Implementation Guide: Transcribe client intake meetings and draft matter summaries for attorney review

Hardware Procurement

Jabra Speak2 75 Wireless Speakerphone

USB-A/USB-C Hub (for workstations lacking ports)

Software Procurement

Prerequisites

• Active Microsoft 365 Business Premium (or equivalent) with Azure AD / Entra ID configured for all attorney and staff accounts
• Clio Manage subscription (Advanced or Complete tier) with admin access and API keys available
• Minimum 25 Mbps symmetrical internet at all office locations where intake meetings occur
• Wi-Fi network with WPA2-Enterprise or WPA3 encryption; guest network isolation if clients connect devices during meetings
• Each attorney workstation running Windows 10/11 or macOS 12+ with Chrome or Edge browser
• iOS 15+ or Android 12+ smartphones for attorneys using PLAUD Note Pro mobile app for device management
• Firm-approved engagement letter template updated to include AI transcription disclosure language per ABA Opinion 512
• State-specific recording consent form drafted and approved by firm's managing partner or ethics counsel
• Clio admin user credentials with permissions to create custom fields, matter templates, and API tokens
• DNS and firewall rules allowing outbound HTTPS (443) to: otter.ai, api.otter.ai, plaud.ai, api.clio.com, hooks.zapier.com, api.openai.com
• Firm's data retention policy documented (required for configuring auto-deletion in Otter.ai and Clio)
• Designated pilot group of 1–2 tech-forward attorneys willing to test the system for 2 weeks before firm-wide rollout

Installation Steps

Step 1: Compliance & Consent Framework Setup

Before any technology is deployed, establish the legal and ethical framework for AI-assisted intake recording. This step is non-negotiable for legal services clients and must be completed first. Work with the firm's managing partner or ethics counsel to: (1) update the client engagement letter to disclose AI transcription use per ABA Opinion 512, (2) create a standalone recording consent form that meets the most restrictive state law applicable to the firm's practice, (3) draft an internal AI use policy for firm staff, and (4) establish data retention periods aligned with state bar requirements.

Step 2: Provision Otter.ai Enterprise Accounts

Contact Otter.ai sales to set up an Enterprise account for the firm. Enterprise tier is mandatory for legal deployments — do NOT use Business or Pro tiers as they lack BAA, SSO, and data training opt-out guarantees. During the sales process: (1) execute a Data Processing Agreement (DPA) confirming no model training on client data, (2) execute a BAA if the firm handles any health-related legal matters, (3) confirm US-only data residency, and (4) obtain written confirmation of SOC 2 Type II certification.

# replace {tenant-id} with your organization's tenant ID

https://login.microsoftonline.com/{tenant-id}/federationmetadata/2007-06/federationmetadata.xml

Step 3: Configure Otter.ai Data Retention & Security Policies

Once the Enterprise account is provisioned, configure security and data governance settings before any recordings are made. This ensures compliance from the first transcript.

Step 4: Configure Custom Legal Vocabulary in Otter.ai

Improve transcription accuracy for legal terminology by configuring custom vocabulary. Legal jargon, Latin phrases, jurisdiction-specific terms, and proper nouns (judge names, court names, opposing counsel names) significantly impact transcript quality.

• General Legal Terms: pro bono, pro se, amicus curiae, voir dire, habeas corpus, mens rea, actus reus, res judicata, stare decisis, prima facie, subpoena duces tecum, in limine, ex parte, nolo contendere, nolle prosequi, certiorari, mandamus, interlocutory, demurrer, interrogatories, deposition, affidavit, stipulation, continuance
• Family Law (practice-area-specific): dissolution, custody, parenting time, child support, alimony, spousal maintenance, marital property, equitable distribution, QDRO
• Personal Injury (practice-area-specific): negligence, comparative fault, contributory negligence, subrogation, med-pay, PIP, loss of consortium, demand letter
• Estate Planning (practice-area-specific): testamentary, intestate, probate, executor, trustee, revocable trust, irrevocable trust, pour-over will, power of attorney, healthcare proxy, advance directive
• Jurisdiction-Specific (customize per firm's courts): local court names, judge names, opposing counsel firm names, county names, and statute numbers commonly referenced

Step 5: Deploy PLAUD Note Pro Hardware

Unbox, configure, and deploy PLAUD Note Pro recorders for each attorney and shared conference room. Each device must be paired with the attorney's smartphone and linked to their PLAUD AI subscription.

