How it works
Look at the Evidence.Make Changes.Check Whether It Actually Worked.
Five steps
From Context to Memory
Understand the context
Fix the questions, models and market conditions; bring spend, outcomes, AI answers and customer signals into one context.
Facts / inferences / unknowns evidence panel
Analyze the causes
Test competing explanations and find the information gaps you control, with sources.
Hypotheses, each with its evidence level
Decide the action
Written as a Decision Card: one change, stop conditions, approver, outcome window. You approve; the system or your team executes.
Decision Card · awaiting approval
Check the result
After the outcome window, verify on three levels: did it ship, did AI see it, did results move — and check side effects and other causes.
Three-level verification
Update the memory
The conclusion, with its context, evidence, action and applicability, enters Decision Memory — checked first before the next decision.
Saved to Decision Memory
How results are judged
Three Levels of Verification, Three Possible Conclusions.
A successful API call isn’t the finish line; a verified external result and a completed outcome window are. AI visibility, inquiry quality and pipeline progress are assessed separately, with other causes checked.
Three levels
- Did it ship?
- Did AI see it?
- Did results move?
Three conclusions · equally weighted, each reported as it is
- Improved
- Regressed
- Inconclusive
The role of people
AI Does the Variable Work. You Own the Facts, Goals, Risk and Approvals.
What AI does
- Organize facts, inferences and unknowns across channels
- Propose hypotheses as an approvable Decision Card
- Execute through official APIs once approved
- Verify external results after the outcome window
What you own
- Data authorization: read-only first, least privilege
- Goals and stop conditions
- An approver for every change
- Reversible, auditable execution records