A polished memo can make weak evidence look stronger than it is. That is one of the main risks of written operating culture. The artifact feels serious, the paragraphs flow, the recommendation sounds calm, and the organization mistakes narrative confidence for decision confidence.

Good memos separate evidence, assumptions, and confidence. Evidence is what the company has observed or sourced. Assumptions are what must be true for the recommendation to work. Confidence is the writer's current judgment about how much weight the decision can safely carry. Those three things belong near each other, but they should not be blurred together.

Evidence quality varies. A customer interview is not the same as a signed contract. A sales anecdote is not the same as a pattern across twenty deals. A benchmark is not the same as production behavior. A competitor claim is not the same as buyer pull. A model-generated summary is not the same as the source. A useful memo tells the reader what kind of evidence is being used and where it is thin.

Assumptions deserve more respect than they usually get. Every important decision contains them. The buyer will care enough. The team can execute. The migration will be manageable. The partner will stay aligned. The model will improve. The regulation will not block us. The customer will accept the workflow change. The question is not whether assumptions exist. The question is whether the memo names the ones that matter.

A strong memo highlights load-bearing assumptions. These are the assumptions that would change the recommendation if they proved false. If the entire argument depends on enterprise buyers accepting a new procurement route, say that. If the plan depends on a fragile integration, say that. If the economics require a support load that the team has never handled before, say that.

Confidence should be explicit. The memo does not need fake precision, but it should give the decision-maker a sense of how certain the author is and why. High confidence with strong evidence is different from high confidence based on founder judgment. Low confidence does not always mean the decision should wait. Sometimes the right move is a small experiment because uncertainty is high and the cost of learning is reasonable.

This is where decision memos can improve risk-taking. Companies often treat uncertainty as a reason to keep talking. A better memo asks what decision is appropriate for the current evidence. Full commitment, staged commitment, reversible test, deeper research, or deliberate no. The evidence standard should match the commitment size.

Evidence also has time decay. A memo written six months ago may contain customer facts, market conditions, pricing assumptions, or technical constraints that no longer hold. Written culture should make old reasoning retrievable, but it should not turn old reasoning into scripture. Decision logs need review dates for decisions whose evidence may expire.

AI complicates this layer. It can help gather source material, summarize interviews, find contradictions, and draft evidence sections. It can also flatten source quality and make unsupported claims sound reasonable. Any AI-assisted memo should preserve links to source material, distinguish model synthesis from observed fact, and make the human owner accountable for the final judgment.

A practical memo can use a simple evidence table: claim, source, strength, uncertainty, and implication. Not every decision needs that much structure, but high-consequence decisions benefit from it. The table prevents the writer from hiding a fragile argument inside confident prose.

The memo should also name missing evidence. What would we want to know but do not know yet? Why are we deciding anyway? What is the cost of waiting? This is how the organization avoids the fantasy that every decision can be fully informed. The point is not perfect information. The point is honest information.

The best reviewers focus on evidence quality before word choice. They ask whether the memo has enough signal to support the recommendation. They challenge the strongest assumption. They add missing context. They distinguish a writing problem from a thinking problem. That review discipline is what keeps written culture from becoming document polish.

Confidence after the decision matters too. The decision log should record whether the choice was high-confidence, judgment-led, experiment-led, or time-constrained. Later, when the organization reviews the outcome, it can learn the right lesson. A low-confidence experiment that fails is different from a high-confidence strategy that fails. The first may have been a useful learning move. The second may reveal a deeper judgment problem.

Evidence, assumptions, and confidence make decisions more honest. They do not remove uncertainty. They prevent the company from hiding it from itself.

One simple practice is to mark each major claim as observed, inferred, or believed. Observed means the company has direct evidence. Inferred means the company is extrapolating from related evidence. Believed means the claim is a judgment call that still needs reality contact. That small distinction can change the entire discussion.

For example, "three enterprise prospects asked for this workflow" is observed. "Enterprise buyers will pay more for this workflow" may be inferred. "This workflow should become the center of the roadmap" is a belief until the company has stronger proof. A memo that separates those statements gives the decision-maker a cleaner view of the risk.

The review date should match the confidence level. A high-confidence infrastructure decision may need a longer review cycle. A low-confidence market bet may need a thirty-day learning checkpoint. Written culture improves when confidence changes the operating cadence, not just the wording of the memo.

Evidence note: this post draws on local Right Depth for the Problem and AI Operating Structure themes, plus public written-work examples including https://handbook.gitlab.com/.

This is part 4 of 10 in Decision Memos and Written Operating Culture.