Per-Device Setup

For Conference Room Units

Step 6: Deploy Jabra Speakerphones

Install Jabra Speak2 75 units in conference rooms and Jabra Speak2 55 units in attorney offices. These serve dual purpose: improved audio capture for transcription AND better audio quality for remote intake meetings via Teams/Zoom.

Jabra Speak2 75 (Conference Rooms)

Jabra Speak2 55 (Attorney Offices)

Step 7: Configure Zapier Integration Hub

Set up the Zapier Team account and build the core automation workflows (Zaps) that connect Otter.ai transcriptions to Clio Manage. This is the integration backbone of the entire solution.

Step 8: Build Primary Zapier Workflow: Otter Transcript → Clio Matter Note

Create the main Zap that automatically takes completed Otter.ai transcripts, formats them into structured matter summaries, and creates notes in the corresponding Clio matter. This is the core automation of the entire solution.

// Step 3 (Optional): OpenAI GPT-5.4 — Generate structured legal summary
Model: gpt-5.4
System prompt: (See custom_ai_components for full prompt)
User message: {{formatted_transcript_from_step_2}}
Max tokens: 2000
Temperature: 0.3 (low creativity for factual accuracy)

// Step 4: Clio → Create Note
Matter: Search by client name or matter number (from transcript metadata)
Subject: 'AI-Generated Intake Summary — {{date}} — PENDING ATTORNEY REVIEW'
Detail: {{summary_output_from_step_3}}
Category: 'Intake Notes'

// Step 5: Microsoft Teams → Send Channel Message
Team: 'Legal Operations'
Channel: 'Intake Summaries'
Message: '📋 New intake summary ready for review
Client: {{client_name}}
Attorney: {{attorney_name}}
Matter: {{matter_link}}
⚠️ Status: PENDING REVIEW — Do not rely on until attorney-approved'

Step 9: Configure Clio Manage for AI Intake Workflow

Set up Clio Manage to properly receive, display, and track AI-generated intake summaries. Create custom fields, matter templates, and task workflows that integrate with the Zapier automation.

1. Create Custom Note Category

• Navigate to Settings → Note Categories → Add
• Name: 'AI Intake Summary — Draft' | Color: Yellow (visual indicator of unreviewed status)
• Name: 'AI Intake Summary — Approved' | Color: Green

2. Create Matter Template for Intake

• Navigate to Settings → Matter Templates → Add
• Name: 'New Client Intake'
• Default practice area: (firm's primary area)
• Auto-create tasks: a. 'Review AI intake summary' (assigned to attorney, due: 1 business day)
• Auto-create tasks: b. 'Conflicts check' (assigned to paralegal, due: 1 business day)
• Auto-create tasks: c. 'Send engagement letter' (assigned to paralegal, due: 2 business days)
• Auto-create tasks: d. 'File consent form' (assigned to admin, due: same day)

3. Create Custom Fields on Matter

• 'AI Transcription ID' (text) — links back to Otter.ai transcript
• 'Intake Recording Consent' (checkbox) — confirms consent was obtained
• 'AI Summary Status' (picklist: Draft / Under Review / Approved / Rejected)

4. Enable Clio Manage AI

• Navigate to Settings → AI Features → Enable Manage AI
• This allows attorneys to use Clio's built-in AI to further refine summaries

5. Generate API Token

• Navigate to Settings → API → Generate new API key
• Label: 'Zapier Integration — Intake Workflow'
• Permissions: Matters (read/write), Notes (read/write), Tasks (read/write)
• Copy token and store securely in MSP password manager

Step 10: Build Notification and Escalation Workflows

Create secondary Zapier workflows for notifications, reminders, and escalation to ensure no AI-generated summary goes unreviewed. This is critical for compliance — unreviewed AI summaries sitting in the system create liability risk.

