The pipeline from Part 6 is not a demo. It runs. This is what it runs on.

Same structure every time: audio in, transcribe, extract, deliver. What changes is the prompt and the destination. That’s it. Four use cases. All real.

This is the part most tools skip.

People show transcription. Then stop.

I care about the step after transcript. That is where value is created.

Sales Call → Draft Email

Input: A 40-minute discovery call. Two people talking. One of them is you.

Steps:

1. Record the call. Whisper transcribes it.
2. Send the transcript to the extraction step with this prompt: “From this sales call transcript, extract: (1) the prospect’s stated pain, (2) what was offered, (3) agreed next steps, (4) any objections raised.”
3. Send the extracted JSON to a second prompt: “Write a follow-up email using these notes. Professional, short, no fluff.”

Output:

Subject: Following up — [pain point]

Hi [Name],

Good talking today. You mentioned [pain]. We discussed [offer].

Next step: [action] by [date].

Let me know if anything changed.

[Your name]

That email is in your inbox before the call window closes. You didn’t write it. You didn’t forget anything. The call happened, the pipeline ran.

No CRM entry required either. Add one more step. Send the extracted JSON straight to HubSpot or Pipedrive. The deal updates itself.

Team Meeting → Slack + Notion

Input: A 60-minute team meeting. Six people. No one is taking notes.

Steps:

1. Record. Transcribe.
2. Extract: “From this meeting transcript, extract action items. For each item: who is responsible, what is the task, what is the deadline if mentioned.”
3. Extract again in a separate pass: “Write a 3–5 sentence meeting summary. What was decided. What was not decided. What’s next.”
4. Post summary to Slack. Post action items to a Notion database.

Output (Slack):

Reviewed Q1 pipeline. Decided to pause the enterprise tier launch until March 30. Design to finalize onboarding flow by EOW. Pricing doc still open — decision next Tuesday.

Output (Notion):

Owner	Task	Due
Maria	Finalize onboarding designs	Friday
Tom	Draft revised pricing doc	Tuesday
Dev team	Pause enterprise launch branch	Today

Nobody took notes. Nobody will send a follow-up asking what was decided. The meeting ended. The record exists. People know what they own.

The pipeline doesn’t care how many people were on the call. It doesn’t get tired. It doesn’t miss the thing that was said in the last five minutes when everyone’s attention was already gone.

Research Call / User Interview → Structured Note

Input: A 30-minute user interview. You’re asking questions. The user is talking.

Steps:

1. Record. Transcribe.
2. Extract: “From this user interview transcript, extract: (1) key quotes — verbatim, (2) main themes, (3) objections or frustrations mentioned, (4) any feature requests or workarounds the user described.”
3. Format as a structured research note. Save to your research database.

Output:

## Interview: [User name / segment]
Date: [date]

### Key quotes
- "I always end up going back to the spreadsheet because I can't find what I need in the app."
- "The onboarding took me three days. I figured it out eventually."

### Themes
- Trust gap with automated suggestions
- Onboarding friction for non-technical users
- Manual fallback behavior

### Objections
- Skeptical of AI-generated outputs without source visibility
- Doesn't want another tool that "learns" without showing its work

### Feature mentions
- Requested: export to CSV from any view
- Workaround described: copy-paste to Google Sheets weekly

That note is ready before your next meeting. You can pull quotes directly into a deck. You can tag themes across twenty interviews and start seeing patterns.

The alternative is a page of your own shorthand that you’ll half-remember in two weeks.

Voice Memo → Structured Note

Input: You’re walking. You have a thought. You open your phone and talk for two minutes.

Steps:

1. Voice memo recorded. Sent to the pipeline automatically — via Shortcut, via folder watch, however you’ve set it up.
2. Transcribe.
3. Extract: “This is a voice memo. Clean up the transcript. Remove filler words. Organize the thought. If there’s an action item, label it. If it’s a question, label it. Return a clean note.”
4. Save to Obsidian, Notion, Apple Notes — wherever.

Output:

## Note [timestamp]

The pricing page problem isn't copy. It's structure. Visitors are landing on it from the blog, not the homepage. They don't have context. The page assumes they've already seen the value prop.

**Action:** Redesign pricing page flow for cold traffic. Consider adding a one-liner above the fold that explains what the product is.

**Question:** Does the current analytics setup distinguish homepage traffic from blog traffic on pricing page views?

You said that in two minutes while walking. It landed formatted, tagged, ready to act on. You didn’t have to sit down and write it up. You didn’t lose the thought.

The voice memo becomes a first-class note. Not a scratchy recording you’ll never listen to again.

The Point

Four scenarios. One pipeline. What changed:

• The extraction prompt
• The output format
• The delivery destination

The audio is just input. It doesn’t matter if it came from a sales call or a walk. The model doesn’t know. It reads the transcript and does what the prompt says.

If you need a fifth use case, write a new prompt. Point it at the same pipeline. You’re done in twenty minutes.

The infrastructure doesn’t change. The shape of what comes out does.

---

Next: You Own the Pipeline. Or You Don’t..

• #nbp
• #pipelines
• #use-cases