Implementation Guide: Transcribe client briefing calls and generate creative brief documents for team review

Hardware Procurement

Software Procurement

Prerequisites

• Active Zoom Pro/Business, Google Meet (Workspace Business Standard+), or Microsoft Teams license for all users who conduct client briefing calls
• Google Workspace Business Standard or Microsoft 365 Business Standard for calendar integration (Fireflies needs read access to user calendars to auto-join meetings)
• Slack workspace (Pro or higher) with admin permissions to create channels and install the Fireflies Slack app
• Asana or Monday.com account with admin permissions to create projects and install integrations (Business tier or higher for Asana; Standard tier or higher for Monday.com)
• Stable internet connection: minimum 10 Mbps symmetrical per concurrent video call participant; <150ms latency to cloud services
• Google Drive or OneDrive with sufficient storage (~2 GB/month for a 10-person agency with daily calls; ~25 MB per hour of recorded audio)
• Client workstations running Windows 10/11 or macOS 12+ with Chrome, Edge, or Firefox browser (latest version) for Fireflies web dashboard access
• Admin access to the agency's DNS and email for domain verification during Fireflies workspace setup
• OpenAI API account with billing enabled and a valid payment method (credit card) at https://platform.openai.com
• Zapier account (Professional plan or higher) with admin access to create and manage multi-step Zaps
• Firewall/network policy allowing outbound HTTPS (TCP 443) to: *.fireflies.ai, api.openai.com, *.zapier.com, hooks.slack.com, *.zoom.us, and *.googleapis.com
• Written client approval for the recording consent policy—all meeting participants must be informed that calls are recorded and processed by AI. Update client Master Service Agreements (MSAs) and engagement letters with recording/AI processing consent clauses before deployment
• Inventory of all recurring client meetings (name, cadence, participants, conferencing platform) to configure Fireflies auto-join rules
• Designated agency 'AI Champion'—an internal point of contact (typically an account director or operations manager) who will approve brief templates, provide feedback on AI output quality, and serve as the primary liaison during deployment

Installation Steps

Step 1: Provision Fireflies.ai Workspace and Configure Admin Settings

Create the agency's Fireflies.ai workspace, configure organization-level settings, and invite all users. This is the foundation of the entire solution—Fireflies handles call recording, transcription, and serves as the data source for downstream automation.

Step 2: Install and Configure Hardware Audio Devices

Deploy Jabra Speak2 75 speakerphones in conference rooms and Poly Sync 20+ units at individual workstations. Proper audio hardware dramatically improves transcription accuracy—built-in laptop microphones produce significantly worse results due to background noise and distant speaker pickup.

Step 3: Connect Fireflies.ai Calendar Integration for Auto-Join

Configure Fireflies to automatically detect and join scheduled client meetings via Google Calendar or Outlook calendar integration. This eliminates the need for users to manually invite the Fireflies bot to each call—the system works passively once configured.

Step 4: Configure Fireflies Native Integrations (Slack, Google Drive, CRM)

Connect Fireflies directly to the agency's existing tools so that raw transcripts and AI summaries are automatically distributed to the right channels. These native integrations run independently of the Zapier automation built in later steps.

Step 5: Set Up OpenAI API Account and Generate API Key

Create an OpenAI platform account, configure billing, set usage limits, and generate the API key that will be used by the Zapier workflow to call GPT-5.4 for creative brief generation.

curl https://api.openai.com/v1/chat/completions -H 'Content-Type: application/json' -H 'Authorization: Bearer sk-YOUR_KEY_HERE' -d '{"model":"gpt-5.4","messages":[{"role":"user","content":"Say hello"}],"max_tokens":50}'

Step 6: Design the Creative Brief Template and System Prompt

Create the structured creative brief template that defines what sections the AI should produce, and craft the system prompt that instructs GPT-5.4 how to transform a raw transcript into a completed brief. This is the most critical step for output quality—invest time in prompt engineering and iterate based on agency feedback.

Step 7: Build the Zapier Automation Workflow (Transcript → GPT-5.4 → Brief → Distribution)

Create the multi-step Zapier workflow (Zap) that triggers when Fireflies completes a new transcript, sends it to GPT-5.4 for brief generation, formats the output as a Google Doc, creates an Asana task, and posts a notification to Slack. This is the core automation that transforms raw transcripts into actionable creative briefs.