Zap 2: Reminder for Unreviewed Summaries

Zap 3: Weekly Compliance Report

Step 11: Configure Recording Consent Workflow

Implement a standardized process for obtaining and documenting client consent before any recording begins. This must be the FIRST step of every intake meeting and is non-negotiable.

Physical Consent Workflow (posted in every intake room)

Digital Consent Option (for remote/Teams intake)

Step 12: Pilot Deployment with Selected Attorneys

Deploy the complete solution to 1–2 tech-forward attorneys for a 2-week pilot period. Monitor closely for transcription accuracy, workflow reliability, summary quality, and user experience issues before firm-wide rollout.

Step 13: Firm-Wide Rollout and Training

After successful pilot, deploy to all attorneys and train all staff (attorneys, paralegals, receptionists, office manager) on their respective roles in the AI intake workflow.

Custom AI Components

Legal Intake Summary Generator Prompt

Type: prompt System prompt for GPT-5.4 that transforms raw meeting transcripts into structured legal matter summaries. This prompt is used in the Zapier workflow's OpenAI step to generate attorney-ready summaries from Otter.ai transcripts. It enforces a consistent format that aligns with standard legal intake documentation requirements.

Implementation:

You are a legal intake summarization assistant working for a law firm. Your task is to transform a raw meeting transcript into a structured matter summary for attorney review.

IMPORTANT RULES:
1. You are creating a DRAFT summary that MUST be reviewed by an attorney before any reliance.
2. Do NOT provide legal advice, opinions, or case assessments.
3. Do NOT infer information not explicitly stated in the transcript.
4. If information for a section is not available in the transcript, write '[NOT DISCUSSED]'.
5. Flag any potential conflicts of interest mentioned.
6. Use formal legal writing style — concise, factual, third person.
7. Preserve exact quotes for critical statements using quotation marks.
8. Include timestamps from the transcript for key statements when available.

OUTPUT FORMAT:

# CLIENT INTAKE SUMMARY — DRAFT
## ⚠️ AI-GENERATED — PENDING ATTORNEY REVIEW

### 1. CLIENT INFORMATION
- Full Name:
- Contact Phone:
- Contact Email:
- Address:
- Date of Birth:
- Preferred Contact Method:
- Referred By:

### 2. MATTER OVERVIEW
- Practice Area:
- Type of Matter:
- Jurisdiction:
- Statute of Limitations Concern: [Yes/No/Unknown — flag if mentioned]
- Brief Description (2-3 sentences):

### 3. KEY FACTS
[Chronological bullet-point list of facts as stated by the client. Each bullet should include the approximate timestamp if available. Do not editorialize.]

### 4. PARTIES INVOLVED
[List all persons, entities, organizations mentioned with their role/relationship]
- Opposing Party:
- Witnesses Mentioned:
- Other Attorneys/Firms Mentioned:
- Insurance Companies:
- Government Agencies:

### 5. DOCUMENTS & EVIDENCE MENTIONED
[List any documents, photos, contracts, communications the client referenced]

### 6. CLIENT'S STATED OBJECTIVES
[What does the client want to achieve? Use their words where possible.]

### 7. FINANCIAL INFORMATION
- Damages/Amount in Controversy:
- Fee Discussion: [Contingency/Hourly/Flat fee discussed?]
- Retainer Discussion:
- Insurance Coverage:

### 8. CONFLICTS CHECK ITEMS
[List all names and entities that need to be run through the firm's conflicts database]

### 9. URGENT/TIME-SENSITIVE ITEMS
[Any deadlines, court dates, statute of limitations, or time-sensitive matters mentioned]

### 10. ATTORNEY NEXT STEPS
[Suggested next steps based on what was discussed — NOT legal advice, just administrative actions]
- [ ] Review and approve this summary
- [ ] Run conflicts check on all parties listed in Section 4
- [ ] [Additional items based on transcript content]

