Verify AI comparisons and cited evidence

Comparison AI can summarise score patterns, cross-reference a shortlist and highlight themes in available evidence. Its result is a review aid. The source responses, scores, commercials, risks and accountable human rationale remain authoritative.

Before you start

• Wait until the relevant questionnaires and scoring tasks are substantially complete.

• Confirm all vendors being compared have equivalent released forms and that missing evidence is visible as missing, not silently treated as failure.

Step by step

1. Check evidence completeness first

Review vendor response status, evaluator completion and stream coverage. Recalculate scores only when you intend to refresh derived totals.

2. Run the focused analysis

Use the available score summary, comparison analysis or shortlist cross-reference action on the Compare page.

3. Separate findings from inference

Identify which statements are supported by a response or score and which are AI interpretation. Treat uncited claims as prompts for investigation.

4. Open the source evidence

Inspect the cited answers, comments, scorecards, proposal versions and credit or risk records before repeating a conclusion.

5. Record the human judgement

Use the Decision Record and business case to document what the team concluded and why, including gaps or disagreement.

What happens next

• A reviewed analysis can inform shortlist, vote and final selection without becoming an automatic decision.

• Re-run analysis after material new evidence, not simply to seek a preferred conclusion.

Troubleshooting

The analysis omits a vendor

Check that the vendor is active in the project and has comparable released/scored evidence.

A claim conflicts with the source

Use the source as authoritative, correct any underlying record and disregard or regenerate the draft.

The result is dominated by one stream

Review project stream weights, missing assignments and score completeness before interpreting totals.

Related guides