URL: https://api.openai.com/v1/chat/completions
Payload Type: JSON
Headers: Authorization = Bearer sk-YOUR_API_KEY | Content-Type = application/json

# map Fireflies fields to dynamic values

{"model":"gpt-5.4","messages":[{"role":"system","content":"YOUR_SYSTEM_PROMPT_HERE"},{"role":"user","content":"TRANSCRIPT: {{transcript_text}}\n\nMEETING TITLE: {{meeting_title}}\n\nATTENDEES: {{attendees}}\n\nDATE: {{meeting_date}}"}],"temperature":0.3,"max_tokens":4000}

📋 *New Creative Brief Generated*
*Meeting:* {{meeting_title}}
*Date:* {{meeting_date}}
*Brief:* {{google_doc_url}}
*Asana Task:* {{asana_task_url}}

_Please review within 48 hours._

Step 8: Configure Asana Project and Brief Approval Workflow

Set up the Asana project structure that receives generated creative briefs and enables a lightweight approval workflow. The goal is to ensure every AI-generated brief is reviewed by a human before being shared with the creative team.

Step 9: Implement Recording Consent and Compliance Procedures

Establish the legal and procedural framework for recording client calls. This is a non-negotiable compliance requirement—failure to obtain proper consent exposes the agency (and the MSP) to significant legal liability, particularly in two-party consent states like California, Florida, Illinois, and Pennsylvania.

Step 10: User Training and Phased Rollout

Train agency staff on the new system and roll out in phases to manage change and gather feedback before full deployment. Start with the account management team (primary users), then expand to creative and strategy teams (consumers of the briefs).

Step 11: Configure Monitoring, Alerting, and Ongoing Optimization

Set up monitoring to ensure the system continues operating correctly after deployment, and establish a feedback loop for continuous prompt improvement.

Custom AI Components

Creative Brief Generator System Prompt

Type: prompt The core system prompt that instructs GPT-5.4 to transform a raw meeting transcript into a structured creative brief document. This prompt defines the output format, extraction logic, and quality standards. It is used in the Zapier webhook call to the OpenAI API (Step 7, Action 3). The prompt is designed to handle various types of client briefing calls including discovery calls, campaign kickoffs, content briefings, and brand strategy sessions.

Implementation:

You are an expert creative strategist at a marketing agency. Your job is to transform raw meeting transcripts from client briefing calls into polished, structured creative brief documents that the creative team can immediately act upon.

INSTRUCTIONS:
1. Read the entire transcript carefully, identifying the client's needs, goals, and requirements.
2. Extract and organize information into the structured brief format below.
3. If information for a section is not explicitly discussed in the transcript, mark it as '[Not discussed — follow up required]' rather than inventing details.
4. Use professional, concise language appropriate for an internal creative team.
5. Resolve any contradictions in the transcript by noting both statements and flagging for clarification.
6. Include direct quotes from the client where they provide particularly clear direction or strong preferences.
7. Do NOT include internal agency discussion, small talk, or off-topic conversation in the brief.

OUTPUT FORMAT — Generate the brief using this exact structure:

# CREATIVE BRIEF
## {{Client Name}} — {{Project/Campaign Name}}
**Date:** {{Meeting Date}}
**Prepared by:** AI-Generated from client briefing call (requires human review)
**Meeting Attendees:** {{List all attendees identified in the transcript}}

---

### 1. PROJECT OVERVIEW
A 2-3 sentence summary of what the client is asking for and why. Include the type of project (campaign, content piece, brand refresh, etc.) and the core ask.

### 2. BUSINESS OBJECTIVE
What business outcome does the client want to achieve? Be specific: increase sales by X%, launch a new product, enter a new market, rebrand after acquisition, etc. Distinguish between primary and secondary objectives.

### 3. TARGET AUDIENCE
- **Primary Audience:** Demographics, psychographics, behaviors
- **Secondary Audience:** If mentioned
- **Audience Insights:** Any research, data, or anecdotal information the client shared about their audience
- **Current Audience Perception:** How the audience currently views the client's brand/product (if discussed)

### 4. KEY MESSAGES
- **Core Message:** The single most important thing the audience should take away
- **Supporting Messages:** 2-4 secondary messages that reinforce the core message
- **Tone/Voice Direction:** How the client wants to sound (e.g., authoritative, playful, empathetic)
- **Mandatory Language:** Any specific phrases, taglines, or terminology the client requires