### 11. NOTABLE QUOTES
[2-5 direct quotes from the client that are particularly relevant to the matter]

### 12. TRANSCRIPT METADATA
- Date of Intake: {{meeting_date}}
- Duration: {{meeting_duration}}
- Attending Attorney: {{attorney_name}}
- Recording Consent: Obtained (per firm protocol)
- Transcription Platform: Otter.ai (AI-generated)
- Summary Generated: {{current_datetime}} by GPT-5.4

---
⚠️ DISCLAIMER: This summary was generated by artificial intelligence and has not been reviewed by an attorney. It may contain errors, omissions, or misinterpretations. An attorney must review and approve this document before any reliance. Do not share with clients or opposing counsel.
---

Zapier Transcript Formatter

Type: integration JavaScript code step in Zapier that preprocesses the raw Otter.ai transcript before sending it to GPT-5.4 for summarization. Cleans up the transcript, extracts metadata, enforces naming conventions, and handles edge cases.

Implementation

// Zapier Code Step (JavaScript)
// Input variables from Otter.ai trigger:
//   inputData.transcript_text — Full transcript text from Otter.ai
//   inputData.transcript_title — Title of the transcript
//   inputData.meeting_date — Date of the meeting
//   inputData.duration — Duration in minutes
//   inputData.speakers — Speaker names (comma-separated)

const transcript = inputData.transcript_text || '';
const title = inputData.transcript_title || 'Untitled Intake';
const meetingDate = inputData.meeting_date || new Date().toISOString().split('T')[0];
const duration = inputData.duration || 'Unknown';
const speakers = inputData.speakers || 'Unknown speakers';

// Check if this is actually an intake meeting (safety check)
const isIntake = /intake|initial consultation|new client|first meeting/i.test(title);

if (!isIntake) {
  // Return empty to stop the Zap if not an intake meeting
  return { should_continue: false, reason: 'Transcript title does not indicate an intake meeting' };
}

// Clean transcript: remove filler words, normalize spacing
let cleaned = transcript
  .replace(/\b(um|uh|you know|like|basically|actually|so yeah)\b/gi, '')
  .replace(/\s{2,}/g, ' ')
  .replace(/\n{3,}/g, '\n\n')
  .trim();

// Truncate if too long (GPT-5.4 context window management)
const MAX_CHARS = 80000; // ~20K tokens, safe for GPT-5.4
if (cleaned.length > MAX_CHARS) {
  cleaned = cleaned.substring(0, MAX_CHARS) + '\n\n[TRANSCRIPT TRUNCATED — Full version available in Otter.ai]';
}

// Extract attorney name from speakers list (assumes attorney speaks first)
const speakerList = speakers.split(',').map(s => s.trim());
const attorneyName = speakerList[0] || 'Unknown Attorney';

// Build the formatted output for GPT-5.4
const formattedForGPT = `MEETING METADATA:
- Title: ${title}
- Date: ${meetingDate}
- Duration: ${duration} minutes
- Speakers: ${speakers}
- Attorney: ${attorneyName}

TRANSCRIPT:
${cleaned}`;

return {
  should_continue: true,
  formatted_transcript: formattedForGPT,
  attorney_name: attorneyName,
  meeting_date: meetingDate,
  duration: duration,
  client_name: speakerList.length > 1 ? speakerList[1] : 'Unknown Client',
  transcript_length: cleaned.length,
  was_truncated: cleaned.length >= MAX_CHARS
};

Clio Matter Note Creator

Type: integration Zapier workflow step configuration that creates a properly formatted and categorized note in Clio Manage from the AI-generated summary. Includes matter lookup logic, error handling, and proper metadata tagging.

