Draft customer case studies from call transcripts, email and notes
This pipeline gathers call recordings, email threads and project notes about a client, pulls out the facts and quotes, and drafts a case study for an editor to shape. People still choose the story and get the client's sign-off.
Paste this into ChatGPT, Claude or your agent tool. See the prompt
- Content and Collateral
- intermediate
- Sales leader
- Founder
- CS
- Category
- Content and Collateral
- Difficulty
- Intermediate
- Est. cost
- Not stated by source
Works with
Problem
A case study is pieced together from many sources (client emails, call recordings, project notes and client documents), so it takes days of team time and often never gets written.
Setup
An n8n pipeline, triggered by a chat command or webhook, pulls the client's email thread, transcribes call recordings, and collects tagged pages from Notion, Confluence or Google Docs into one markdown file. An LLM extracts structured facts: names and titles tied to their source, metrics with period and units, key dates and attributed quotes. A person picks the angle and agrees it with the client, then the LLM drafts the narrative in house style, and an editor fact-checks, edits and gets client sign-off.
Prompt
Starter prompt by The GTM Agent Weekly, written from the setup described on this page. It is not the source's own prompt; see the source link below for their write-up.
# ROLE
You help produce a customer case study. Stage 1: extract verified facts. Stage 2 (after a
person picks the angle): draft the story in house style.
# INPUTS
- {{SOURCE_PACK}}: one markdown file with the client email thread, call transcripts and
tagged notes pages, each section labeled with its source.
- {{HOUSE_STYLE}}: our case study style guide and one example.
- {{ANGLE}}: the angle agreed with the client (stage 2 only).
# STAGE 1: FACT SHEET
Extract only what is stated, each with its source label:
- People: name, title, company.
- Metrics: value, unit, period, and what it measures.
- Dates: key events.
- Quotes: exact words and who said them.
- Possible angles: two or three, with the facts that support each.
# STAGE 2: DRAFT
Write the case study for {{ANGLE}} in {{HOUSE_STYLE}}: challenge, approach, results, quote.
Use only facts from the fact sheet. Mark each metric and quote with its source label.
# GUARDRAILS
- Never round, combine or estimate numbers. Never put words in someone's mouth.
- Anything unclear goes in "Questions for the client".
- An editor fact-checks and the client signs off before publishing. Replace every {{PLACEHOLDER}} before running. Works with any capable LLM and runner.
Tools
Steps
- Trigger the run from a chat command or webhook for one client.
- Gather the email thread, call transcripts and tagged notes and documents into one file.
- Have the LLM extract facts, metrics with context and attributed quotes as structured data.
- A person chooses the 3 to 5 moments to feature and agrees the structure with the client.
- Generate the draft, then fact-check against sources, edit by hand and get client sign-off on quotes and numbers.
Results
Vendor describing its own pipeline; it reports that the gathering and drafting stages get much faster while structure, editing and client sign-off do not. Self-reported.
What goes wrong
Gotchas
- Transcript quality depends on call audio, and email parsing misses attachments and embedded tables.
- Letting the model choose what matters produces generic case studies; the angle has to be a human decision.
- Fact extraction is not perfect, so every number and name needs checking against the source before the client sees it.
- Quotes and sensitive numbers need the customer's explicit approval before anything is published. (editor note)
Get 5 like this every week, free.
Real agent setups with sources and failure modes, in one short weekly email.
Free. One email a week. Unsubscribe anytime.
Source and credit
Slava Pospelov, Kelva · vendor
Source dated 2026-05-16 · last reviewed 2026-10-11. Summarized in our own words; see the source for details.