Review an AI-assisted management output against the decision it needs to support, then inspect the prompt for missing evidence, constraints and an explicit standard for useful work.
Start with the decision
Write down who needs to decide, what choice is due and what an accepted output would help them do. A fluent answer can still miss that decision. This is an editorial review method, not a report of measured performance.
Use the five-step review
The Upskill Decision Loop makes the review concrete:
- Signal: Can you trace each observation to a source, period and definition? Identify missing comparisons.
- Structure: Does the output distinguish a possible explanation from an established cause? Name the evidence that could change it.
- Options: Can you compare commitments, including doing less or waiting for better evidence? Check costs, constraints and trade-offs.
- Pilot: Is the next step bounded, assigned and subject to manager approval? A proposed action is not an accepted commitment.
- Measure: Will the review show whether to extend, revise or stop? Include quality, checking and rework alongside effort.
Inspect a small illustration
Suppose a synthetic monthly pack reports revenue of 110 against the previous month’s 100, and positive operating cash flow of 10 against 30. These normalized figures are authored examples, not customer data or product output.
Revenue increased 10%. Positive cash flow decreased by 20; it did not become negative. Without invoice ageing, payment records and other cash movements, the answer cannot establish that late collections caused the change.
A useful brief names the unresolved question and requests the evidence. It can propose a receivables review; a manager still approves the scope and any customer contact. The cash-collection illustration shows what that evidence register and next step could contain.
Revise the prompt after the review
Supply the missing source definitions, the decision owner, constraints and acceptance criteria. Ask for observations, hypotheses and proposed actions to remain distinguishable. Review the revised output with the same criteria rather than searching for a phrase that sounds more authoritative.
Keep the corrections with the brief. If another teammate continues the work, share the authorized context and approved summary deliberately; do not assume the original private conversation is shared.