Implementation Guide: Transcribe client onboarding calls and generate engagement letter drafts

Hardware Procurement

Software Procurement

Prerequisites

• Active Zoom (5.x+), Microsoft Teams, or Google Meet subscription with admin-level access to authorize Fireflies.ai bot integration
• Microsoft 365 or Google Workspace email accounts for all staff who conduct onboarding calls
• Reliable internet connectivity: minimum 5 Mbps upload/download per concurrent call; 10 Mbps recommended. Wired Ethernet preferred in conference rooms.
• Firewall/proxy configured to allow outbound HTTPS (port 443) to: *.fireflies.ai, api.openai.com, *.make.com, *.ignitionapp.com, and *.smartvault.com
• Existing practice management system (Karbon, TaxDome, Canopy, or equivalent) with API access or Zapier integration capability
• Written Information Security Plan (WISP) on file — will be updated during implementation to address AI transcription tools
• Firm principal/partner authorization to implement AI transcription, including willingness to update engagement letter templates with recording consent and IRS Section 7216 consent language
• Inventory of states where clients are located to determine one-party vs. all-party recording consent requirements (all-party consent states: CA, CT, DE, FL, IL, MD, MA, MT, NV, NH, PA, WA, DC)
• Admin credentials for the firm's practice management system, document management system, and accounting software (QuickBooks Online, Xero) for integration configuration
• At least one partner or manager designated as the 'AI Champion' who will review and approve engagement letter drafts during the pilot phase

Installation Steps

Step 1: Deploy Speakerphone Hardware

Physically install and configure USB speakerphones in all rooms where client onboarding calls will be conducted. The Jabra Speak2 75 goes in the primary conference room; Jabra Speak2 55 units go in individual partner/manager offices. Connect via USB to the room's primary computer or docking station. Pair via Bluetooth to the partner's laptop as a secondary connection option for mobile use.

Get-PnpDevice -FriendlyName '*Jabra*' | Format-Table Status, FriendlyName

system_profiler SPUSBDataType | grep -A 5 'Jabra'

Step 2: Create and Configure Fireflies.ai Business Accounts

Set up the Fireflies.ai organization account with Business-tier licenses for all partners and managers who conduct onboarding calls. Configure SSO if available, set organization-level privacy defaults, and connect to the firm's video conferencing platform.

1040, 1065, 1120-S, S-Corp, C-Corp, LLC, GAAP, GAAS, QBO, W-2, W-9,
990, 501(c)(3), basis, depreciation, amortization, accrual, cash-basis,
engagement letter, retainer, tax preparation, bookkeeping, advisory

Step 3: Configure Recording Consent Workflow

Implement a multi-layered consent system that satisfies state recording laws, AICPA confidentiality rules, and IRS Section 7216 requirements. This includes written consent in the engagement letter, a verbal disclosure script read at the start of each call, and automated notification from Fireflies.ai.

Document 1: Verbal Disclosure Script

To be read at the start of every onboarding call. Save as: /templates/verbal-recording-disclosure.docx

# read verbatim at the start of each onboarding call

Before we begin, I want to let you know that we use an AI-powered
transcription service to record and transcribe our meetings. This helps us
accurately capture your information and generate your engagement documents
more efficiently. The recording and transcript are stored securely and are
subject to the same confidentiality protections as all your financial
information. Are you comfortable proceeding with the recording? If not,
I am happy to take notes manually instead.

Document 2: Written AI Transcription Consent Addendum

Save as: /templates/ai-transcription-consent-addendum.docx. Must include IRS Section 7216 consent language per Rev. Proc. 2013-14.

Document 3: Updated Engagement Letter Template

Add an AI disclosure paragraph to all engagement letter templates in Ignition/GoProposal.

