Start with outcomes, not capabilities
'Does the supplier have ISO 9001?' is a capability question and tells you almost nothing. 'What was their on-time delivery rate to their three largest US customers in the last 12 months?' is an outcome question and tells you almost everything.
Outcome questions also produce data AI can actually compare across vendors — capability questions produce marketing copy.
Weighting: pick five dimensions, not fifteen
Across 1,400+ RFx events on our platform, the highest-quality awards used 5-dimension scorecards. Anything more dilutes the signal.
- Price (25–35%) — total landed cost in USD, not list price
- Delivery performance (20%) — outcome metric with evidence
- Quality (20%) — defect rate per 1,000 units or equivalent
- Financial stability (15%) — D&B or equivalent
- Strategic fit (10–20%) — references and continuity risk
Where AI helps
AI is genuinely good at: extracting structured fields from supplier responses, flagging contradictions across answers, summarizing 80-page proposals into evidence packs, and surfacing missing or vague answers for follow-up.
Where AI quietly hurts
AI is bad at — and you should never let it own — final scoring, weighting decisions, qualitative cultural fit, and red-flag judgment calls. Keep humans in those seats. Treat AI scores as inputs to the conversation, never outputs of it.