Implementation

Action: Clio → Find Matter
Search Configuration:
  - Search field: Client name
  - Search value: {{client_name_from_formatter}}
  - If not found: Create new matter using template 'New Client Intake'
    - Client name: {{client_name}}
    - Responsible attorney: {{attorney_name}}
    - Practice area: (default from template)
    - Status: 'Pending'

Action: Clio → Create Note
Configuration:
  - Matter ID: {{matter_id_from_step_a}}
  - Subject: 'AI Intake Summary — {{meeting_date}} — ⚠️ DRAFT'
  - Detail (rich text): {{gpt4o_summary_output}}
  - Type: 'Intake Note'
  - Date: {{meeting_date}}
  - Regarding: {{client_name}}

Action: Clio → Create Task
Configuration:
  - Matter ID: {{matter_id_from_step_a}}
  - Name: 'Review AI Intake Summary — {{client_name}}'
  - Description: 'An AI-generated intake summary has been created for this matter. Please review for accuracy, approve or reject, and update the note category accordingly.\n\nTranscript source: Otter.ai\nGenerated: {{current_datetime}}'
  - Assignee: {{attorney_name}}
  - Due date: {{tomorrow_date}}
  - Priority: High
  - Status: Not started

# Zapier action configuration

Action: Clio → Update Matter
Configuration:
  - Matter ID: {{matter_id_from_step_a}}
  - Custom field 'AI Transcription ID': {{otter_transcript_id}}
  - Custom field 'Intake Recording Consent': true
  - Custom field 'AI Summary Status': 'Draft'

Error Handling

Attorney Review Status Tracker

Type: workflow A Zapier workflow that monitors the status of AI-generated summaries in Clio and provides analytics on review completion rates, average review time, and rejection rates. Feeds into the weekly compliance report.

Implementation

const notes = JSON.parse(inputData.clio_notes || '[]');

let stats = {
  total_created: notes.length,
  approved: 0,
  pending_review: 0,
  rejected: 0,
  avg_review_hours: 0,
  overdue: 0 // pending > 48 hours
};

let totalReviewHours = 0;
let reviewedCount = 0;
const now = new Date();

notes.forEach(note => {
  const category = (note.category || '').toLowerCase();
  const created = new Date(note.created_at);
  const hoursOld = (now - created) / (1000 * 60 * 60);
  
  if (category.includes('approved')) {
    stats.approved++;
    // Estimate review time as hours between creation and last update
    if (note.updated_at) {
      totalReviewHours += (new Date(note.updated_at) - created) / (1000 * 60 * 60);
      reviewedCount++;
    }
  } else if (category.includes('draft')) {
    stats.pending_review++;
    if (hoursOld > 48) stats.overdue++;
  } else if (category.includes('rejected')) {
    stats.rejected++;
  }
});

stats.avg_review_hours = reviewedCount > 0 ? (totalReviewHours / reviewedCount).toFixed(1) : 'N/A';
stats.approval_rate = stats.total_created > 0 ? ((stats.approved / stats.total_created) * 100).toFixed(0) + '%' : 'N/A';

const report = `
📊 WEEKLY AI INTAKE SUMMARY REPORT
Period: ${inputData.start_date} — ${inputData.end_date}

📝 Total Summaries Generated: ${stats.total_created}
✅ Approved: ${stats.approved}
⏳ Pending Review: ${stats.pending_review}
❌ Rejected: ${stats.rejected}
⚠️ Overdue (>48hrs): ${stats.overdue}

📈 Approval Rate: ${stats.approval_rate}
⏱️ Avg Review Time: ${stats.avg_review_hours} hours

${stats.overdue > 0 ? '🚨 ACTION REQUIRED: ' + stats.overdue + ' summaries are overdue for review. Escalate to managing partner.' : '✅ No overdue items.'}
`;

return { report: report, overdue_count: stats.overdue, total: stats.total_created };

PLAUD-to-Otter Bridge Workflow

Type: workflow

Handles the scenario where attorneys use PLAUD Note Pro hardware for in-person recordings that need to be processed through Otter.ai for consistent summary generation. PLAUD has its own transcription, but routing through Otter.ai ensures a single integration point to Clio.