Step 4: Set Up OpenAI API Access

Create an OpenAI organization account, generate an API key for the engagement letter generation pipeline, configure usage limits to prevent runaway costs, and test the API connection.

curl https://api.openai.com/v1/chat/completions -H 'Content-Type: application/json' -H 'Authorization: Bearer sk-YOUR_API_KEY' -d '{"model": "gpt-5.4", "messages": [{"role": "user", "content": "Respond with OK if you can read this."}], "max_tokens": 10}'

Step 5: Build the Make.com Automation Workflow

Create the core automation scenario in Make.com that orchestrates the full pipeline: Fireflies.ai transcript completion → data extraction → GPT-5.4 engagement letter generation → output routing to practice management and document management systems. This is the central nervous system of the solution.

{ transcript(id: "{{1.transcript_id}}") { sentences { text speaker_name } summary { action_items overview } } }

• Add error handler to Module 3 (OpenAI): retry 2x with 30-second delay.
• Add error handler to Module 5: if API fails, email draft as attachment instead.
• Add a global error notification email to MSP support.

Step 6: Configure Fireflies.ai Webhook to Make.com

Connect Fireflies.ai to the Make.com scenario so that completed transcriptions automatically trigger the engagement letter generation pipeline. This eliminates any manual export or copy-paste steps.

In Fireflies.ai

In Make.com

Test the Connection

Step 7: Create Engagement Letter Templates in Ignition

Configure 4–5 engagement letter templates in Ignition (or the firm's engagement letter platform) that will receive data from the AI pipeline. Each template corresponds to a primary service type and contains merge fields for AI-generated content.

{{client_name}}, {{client_entity_type}}, {{client_ein_or_ssn_last4}},
{{services_scope}}, {{fee_structure}}, {{fee_amount}}, {{billing_frequency}},
{{engagement_period}}, {{client_responsibilities}}, {{firm_responsibilities}},
{{deadline_dates}}, {{special_terms}}, {{ai_transcription_consent_clause}}

Client acknowledges and consents to the use of AI-powered transcription
technology during meetings and calls for the purpose of accurate record-keeping
and service delivery. All recordings and transcripts are maintained in accordance
with our firm's privacy policy and applicable professional standards.

Step 8: Update the Firm's Written Information Security Plan (WISP)

Amend the firm's WISP to document the new AI transcription tools, data flows, vendor risk assessments, and security controls. This is required by the FTC Safeguards Rule (GLBA) and is a critical compliance step that cannot be skipped. ``` # No CLI commands — this step involves policy documentation # # Add the following sections to the firm's existing WISP document: # # SECTION: AI Transcription and Document Generation Tools # # 1. System Description: # - Fireflies.ai Business (transcription...

Step 9: Configure Data Retention and PII Handling

Set up appropriate data retention policies in Fireflies.ai and configure PII redaction where available. Ensure that call recordings, transcripts, and engagement letter drafts are retained only as long as needed and are protected throughout their lifecycle.

Fireflies.ai Data Retention Configuration

OpenAI API Configuration

# Redact SSN patterns:
replace /(\d{3}-\d{2}-\d{4})/ with '[SSN REDACTED]'

# Redact EIN patterns:
replace /(\d{2}-\d{7})/ with '[EIN REDACTED]'

# Redact bank account numbers:
replace /(\d{8,17})/ with '[ACCT REDACTED]'

SmartVault / Document Storage

Step 10: Pilot Testing with Live Onboarding Calls

Conduct 5–10 real onboarding calls using the full pipeline. Each call goes through the complete workflow: verbal consent → Fireflies recording → transcription → Make.com automation → GPT-5.4 engagement letter draft → partner review. Document quality issues and refine the system.

• SCORING (1–5 scale): Accuracy of extracted information: ___
• SCORING (1–5 scale): Completeness of engagement terms: ___
• SCORING (1–5 scale): Professional tone and language quality: ___
• SCORING (1–5 scale): Time saved vs. manual process: ___

Step 11: Staff Training Session

Conduct a 2-hour training session for all firm staff who will interact with the system. Cover the technology workflow, consent procedures, draft review process, and troubleshooting common issues. Leave behind reference documentation.