### 5. DELIVERABLES
List all specific deliverables discussed, including:
- Format/medium (e.g., social media posts, video, print ad, landing page, email campaign)
- Specifications (dimensions, duration, word count) if mentioned
- Quantity of each deliverable
- Any platform-specific requirements (e.g., Instagram Reels, LinkedIn carousel)

### 6. BRAND GUIDELINES & CREATIVE DIRECTION
- **Existing brand guidelines:** Reference any brand books, style guides, or asset libraries mentioned
- **Visual direction:** Colors, imagery style, photography vs. illustration preferences
- **Examples/references:** Any competitor work, inspiration, or examples the client referenced (positive or negative)
- **Restrictions:** Anything the client explicitly does NOT want

### 7. TIMELINE & MILESTONES
- **Project start date:** 
- **Key milestones:** (e.g., first concepts, client review, revisions, final delivery)
- **Final delivery date:**
- **Hard deadlines:** Any immovable dates (e.g., product launch, event, seasonal window)
- **Review/approval process:** Who approves on the client side, expected turnaround for feedback

### 8. BUDGET INDICATORS
- **Stated budget:** Exact figures if provided
- **Budget range:** If the client gave a range or relative indication
- **Budget constraints:** Any cost sensitivities or allocation preferences mentioned
- **[Note: If budget was not discussed, write 'Budget not discussed in this call — follow up required']

### 9. COMPETITIVE CONTEXT
- **Key competitors mentioned:**
- **Competitive positioning:** How the client wants to differentiate
- **Competitor examples:** Any competitor work referenced (positively or negatively)

### 10. SUCCESS METRICS
How will the client measure success? Include:
- KPIs mentioned (e.g., impressions, click-through rate, conversions, brand awareness lift)
- Measurement tools or reporting expectations
- Benchmarks or targets

### 11. OPEN QUESTIONS & FOLLOW-UPS
List any unresolved items, missing information, or topics that need clarification before creative work begins. Format as a numbered checklist:
- [ ] Question 1
- [ ] Question 2

### 12. ADDITIONAL NOTES
Any other relevant information from the call that doesn't fit the above sections, including:
- Stakeholder dynamics (e.g., 'CMO has final say; marketing manager is day-to-day contact')
- Political/organizational context that may affect the project
- Client's emotional state or urgency level
- References to previous projects or agency relationship history

---
⚠️ **AI-GENERATED DOCUMENT — REQUIRES HUMAN REVIEW BEFORE DISTRIBUTION**
This brief was automatically generated from a meeting transcript. Please verify all details for accuracy, completeness, and nuance before sharing with the creative team or the client.

---

Zapier Webhook Payload Builder

Type: integration The complete JSON payload structure for the Zapier Webhooks by Zapier action that calls the OpenAI API. This integration component connects the Fireflies transcript data to the GPT-5.4 creative brief generator and handles the API communication.

Implementation

{
  "model": "gpt-5.4",
  "messages": [
    {
      "role": "system",
      "content": "[PASTE THE FULL SYSTEM PROMPT FROM THE 'Creative Brief Generator System Prompt' COMPONENT HERE]"
    },
    {
      "role": "user",
      "content": "Please generate a creative brief from the following client briefing call transcript.\n\nMEETING TITLE: {{1. Meeting Title}}\nMEETING DATE: {{1. Date}}\nATTENDEES: {{1. Attendees}}\nDURATION: {{1. Duration}} minutes\n\n--- TRANSCRIPT START ---\n{{1. Transcript Text}}\n--- TRANSCRIPT END ---\n\nGenerate the creative brief now, following the exact format specified in your instructions."
    }
  ],
  "temperature": 0.3,
  "max_tokens": 4000,
  "top_p": 0.95
}

Parsing the Response

In the next Zapier step, reference the GPT-5.4 response using the variable below. This contains the full generated creative brief as a text string.

{{2. Choices 0 Message Content}}

Error Handling

Add a Zapier 'Paths' step after the webhook and configure the following two paths:

• Path A (Success): If {{2. Status Code}} is 200 → continue to Google Docs creation
• Path B (Failure): If {{2. Status Code}} is not 200 → Send Slack DM to MSP admin: 'Brief generation failed for meeting: {{1. Meeting Title}}. Error: {{2. Body}}'

Rate Limiting

OpenAI rate limits for GPT-5.4 are generous (10,000 RPM on Tier 1). For agencies generating fewer than 50 briefs/day, rate limiting will not be an issue. If you encounter 429 errors, add a 'Delay' step (60 seconds) before the webhook.

