? This is the inverse of the threshold for acting and equally important.
• Use the word "we" sparingly. "We recommend" is fine. "We feel that we believe that we should consider" is six words too long.
• Date everything. A memo without a date is a memo nobody will trust in six months.

What Separates an Executive Memo From a Research Write-Up

Three differences worth naming explicitly:

Both documents have a place. A team that ships only research write-ups will be respected and ignored. A team that ships only memos will move fast and lose the audit trail. A working organization ships both, with the memo on top and the research write-up linked from it.