Persona insights analysis

Positioning and copy in service companies are usually written from the seller's memory of calls. Memory keeps the dramatic quote and loses the boring one that came up in eight calls out of ten. This skill reads the transcripts, codes every buyer statement, counts it, and only then draws conclusions.

Answer in the user's language. Quotes stay in the language they were spoken in.

Inputs (ask once, only what is missing)

1. Transcripts: pasted text, a CSV export, DOCX or PDF files, or a call-recording connector if one is connected in this environment (for example Fireflies, Fathom, Gong, tl;dv, Claap). Ask for a date range or tag if pulling from a connector.
2. Which decision this should inform: messaging, ICP, offer, or all three.
3. Known personas to group by, or permission to infer them from the data.
4. Anonymisation: may company names appear in the report? Default: no, use role and company type.
5. Format: markdown report (default), Notion page if a Notion connector is available, or a DOCX via the docx skill if available.

Call metadata, if the user has it, makes the analysis much stronger: call stage, outcome (won, lost, open, no decision), lead source (outbound, referral, inbound).

Step 1. Normalise every call into one record