Meeting Filter Logic

Type: workflow

A Zapier Filter step that ensures only client briefing calls trigger brief generation—preventing internal team meetings, 1:1s, and non-client calls from wasting API credits and cluttering the brief pipeline.

Implementation

• Step Type: Filter by Zapier
• Position: After the Fireflies trigger, before the GPT-5.4 webhook

Filter Rules (ALL conditions must pass — AND logic)

• Condition 1 — Meeting Title: Field: {{1. Meeting Title}} | Condition: (Text) Contains any of: briefing, kickoff, discovery, client call, intake, onboarding, campaign, strategy session, brief, creative review
• OR
• Condition 2 — Meeting Title (custom tag): Field: {{1. Meeting Title}} | Condition: (Text) Contains | Value: [BRIEF]

Exclusion Rules

Use Zapier Paths instead of Filter if needed. If Meeting Title contains any of the following keywords, STOP and do not generate a brief:

• internal
• standup
• 1:1
• one-on-one
• team sync
• sprint
• retro
• happy hour
• lunch

Additional Filters

• Duration: Only proceed if {{1. Duration}} > 10 (ignore very short calls/accidental recordings)
• Attendees: Only proceed if {{1. Attendees Count}} >= 2 (need at least 2 people for a briefing call)

Fallback

If no filter matches (ambiguous meeting), send a Slack DM to the account manager:

Meeting "{{1. Meeting Title}}" was transcribed but didn't match brief generation criteria. If you'd like a brief generated, rename the meeting to include [BRIEF] and contact your admin.

Brief Quality Feedback Loop Workflow

Type: workflow A feedback mechanism that allows agency reviewers to rate AI-generated briefs and automatically flags low-quality outputs for prompt improvement. This creates a continuous improvement cycle that makes the briefs more accurate over time.

Implementation

IMPLEMENTATION — Google Form + Google Sheet + Zapier

Step 1: Create Google Form — 'Creative Brief Quality Review'

• Brief Title (Short text, pre-filled from Asana task link)
• Meeting Date (Date picker)
• Reviewer Name (Dropdown of team members)
• Overall Accuracy (Scale 1-5: 1=Mostly Wrong, 5=Perfectly Accurate)
• Completeness (Scale 1-5: 1=Major Gaps, 5=Comprehensive)
• Specific Issues (Checkboxes: Incorrect client name, Wrong dates/timeline, Missing deliverables, Fabricated details not from call, Wrong audience description, Missed key discussion points, Formatting issues, Other)
• Comments/Corrections (Long text — free-form feedback)
• Prompt Change Suggested (Long text — what should the AI do differently?)

Step 2: Link Form to Google Sheet — 'Brief Quality Tracker'

Sheet auto-populates with form responses. Add the following calculated columns:

• Average Score
• Flagged

Average Score = (Accuracy + Completeness) / 2
Flagged = IF(Average Score < 3, "YES", "NO")

Step 3: Zapier Automation — 'Brief Quality Alert'

Low quality brief detected: {{Brief Title}} scored {{Average Score}}/5. Reviewer: {{Reviewer Name}}. Issues: {{Specific Issues}}. Please review and adjust the GPT-5.4 prompt.

Step 4: Add Review Link to Asana Task

In the Zapier workflow (Step 7, Action 5 — Asana task creation), include a pre-filled Google Form link in the task description:

After reviewing this brief, please submit quality feedback: https://forms.google.com/YOUR_FORM_ID?entry.FIELD_ID={{meeting_title}}

Step 5: Monthly Review Process

MSP reviews the Google Sheet monthly during the health check (Step 11). If average quality score drops below 3.5/5, follow this remediation process:

Client-Specific Context Injection Prompt

Type: prompt An extension prompt module that injects client-specific context (brand guidelines, past brief patterns, key stakeholders) into the GPT-5.4 request to improve brief accuracy for repeat clients. This is implemented as a dynamic user message prefix in the Zapier workflow, populated from a Google Sheet lookup.

Implementation

Please generate a creative brief from the following client briefing call transcript.