Step 12: Go-Live and Monitoring Setup

Transition from pilot to production use. Enable the system for all onboarding calls, set up monitoring alerts, and establish the ongoing support cadence with the firm.

Custom AI Components

Onboarding Call Transcript Processor

Type: prompt System prompt for GPT-5.4 that processes a Fireflies.ai transcription of an accounting firm client onboarding call and extracts structured data needed for engagement letter generation. This prompt is used in the Make.com scenario (Module 3) and produces a JSON output that maps directly to engagement letter template merge fields. Implementation: ``` SYSTEM PROMPT: You are an expert engagement letter drafting assistant for a CPA / accounting firm. You will receive a transc...

Engagement Letter Draft Generator

Type: prompt Second-stage GPT-5.4 prompt that takes the structured JSON output from the Onboarding Call Transcript Processor and generates a professional, ready-to-review engagement letter draft. This produces the actual letter text that will be sent to Ignition or output as a Word document. Implementation: ``` SYSTEM PROMPT: You are a senior CPA firm engagement letter specialist. You will receive a structured JSON object containing extracted engagement details from a client onboardi...

Make.com Scenario: Onboarding-to-Engagement Pipeline

Type: workflow Complete Make.com automation scenario specification that connects Fireflies.ai transcript completion to the two-stage GPT-5.4 processing pipeline and routes the final engagement letter draft to the appropriate output destination. This is the central orchestration workflow. Implementation: ``` MAKE.COM SCENARIO SPECIFICATION ================================ Scenario Name: Onboarding Call → Engagement Letter Draft Scheduling: Instant (webhook-triggered) Max execution tim...

Custom Vocabulary List for Accounting Transcription

Type: skill Pre-built custom vocabulary configuration for Fireflies.ai that dramatically improves transcription accuracy for accounting-specific terminology. This vocabulary list covers tax forms, entity types, accounting standards, software names, and industry jargon commonly discussed in onboarding calls.

Implementation:

TAX FORMS:
1040, 1040-SR, 1040-X, 1040-ES, 1065, 1120, 1120-S, 1120-H, 990, 990-EZ, 990-PF, 990-T, 941, 940, 944, 943, W-2, W-3, W-4, W-9, 1099-NEC, 1099-MISC, 1099-INT, 1099-DIV, 1099-B, 1099-R, 1099-K, 1099-S, 1099-G, 1098, 1098-T, K-1, Schedule C, Schedule E, Schedule K-1, Schedule SE, 8829, 4562, 2553, SS-4, 8832, 7004, 2290, 720

ENTITY TYPES:
S-Corp, S-Corporation, C-Corp, C-Corporation, LLC, PLLC, LLP, sole proprietor, sole proprietorship, partnership, general partnership, limited partnership, 501(c)(3), 501c3, non-profit, nonprofit, disregarded entity, single-member LLC, multi-member LLC, professional corporation, P.C., SMLLC

# ACCOUNTING STANDARDS & TERMS

ACCOUNTING STANDARDS & TERMS:
GAAP, GAAS, SSARS, accrual basis, cash basis, modified cash basis, accounts receivable, accounts payable, general ledger, chart of accounts, trial balance, adjusted trial balance, depreciation, amortization, Section 179, bonus depreciation, MACRS, straight-line, cost basis, fair market value, FMV, FIFO, LIFO, weighted average, bank reconciliation, journal entry, adjusting entry, closing entry, fiscal year, calendar year, engagement letter, management representation letter, compilation, review, audit, agreed-upon procedures, attestation

TAX CONCEPTS:
AGI, adjusted gross income, MAGI, modified adjusted gross income, standard deduction, itemized deductions, QBI, qualified business income, Section 199A, estimated payments, quarterly estimates, extension, tax extension, amended return, carryforward, carryback, NOL, net operating loss, capital gains, capital losses, wash sale, passive activity, material participation, hobby loss, home office deduction, self-employment tax, FICA, FUTA, SUTA, nexus, apportionment, multistate

