ethos
← Back to Projects
🎙️

Intake

Client intake calls to structured requirements

Thesis

Client intake calls are unstructured and notes get lost. Transcription APIs are mature enough to extract structured requirements automatically, saving 2+ hours per client.

The Problem

Discovery calls are where projects go wrong before they start: Someone takes notes (maybe). Requirements get scattered across emails, docs, and memory. By the time you write the SOW, you've forgotten half of what was discussed. Gaps don't surface until the project is underway. The irony? The client told you everything you needed. You just didn't capture it in a usable way.

Implementation Approaches

Approach 1

Meeting Bot + Extraction (Recommended)

Recommended

Bot joins calls, transcribes, and extracts structured requirements

⏱️2 weeks
📊High complexity

Implementation

  • Zoom/Meet/Teams bot joins scheduled intake calls
  • Real-time transcription via Deepgram or Assembly
  • Claude extracts: goals, requirements, constraints, timeline, budget signals
  • Flags gaps: 'Budget not discussed', 'Timeline unclear'
  • Outputs draft SOW sections ready for review

Pros

  • +Captures everything, not just what someone remembers to write
  • +Flywheel: learns what good intake questions look like over time
  • +Connects to Guildry's scoping module directly
  • +High value per call justifies premium pricing

Cons

  • Meeting integrations are complex
  • Privacy concerns, some clients won't want recording
  • Depends on call quality and clear discussion
Approach 2

Post-Call Upload

Upload recordings after the fact for processing

⏱️1 week
📊Low complexity

Implementation

  • Upload audio/video from any source
  • Same extraction pipeline, no live integration needed
  • Works with existing recording workflows
  • Can process historical calls to build training data

Pros

  • +Fastest to build, no meeting integrations
  • +Works with any recording tool or phone call
  • +Lower privacy friction, user controls what gets uploaded
  • +Can process backlog of past calls

Cons

  • Extra step, users might not bother
  • No real-time value during the call
  • Harder to build habit
Approach 3

Guided Interview Tool

AI-guided intake form that asks the right questions

⏱️1 week
📊Medium complexity

Implementation

  • Send client a link before/after the call
  • AI asks clarifying questions based on project type
  • Combines with call notes for complete picture
  • Generates structured brief automatically

Pros

  • +No recording needed, pure async
  • +Client does some of the work
  • +Can use before call to prep, or after to fill gaps
  • +Reuses Brief's conversation patterns

Cons

  • Requires client participation
  • Doesn't capture the nuance of live conversation
  • May feel redundant if call happened

Validation Plan

Hypothesis to Test

Agencies will pay $29/mo to turn intake calls into structured SOW drafts automatically

Validation Phases

1

Manual Extraction

1 week
  • Get 5 intake call recordings from agency contacts
  • Manually extract requirements using Claude
  • Generate draft SOW sections
  • Show output to agencies, gauge reaction
Strong positive reaction, 'this would save us hours'
2

Upload MVP

1 week
  • Build simple upload and extraction flow
  • Partner with 3 agencies for pilot
  • Process their real intake calls
  • Measure: time saved, quality of extraction
2+ hours saved per call, agencies want to keep using it
3

Pricing + Integration

2 weeks
  • Add meeting bot integration for one platform
  • Propose pricing: $29/mo for 10 calls, $79/mo unlimited
  • Connect output to Guildry's project creation
  • Track: calls processed, conversion to paid
3+ paying agencies, clean handoff to Guildry

Kill Criteria

Stop and move on if any of these become true:

  • Extraction quality too inconsistent for real use
  • Agencies don't actually process enough calls to justify subscription
  • Privacy concerns kill adoption
  • Existing transcription tools are 'good enough'