CLIENT CONTEXT (use this to enrich the brief with background knowledge):
- Client: {{lookup. Client Name}}
- Industry: {{lookup. Industry}}
- Brand Voice: {{lookup. Brand Voice Keywords}}
- Primary Contact: {{lookup. Primary Contact Name & Role}}
- Key Products/Services: {{lookup. Key Products/Services}}
- Known Target Audience: {{lookup. Target Audience Summary}}
- Brand Guidelines: {{lookup. Brand Guidelines URL}}
- Historical Notes: {{lookup. Previous Brief Patterns/Notes}}
- Special Instructions: {{lookup. Special Instructions}}

MEETING TITLE: {{1. Meeting Title}}
MEETING DATE: {{1. Date}}
ATTENDEES: {{1. Attendees}}
DURATION: {{1. Duration}} minutes

--- TRANSCRIPT START ---
{{1. Transcript Text}}
--- TRANSCRIPT END ---

Generate the creative brief now. Use the client context to fill in background information and ensure consistency with known brand guidelines. However, always prioritize what was explicitly discussed in the transcript over historical context.

Testing & Validation

• TEST 1 — Audio Device Quality: Record a 2-minute test call using each deployed audio device (Jabra Speak2 75, Poly Sync 20+). Play back through Fireflies and verify the transcript has fewer than 5% word errors. Have 3 people speak simultaneously for 10 seconds and verify speaker diarization correctly separates them.
• TEST 2 — Fireflies Auto-Join: Create a test calendar event with a Zoom/Google Meet link titled 'Test Client Briefing Call' scheduled for 5 minutes in the future. Verify the Fireflies bot joins within 2 minutes of the meeting start time. Confirm the consent notification appears in the meeting chat.
• TEST 3 — Fireflies Transcription Accuracy: Conduct a simulated 15-minute client briefing call using a prepared script with specific names, dates, dollar amounts, and industry terminology. Compare the Fireflies transcript against the script. Target: >95% word accuracy for clearly spoken content, correct speaker labels for all participants.
• TEST 4 — Zapier Filter Logic: Create three test transcripts in Fireflies: (a) titled 'Client Discovery Call - Acme Corp' (should trigger brief generation), (b) titled 'Weekly Team Standup' (should be filtered out), (c) titled 'Lunch Planning' (should be filtered out). Verify only transcript (a) proceeds through the Zapier workflow.
• TEST 5 — GPT-5.4 Brief Generation Quality: Send a real client call transcript (with client permission) through the Zapier webhook to GPT-5.4. Review the output brief against the original transcript with the agency's AI Champion. Verify all 12 sections of the brief template are present, key client requirements are captured accurately, no hallucinated details appear, and '[Not discussed]' markers appear for sections not covered in the call.
• TEST 6 — Google Drive Document Creation: After a successful brief generation, verify a new Google Doc appears in the correct folder within the 'Client Call Transcripts' directory. Open the document and verify formatting is clean (headings, bullet points, bold text render correctly). Verify the document title follows the naming convention: 'Creative Brief - [Meeting Title] - [Date]'.
• TEST 7 — Asana Task Creation: After brief generation, verify a new task appears in the 'Client Briefs for Review' Asana project in the 'AI Generated' column. Verify the task includes: correct title, link to Google Doc, assignee (account manager), and due date (2 business days out). Move the task to 'Approved' and verify the Asana Rule triggers notification to the Creative Director.
• TEST 8 — Slack Notification: After brief generation, verify a formatted message appears in the #client-briefs Slack channel containing: meeting title, date, Google Doc link, and Asana task link. Click both links to verify they resolve to the correct documents.
• TEST 9 — End-to-End Latency: Time the full workflow from call end to Slack notification. Target: brief should be generated, filed, tasked, and notified within 10 minutes of call completion. If latency exceeds 15 minutes, investigate Fireflies processing time (typically 5 min) and Zapier execution time.
• TEST 10 — Error Handling: Temporarily invalidate the OpenAI API key in Zapier and trigger a test transcript. Verify the error path sends a Slack DM to the MSP admin with the failure details. Restore the valid API key and re-run to confirm recovery.
• TEST 11 — Consent Compliance: Join a test call as an external participant (simulating a client). Verify the Fireflies consent message appears in the meeting chat within 30 seconds of the bot joining. Verify the account manager delivers the verbal consent script. Record the test for documentation.
• TEST 12 — Volume Test: Generate 5 briefs in rapid succession (5 test calls within 30 minutes). Verify all 5 are processed correctly without Zapier task queue errors, API rate limiting, or duplicate documents. Check Zapier task usage to confirm consumption matches expectations (approximately 25-30 tasks for 5 briefs).