SOFTWARE & PLATFORMS:
QuickBooks, QuickBooks Online, QBO, QuickBooks Desktop, Xero, FreshBooks, Sage, Sage Intacct, Wave, Gusto, ADP, Paychex, Bill.com, BILL, Divvy, Expensify, Dext, Hubdoc, Receipt Bank, TSheets, Harvest, Karbon, TaxDome, Canopy, Drake, Lacerte, UltraTax, ProConnect, ProSeries, CCH Axcess, GoSystem

INDUSTRY TERMS:
retainer, engagement, onboarding, offboarding, bookkeeping, write-up, bank feeds, categorization, reconciliation, month-end close, year-end close, 1099 filing, W-2 processing, payroll processing, quarterly review, annual review, tax planning session, tax projection, estimated tax payment, IRS, state revenue department, notice response, audit representation, power of attorney, Form 2848, CAF number

Opt-Out Manual Workflow Fallback

Type: workflow

A documented manual workflow for situations where a client declines recording consent. Ensures the firm can still generate engagement letters efficiently without the AI pipeline, maintaining service consistency.

Implementation

TRIGGER: Client declines recording consent during verbal disclosure or in writing.

Onboarding Call Notes Template Fields

Testing & Validation

• AUDIO QUALITY TEST: From each speakerphone location, record a 60-second sample conversation using Windows Voice Recorder or macOS QuickTime. Play back and verify: (1) all speakers are clearly audible, (2) no echo or feedback, (3) accounting terms like '1120-S' and 'S-Corp' are distinguishable. Score each location pass/fail.
• FIREFLIES BOT JOIN TEST: Create a test Zoom/Teams meeting titled 'Onboarding: Test Client ABC'. Verify Fireflies bot auto-joins within 30 seconds of meeting start. Confirm the notification message appears to all participants. If bot does not join, check calendar integration and auto-join settings.
• TRANSCRIPTION ACCURACY TEST: Conduct a 10-minute scripted mock onboarding call that includes at least 15 accounting-specific terms (entity types, form numbers, dollar amounts). After transcription, compare the AI output against the script. Target: 95%+ accuracy on general speech, 90%+ on accounting terms. If accounting term accuracy is below 90%, add more terms to custom vocabulary.
• SPEAKER DIARIZATION TEST: Conduct a mock call with 3 participants (partner, client, spouse/business partner). Verify Fireflies correctly identifies and labels each speaker at least 85% of the time. Mislabeled speakers will cause confusion in the transcript and engagement letter.
• WEBHOOK TRIGGER TEST: After a test transcription completes, verify the Make.com scenario execution log shows a successful trigger within 5 minutes. Check that the webhook payload contains all expected fields (transcript_id, title, participants, duration).
• PII REDACTION TEST: Include a fake SSN (e.g., '123-45-6789') and fake EIN (e.g., '12-3456789') in a test call transcript. Verify the Make.com text replacement module correctly redacts both before they reach the OpenAI API. Check the OpenAI input in Make.com execution log to confirm redaction.
• STAGE 1 EXTRACTION TEST: Feed a sample onboarding transcript through the Onboarding Call Transcript Processor prompt. Verify the JSON output: (1) all discussed fields are populated, (2) undiscussed fields are null, (3) client name, entity type, services, and fee amount are 100% accurate, (4) no hallucinated information.
• STAGE 2 ENGAGEMENT LETTER TEST: Feed the Stage 1 JSON output through the Engagement Letter Draft Generator prompt. Verify: (1) letter follows the correct template structure, (2) all merge fields are populated, (3) [BRACKET] placeholders appear for missing data, (4) tone is professional and appropriate, (5) AI consent clause is included, (6) letter is between 1.5–3 pages.
• END-TO-END PIPELINE TEST: Conduct a full mock onboarding call covering: new S-Corp tax client, $3,500 annual fee, monthly bookkeeping at $500/month, fiscal year end December 31. Verify the complete pipeline produces an engagement letter draft within 10 minutes of call completion that accurately reflects all discussed terms.
• NOTIFICATION EMAIL TEST: After an end-to-end test, verify the partner receives the notification email with: (1) correct client name and service type, (2) fee amount and frequency, (3) clickable link to the draft document, (4) list of items requiring attention, (5) clear AI-generated disclaimer.
• CONSENT WORKFLOW TEST: Have a test participant decline recording consent during the verbal disclosure. Verify: (1) the partner smoothly transitions to manual note-taking, (2) the Fireflies bot is removed or not started, (3) the opt-out is logged in the tracking spreadsheet, (4) an engagement letter is still produced manually within 24 hours.
• COMPLIANCE DOCUMENTATION TEST: Verify all compliance artifacts are in place: (1) WISP has been updated with AI tool sections, (2) SOC 2 reports have been obtained from Fireflies and OpenAI, (3) engagement letter templates include AI consent and Section 7216 language, (4) verbal disclosure script is printed and available in each office.
• ERROR RECOVERY TEST: Simulate an OpenAI API failure (temporarily invalidate the API key in Make.com). Verify: (1) Make.com retries twice, (2) after failure, the MSP support email receives an error notification, (3) the transcript is preserved and can be re-processed manually.
• MULTI-SERVICE TYPE TEST: Conduct a test call where the client needs both tax preparation AND monthly bookkeeping. Verify the system correctly identifies it as a 'Multi-Service' engagement and generates a comprehensive letter covering both service types with separate fee sections.
• VOLUME TEST: Process 5 test transcripts in rapid succession (simulate a busy day). Verify all 5 complete successfully without timeouts, API rate limiting, or missed webhooks. Check Make.com operation consumption against the plan limit.

