What should a marketing AI agent do with customer interviews?

The useful output is a traceable account of what respondents said, how themes differ between segments and where the evidence conflicts. A list of persuasive phrases alone is not enough to support positioning decisions.

Revensi’s Audience Insight Synthesis blueprint counts respondents rather than repeated mentions and retains contrary evidence. Give it attributed notes or transcripts and a clear definition of the groups you want to compare.

Use a compact evidence record:

• Observation: what the respondent actually described.
• Context: their segment, role and relevant circumstances.
• Interpretation: the proposed meaning of the observation.
• Limit: contradictory evidence or a question the sample cannot answer.

These fields separate source material from analysis. They also let a reviewer revisit an interpretation without searching every interview from the beginning.

Treat one person's repeated comments as one respondent's evidence, not several independent endorsements. Frequency within a conversation may show emphasis, but it does not increase the number of people represented.

In a hypothetical set of interviews, one operations manager mentions reporting delays repeatedly while several other respondents describe different problems. The synthesis should not label reporting delays the universal priority merely because that phrase appears most often.

Compare the evidence in context:

Keep these distinctions beside the theme. A caveat in a separate appendix is easy to lose when a phrase is copied into a campaign brief.

Use a theme in positioning when it describes a relevant buyer problem and the product evidence supports the proposed response. The Positioning Evidence Builder combines approved product claims with segment-specific needs and keeps proof gaps visible.

A customer saying that handoffs are frustrating does not establish that your product eliminates handoff errors. The positioning draft needs a supported explanation of what the product does, for whom and under which conditions.

Before approving a message, check:

• The problem is supported by attributable evidence.
• The promised capability exists in the approved product information.
• Comparisons with alternatives use comparable evidence.
• The wording does not turn an aspiration into a measured result.

Retain the customer's vocabulary where useful, but do not imply permission to publish a quote merely because it appears in internal research notes.

Pass the approved positioning, evidence boundaries and unresolved hypotheses into the campaign plan. The Campaign Brief Orchestrator translates an idea into audience, message, budget, tasks and owners.

For a hypothetical campaign, a supported theme about reporting delays might justify testing a message focused on clearer review queues. Record that as a proposed message, alongside the outcome the test will examine. Do not promise that the theme will produce a particular conversion rate.

Review the chain from interview to campaign: can the team trace the final claim back to the original evidence and approved capability? If not, return it to the appropriate owner before scheduling production.

Use interview agents to make evidence easier to inspect, then use that evidence to shape bounded positioning and campaign hypotheses. Preserve disagreement and uncertainty throughout the handoff.

Choose one segment and one proposed message. Build a short evidence record for it, confirm the product proof and agree what a campaign test would need to establish before expanding the claim.