Option A: Manual Upload (Simplest)

Option B: Automated via Cloud Storage (Recommended)

# POST to Otter.ai upload endpoint (Enterprise API)

POST https://otter.ai/api/v1/upload
Authorization: Bearer {{otter_api_token}}
Content-Type: multipart/form-data

<audio file>

If Otter.ai Enterprise API upload is not available or too complex:

Testing & Validation

• CONSENT WORKFLOW TEST: Conduct a mock intake meeting with a staff member playing the client. Verify that: (1) the consent form is presented before any recording begins, (2) the verbal consent statement is captured at the beginning of the recording, (3) the signed consent form is scanned and uploaded to the correct Clio matter, and (4) the 'Intake Recording Consent' custom field is checked in Clio.
• AUDIO QUALITY TEST: Record a 10-minute sample conversation in each intake room and attorney office using both the PLAUD Note Pro and Jabra speakerphone. Play back recordings and verify: (1) both speakers are clearly audible, (2) no significant background noise interference, (3) speaker separation is functioning (PLAUD identifies different speakers), and (4) audio quality is sufficient for accurate transcription (target: 90%+ word accuracy).
• TRANSCRIPTION ACCURACY TEST: Record a scripted 15-minute mock intake containing at least 20 legal terms from the custom vocabulary list. Upload to Otter.ai and compare the transcript against the known script. Calculate word error rate (WER). Accept if WER < 10% overall and < 15% for legal-specific terms. If accuracy is below threshold, review and expand custom vocabulary.
• SPEAKER DIARIZATION TEST: Conduct a 3-person mock intake (attorney, client, interpreter or second client). Verify that Otter.ai correctly identifies and labels all three speakers for at least 85% of utterances. Retrain speaker profiles if accuracy is below threshold.
• END-TO-END PIPELINE TEST: Record a complete mock intake meeting, let it flow through the entire pipeline (PLAUD → Otter.ai → Zapier → GPT-5.4 → Clio), and verify: (1) transcript appears in Otter.ai within 15 minutes, (2) AI summary is generated and correctly formatted, (3) Clio note is created in the correct matter with 'Draft' category, (4) Clio task is created and assigned to the correct attorney, (5) Teams notification is delivered, and (6) all 12 sections of the summary template are populated or marked '[NOT DISCUSSED]'.
• AI SUMMARY QUALITY TEST: Have the pilot attorney review 5 AI-generated summaries against the original transcripts. Score each summary on: (1) factual accuracy — no hallucinated information (critical), (2) completeness — all discussed topics captured, (3) organization — easy to scan and review, (4) actionable — next steps are clear. All summaries must score 4/5 or higher on each criterion. Any factual hallucination is an automatic fail requiring prompt template revision.
• ZAPIER FAILURE HANDLING TEST: Deliberately trigger error conditions and verify graceful handling: (1) disconnect Clio API token — verify error notification is sent to MSP, (2) submit a transcript with no identifiable client name — verify the workflow creates an 'Unmatched Intake' note for manual routing, (3) submit a very short transcript (< 2 minutes) — verify the formatter flags it as potentially incomplete.
• SSO AND ACCESS CONTROL TEST: Verify that: (1) attorneys can log into Otter.ai only via Azure AD SSO (direct password login is disabled), (2) Conditional Access policy requires MFA for Otter.ai access, (3) a terminated user cannot access Otter.ai after their Azure AD account is disabled, and (4) Otter.ai audit logs show login events for compliance.
• DATA RETENTION TEST: Create a test transcript, then manually trigger the retention policy deletion (or set a short test retention period of 1 day). Verify that: (1) the owner is notified before deletion, (2) the transcript and audio are permanently deleted from Otter.ai, (3) the corresponding Clio note is NOT deleted (Clio has its own retention policy), and (4) no copies remain in Zapier task history.
• REMOTE INTAKE TEST: Conduct a mock intake via Microsoft Teams with the Jabra Speak2 75 as the audio device. Verify that: (1) Otter.ai captures the Teams meeting audio correctly (either via bot or system audio capture), (2) remote participant audio quality is sufficient for transcription, (3) speaker diarization works for remote participants, and (4) the full pipeline completes identically to in-person intakes.
• REJECTION WORKFLOW TEST: Have an attorney review an AI summary and mark it as 'Rejected' in Clio (change note category). Verify that: (1) the status change is reflected in the weekly report, (2) no automated follow-up messages are sent for rejected summaries, and (3) the original transcript link in Otter.ai is preserved so the attorney can manually create notes from the raw transcript.
• VOLUME/LOAD TEST: On a single day, process 8–10 mock transcripts through the pipeline in rapid succession (simulating a busy intake day). Verify that: (1) Zapier does not throttle or drop any tasks, (2) all summaries are generated and delivered within 30 minutes, (3) Clio API rate limits are not exceeded, and (4) task count remains within the Zapier Team plan allocation.