Client Handoff

The client handoff session should be a structured 90-minute meeting with the firm's managing partner, all partners/managers who conduct onboarding calls, and the firm's administrative coordinator. Cover the following topics:

Maintenance

ONGOING MAINTENANCE RESPONSIBILITIES: 1. WEEKLY (15 min MSP time): - Check Make.com execution logs for failed scenarios — investigate and resolve any errors - Review OpenAI API usage dashboard — verify spend is within expected range ($5–$15/month) - Scan Fireflies.ai admin dashboard for unusual activity or transcription failures 2. MONTHLY (1–2 hours MSP time): - Review engagement letter quality metrics with the firm (from tracking spreadsheet) - Analyze common edit patterns — if partners cons...

Alternatives

Microsoft Teams Premium with Copilot

Instead of Fireflies.ai + OpenAI, leverage Microsoft Teams Premium ($10/user/month) which includes built-in intelligent meeting recap with AI-generated notes, action items, and task assignments via Microsoft Copilot. Engagement letter generation would still require a separate LLM step but could use Microsoft Copilot Studio or Azure OpenAI Service for a fully Microsoft-native stack. Tradeoffs: PROS: Single vendor ecosystem (Microsoft), no third-party bot joining meetings (better client optic...

Fully Custom Pipeline with Deepgram + AssemblyAI

Replace the SaaS transcription platform entirely with direct API calls to Deepgram (for real-time streaming transcription) or AssemblyAI (for batch transcription with built-in PII redaction). Build a custom web application or use a low-code platform like Retool to create a firm-branded transcription and engagement letter portal.

Engagement Letter Platform Native AI (Ignition AI / GoProposal)

Wait for Ignition and GoProposal to release their own AI-powered engagement letter generation features (both have announced AI roadmap items). Use a simpler transcription-only setup and rely on the engagement letter platform's built-in AI to generate drafts from imported meeting notes rather than building a custom GPT-5.4 pipeline.

Budget Approach: Fathom Free + Manual Letter Creation

Use Fathom's unlimited free transcription tier for all onboarding calls. Partners manually review the AI-generated summary and action items from Fathom, then create engagement letters using existing templates in Word/Google Docs. No automation pipeline, no LLM API costs.