1.Begin with the decision
We define the choice, the decision owner, the deadline, the consequence of delay, and what a useful answer must enable. A broad topic such as “healthcare AI” is not yet an intelligence question. “Should this hospital invest in an AI-assisted medication reconciliation workflow during the next budget cycle?” is.
2.Build a source plan
Sources are selected for relevance, authority, freshness, independence, and proximity to the underlying event. The plan may include public records, company materials, regulatory documents, research, patents, market data, expert interviews, and client-provided information.
3.Move from evidence to interpretation
Analysts compare signals, identify patterns, map capabilities, test alternative explanations, and develop scenarios. Each material claim should be traceable to evidence or clearly labeled as interpretation.
Five layers remain distinct
Fact: what the evidence directly supports.
Interpretation: what the fact likely means in context.
Assumption: what must be true for the analysis to hold.
Scenario: a plausible future and its trigger conditions.
Recommendation: the proposed action based on the first four layers.
See these layers in action in the Evidence Map and the Scenario Studio.
4.Express confidence honestly
Confidence reflects source quality, source agreement, evidence completeness, and the stability of the environment. It is not a claim of certainty. Material uncertainty is written into the recommendation.
Read the confidence framework5.Deliver for the decision moment
The final output is structured so a leader can see the decision, recommended move, supporting evidence, alternatives, risks, dependencies, and next milestones without searching through the document.
- Executive answer
- Decision context
- Evidence base
- Scenarios and triggers
- Recommendation and rationale
- Risks and counterarguments
- 30, 60, and 90-day actions
6.Update when the picture changes
Some decisions require ongoing monitoring. New evidence may increase or decrease confidence, invalidate an assumption, activate a scenario trigger, or change the recommended action. Updates should show what changed from the previous version.
Bring us a decision