Client Handoff

The client handoff meeting should be a 2-hour session attended by the managing partner, all attorneys, the office manager, and the designated IT liaison. Cover the following:

1. Live System Demonstration (30 minutes)

• Walk through a complete intake cycle: consent → recording → transcription → AI summary → Clio note → attorney review → approval
• Show the Teams notification flow and how summaries appear in Clio
• Demonstrate how to review, edit, approve, or reject an AI summary in Clio

2. Compliance Review (20 minutes)

• Review the updated engagement letter language with AI disclosure
• Walk through the consent form and verbal consent script
• Explain ABA Opinion 512 obligations: competence, confidentiality, informed consent, reasonable fees
• Clarify that attorneys CANNOT bill clients for time spent learning the tool, but CAN bill for review time
• Review the firm's AI use policy document

3. Troubleshooting & Escalation (15 minutes)

• What to do if PLAUD doesn't sync (restart Bluetooth, check Wi-Fi, manual upload)
• What to do if Otter.ai is down (record anyway, upload later; take backup handwritten notes)
• What to do if a client declines recording (proceed with traditional intake, no pressure)
• MSP support contact and SLA: response within 2 business hours for non-critical, 1 hour for pipeline failures

4. Documentation Handoff

Leave the following documents (digital and printed):

• Quick Reference Card for Attorneys (1 page, laminated, for each intake room)
• Quick Reference Card for Paralegals/Admin (1 page)
• Quick Reference Card for Reception (1 page)
• Full System Documentation binder/PDF including: architecture diagram, all account credentials (in password manager), Zapier workflow documentation, Clio configuration baseline, vendor contact information, consent form templates, AI use policy
• Recording of the attorney training session (uploaded to firm's SharePoint)

5. Success Criteria Review (15 minutes)

• Confirm pilot results: X summaries generated, Y% accuracy, Z minutes saved per intake
• Agree on 30-day and 90-day success metrics: target 80%+ adoption rate, <24hr average review time, <5% summary rejection rate
• Schedule 30-day check-in call
• Schedule first quarterly review

6. Billing & Ongoing Service Agreement (10 minutes)

• Review the managed service agreement and monthly fee
• Confirm what is included vs. billable (e.g., prompt template changes are included; new Zapier integrations are project-billed)
• Review escalation path: firm admin → MSP helpdesk → MSP project team

Maintenance

Monthly Maintenance (2-4 hours MSP time)

• Review Zapier task usage and error logs; resolve any failed Zaps
• Check Otter.ai admin console for: storage usage, user activity, security alerts
• Review PLAUD device sync status across all units; replace any malfunctioning devices
• Pull and review the weekly compliance reports; flag any persistent overdue summaries to firm management
• Verify SSO and Conditional Access policies are functioning (test login)
• Check for software updates: Otter.ai feature releases, PLAUD firmware, Jabra firmware, Zapier connector updates

Quarterly Maintenance (4-6 hours MSP time)

• Meet with firm's managing partner or office manager to review system performance
• Analyze summary quality trends: approval rate, rejection rate, common rejection reasons
• Update custom vocabulary in Otter.ai based on new practice areas, judges, or terminology
• Refine GPT-5.4 summary prompt template based on attorney feedback and rejection patterns
• Review and update consent forms if state recording laws have changed
• Audit data retention compliance: verify no transcripts are retained beyond policy period
• Review vendor security posture: check for any SOC 2 report updates, security incidents, or policy changes
• Update system documentation with any configuration changes

