It identifies whether you can commission a useful, controlled pilot—not whether fine-tuning will work. A high score means you have more of the evidence and ownership needed to test the decision; it is not a predicted ROI, compliance certificate or supplier recommendation.
This is a desk-designed rubric, not a validated psychological or statistical instrument. Complete it with the product owner and AI lead, then have finance and the relevant data/security reviewer challenge the answers. Use links to your own documents as evidence; do not award points for an intention to create them later.
The design draws on the practical emphasis on error analysis in Hamel Husain’s field guide, grader discipline in Anthropic’s evaluation guide, and documentation/risk considerations in NIST’s framework. The questions, weights and bands below are editorial choices.