Give leadership a competitive brief they can question and verify

State the decision before collecting comparisons

Write the decision in one sentence: for example, whether to prioritize a reporting feature for the next customer segment. Specify the segment, product scope and information cutoff. A general ranking of competitors is likely to mix markets and use cases; a decision-led brief tells you which differences actually matter.

Choose comparison dimensions before reading model summaries. Separate public capability, commercial packaging, documented limitations and your own strategic interpretation. If a vendor does not publish an answer, mark it unknown. A blank cell should not become a disadvantage simply because another vendor describes its product more extensively.

Create an evidence pack that survives a follow-up question

For each vendor, collect permitted primary-source excerpts with their URLs, publication or update dates where available, and your access date. Record the product and plan each source covers. Keep a source’s statement separate from an inference you draw from it. NIST’s Generative AI Profile recommends reviewing and verifying sources and citations in generated outputs; a model-created reference should not bypass that step.

Use labels with defined meanings: documented, directly tested under recorded conditions, inferred, contradicted or unknown. These describe the basis of a statement. Avoid turning them into an arbitrary percentage of confidence, and do not use the number of agreeing models as an evidence score.

• NIST: checking sources and citations in AI outputs↗

Ask for competing readings of the same pack

In Colay, ask available agents separately to examine customer relevance, positioning, implementation constraints and the strongest objection to your proposed recommendation. Give them the same evidence pack. A disagreement is useful when you can locate the assumption that produced it; differences in writing style alone add little to the decision.

After comparing responses, you can ask Consensus to synthesize the argument while explicitly retaining unresolved points. Treat that synthesis as a draft. Check that a minority objection has not disappeared merely because the final paragraph sounds smoother without it.

Prepare a competitive decision brief using only this evidence pack: [sources with IDs]. Decision and segment: [scope]. Information cutoff: [date]. Separate documented facts, interpretations, unknowns and recommendations. Compare vendors on these dimensions: [dimensions]. For each interpretation give the supporting source IDs, the strongest alternative explanation and what new evidence would change it. Preserve conflicting evidence. Do not invent missing features, claim absence from silence or use model agreement as factual confirmation. End with a decision request and a short research backlog.

Show how uncertainty changes the recommendation

Suppose a fictional evidence pack covers three vendors: A documents scheduled reports, B documents manual export, and C’s supplied materials do not discuss reporting. The team is considering a reporting feature. The following table illustrates reasoning from those invented inputs; it is not a market finding.

Write a one-page conclusion with an inspectable appendix

Do not bury a decision-changing gap in a footnote. If the recommendation depends on customer demand you have not established, make that the next investigation instead of disguising it as competitor analysis. Colay helps organize interpretations and produce a coherent draft; your team remains responsible for verifying the evidence and deciding what it justifies.

1. Lead with the decision requested and the most relevant documented observation.
2. Explain the recommendation as an interpretation, including its main assumption.
3. Show one serious alternative and the unresolved fact most likely to change the choice.
4. Attach the source table, research owner and date for revisiting the brief.

Questions, answered