Semi-Annual Maintenance (6-8 hours MSP time)

• Comprehensive compliance audit: review all AI-related policies, consent forms, engagement letters
• Review ABA ethics opinion updates and state bar guidance on AI in legal practice
• Assess whether current transcription platform is still best-in-class (market moves fast)
• Hardware assessment: battery health of PLAUD devices, condition of Jabra units
• Zapier workflow optimization: consolidate Zaps, improve error handling, add new automations requested by firm
• Present ROI report to firm leadership: time saved, matters processed, cost per intake summary

Trigger-Based Maintenance

• Otter.ai or OpenAI API pricing changes → Reassess cost model, notify client, adjust if needed
• New attorney hired → Provision accounts, deploy hardware, schedule 1-hour onboarding training ($150-250 billable)
• Attorney departure → Deprovision accounts, reassign devices, transfer any pending summaries, update Clio assignments
• Vendor security incident → Immediate assessment, notify firm, implement mitigations per incident response plan
• State recording law change → Update consent forms, retrain staff, document changes
• Clio major version update → Test Zapier integrations, verify API compatibility, update workflows if needed
• Firm opens new practice area → Update custom vocabulary, review and potentially revise summary prompt template

SLA Considerations

• Pipeline failure (no summaries being generated): 1-hour response, 4-hour resolution target
• Degraded transcription quality: 4-hour response, 24-hour resolution target
• Hardware failure (single device): next business day replacement from MSP spare inventory
• Account/access issues: 2-hour response
• Feature requests/prompt changes: scheduled for next monthly maintenance window
• Emergency compliance issue (potential privilege breach): 1-hour response, immediate escalation to firm's ethics counsel

Alternatives

Microsoft 365 Copilot + Teams Premium Native Solution

Instead of Otter.ai and PLAUD hardware, leverage Microsoft's native ecosystem. Teams Premium ($10/user/month) provides Intelligent Recap with AI-generated meeting summaries, and Microsoft 365 Copilot ($30/user/month) extends this with full generative AI capabilities across Office apps. All intake meetings are conducted via Teams (even in-person meetings use a Teams room). Copilot generates summaries natively within Teams, which can be pushed to Clio via Zapier or Power Automate.

Fireflies.ai + Custom API Pipeline

Replace Otter.ai with Fireflies.ai Business ($19/user/month annual) as the transcription platform. Fireflies offers superior CRM integrations, conversation intelligence analytics, and a more generous storage policy. For summary generation, use Fireflies' built-in AI summaries combined with a custom Python script that calls the Fireflies API and OpenAI API to generate legal-formatted summaries, bypassing Zapier entirely.

Rev.com AI + Human Hybrid Transcription

Use Rev.com's hybrid model: AI transcription at $0.25/minute for initial transcript, with optional human review at $1.99/minute for critical matters. This provides a human quality assurance layer that may be required for sensitive matters (e.g., criminal defense, high-value litigation). Rev's human transcribers are trained on legal terminology and can handle complex multi-party conversations.

Clio Manage AI + Clio Duo Native Solution (No Third-Party Transcription)

Wait for Clio's expanding AI capabilities and use Clio as the end-to-end platform. Clio Manage AI already includes document summarization and AI-assisted drafting. As Clio continues to build out ambient/meeting capture features (currently in development based on their product roadmap), the firm could potentially handle everything within Clio — eliminating the need for Otter.ai, Zapier, and the custom integration pipeline entirely.

Open-Source Self-Hosted Solution (Whisper + Llama)

For firms with extreme data sensitivity requirements, deploy a fully self-hosted solution: OpenAI Whisper (open-source) for transcription running on a local GPU server or Azure VM, and Meta Llama 3 (70B) or Mistral for summary generation. No client data ever leaves the firm's infrastructure or private cloud.