Client Handoff

Conduct a 90-minute client handoff session with the agency's leadership team (AI Champion, Creative Director, Account Directors, and Operations Manager). Cover the following topics:

Maintenance

ONGOING MSP RESPONSIBILITIES — Budget 2-4 hours/month per agency client:

Weekly (15 min)

• Check Zapier task history for failed runs — investigate and resolve any errors within 24 hours
• Review Slack #client-briefs channel for any user-reported issues
• Spot-check one generated brief against its transcript for quality assurance

Monthly (2-3 hours)

• Review Brief Quality Tracker Google Sheet — calculate average scores and identify trends
• Review OpenAI API usage and costs at https://platform.openai.com/usage — verify within budget
• Review Fireflies admin dashboard — check active users, transcription minutes consumed, and storage usage
• Review Zapier task consumption — ensure within plan limits (2,000 tasks/month on Professional)
• If average quality score is below 4.0/5.0: Pull lowest-rated briefs, analyze failure patterns, and revise the system prompt. Test revised prompt against 3 historical transcripts before deploying.
• Update Client Context Database Google Sheet with any new clients or changed brand guidelines
• Apply any Fireflies platform updates or new feature configurations

Quarterly (3-4 hours)

• Full prompt optimization review: Analyze all brief quality data from the quarter, identify systematic gaps, and perform a major prompt revision if needed
• Review and renew OAuth tokens for all integrations (Google Calendar, Slack, Asana, HubSpot) — re-authorize any expired connections
• Update Fireflies, Jabra Direct, and Poly Lens software/firmware to latest versions
• Review compliance: Verify consent procedures are being followed, data retention policies are enforced, and any new privacy regulations are addressed
• Conduct a brief user satisfaction survey (5 questions, Google Form) with agency team
• Review hardware condition — replace any damaged speakerphones or microphones
• Present quarterly report to agency leadership: usage stats, quality metrics, cost summary, recommendations

Annually

• Renegotiate Fireflies licensing (volume discounts may apply at renewal)
• Review the competitive landscape for transcription and LLM platforms — assess if switching vendors would improve quality or reduce cost
• Review and update the client MSA recording consent clause with agency legal
• Assess whether to upgrade from Zapier to Make (if task volumes have grown) or to a custom integration (if the agency has scaled significantly)

SLA Considerations

• Response time for system-down issues (Zapier failures, API outages): 4 business hours
• Response time for quality issues (bad brief output): 1 business day
• Scheduled maintenance windows: Off-hours only (evenings/weekends)
• Uptime target: 99.5% (dependent on third-party SaaS uptime — Fireflies, OpenAI, Zapier)

Escalation Path

Alternatives

Replace Fireflies.ai with Otter.ai Business ($20/user/month) as the transcription platform. Otter provides similar auto-join, transcription, and summary features with OtterPilot. The Zapier integration and GPT-5.4 brief generation workflow remain identical — only the trigger source changes.

Zoom AI Companion — Zero Additional Cost (Platform-Native)

Use Zoom's built-in AI Companion features (included free with Zoom Pro/Business) instead of a third-party transcription platform. Zoom AI Companion provides meeting transcription, AI-generated summaries, and action items natively within the Zoom interface. The brief generation would still use GPT-5.4 via Zapier, but the transcript source changes from Fireflies to Zoom's API.

Custom API Pipeline (Deepgram + GPT-5.4) — Maximum Control & Margin

Build a fully custom transcription pipeline using Deepgram API ($0.0077/min) for speech-to-text and GPT-5.4 for brief generation, bypassing SaaS transcription platforms entirely. Audio is captured from Zoom/Meet/Teams via their recording APIs and sent to Deepgram for transcription. A custom Python application or n8n workflow orchestrates the pipeline.

Microsoft Teams Premium + Copilot — Enterprise Microsoft Ecosystem

For agencies fully embedded in the Microsoft 365 ecosystem, use Teams Premium ($10/user/month) for Intelligent Recap (AI transcription, notes, action items, and meeting chapters) combined with Microsoft Copilot for M365 ($30/user/month) for document generation. Teams Premium transcribes the call and Copilot generates the creative brief as a Word document in SharePoint.

Self-Hosted Whisper + Local LLM — Maximum Privacy

Deploy OpenAI's Whisper large-v2 model on a local GPU server or private cloud VM for transcription, and use a self-hosted LLM (e.g., Llama 3.1 70B via Ollama or vLLM) for brief generation. No audio or transcript data leaves the agency's infrastructure. Audio is captured from conference room devices and uploaded to the local processing server.