Implementation Guide: Generate vehicle listings, trade-in appraisal summaries, and financing option explanations

Hardware Procurement

Smartphone for Lot Photography & 360° Spins

Software Procurement

Prerequisites

• Active DMS system (CDK Global, Reynolds & Reynolds, or Tekion) with API access or scheduled data export capability. Verify DMS contract includes data access rights—CDK Fortellis APIs are pay-as-you-go; Reynolds may require partnership status.
• Inventory management platform (vAuto, Dealerslink, HomeNet, or equivalent) with inventory feed export in ADF/XML or CSV format, including VIN, stock number, pricing, mileage, and condition fields.
• Dealer website platform (Dealer.com, Dealer Inspire, DealerOn, or equivalent) that accepts inventory feed updates including vehicle description/comments fields via API or feed import.
• CRM system (DriveCentric, AutoRaptor, VinSolutions, Elead, or equivalent) with ability to receive formatted email templates for financing explanations sent to customers.
• Minimum 50 Mbps download / 10 Mbps upload internet connectivity at the dealership. 100+ Mbps symmetrical recommended if doing high-volume photo uploads.
• Firewall configured to allow outbound HTTPS (port 443) to: api.openai.com, *.n8n.cloud (or self-hosted n8n server IP), vpic.nhtsa.dot.gov, and all SaaS platform domains.
• Designated Content Manager or Internet Sales Manager (ISM) at the dealership who will serve as the human-in-the-loop reviewer for AI-generated content, particularly financing explanations.
• Dealership's legal counsel or compliance officer available for a 1-hour review session to approve AI output templates for financing content (TILA/Regulation Z and FTC CARS Rule compliance).
• OpenAI API account with billing configured and API key generated. Organization-level account recommended for MSP management across multiple dealership clients.
• Valid business email domain for SSO/MFA configuration on all AI platform accounts. No shared passwords—each user gets individual credentials per FTC Safeguards Rule requirements.

Installation Steps

...

Step 1: Environment Discovery & Audit

Conduct a thorough audit of the dealership's existing technology stack to map all integration points. Document the DMS vendor and version, inventory management platform, website provider, CRM, and any existing content generation workflows. Identify how vehicle data currently flows from acquisition to website listing. Record API credentials, feed formats, and update frequencies. This audit determines which integration paths are available and identifies any blockers (e.g., locked-down DMS APIs, legacy feed formats).

Step 2: Provision OpenAI API Account & Configure Security

Create or configure the OpenAI organization account that will power the custom trade-in and financing content pipelines. Set up proper API key management, usage limits, and monitoring. This account will be managed by the MSP, not the dealership, to maintain control over costs and security.

openai api usage --organization [org-id]  # Monitor usage after setup

Step 3: Deploy n8n Automation Server

Set up the n8n workflow automation platform that will orchestrate the entire content generation pipeline. n8n serves as the integration hub connecting DMS data feeds, VIN decode APIs, OpenAI, and content distribution channels. Self-hosting on MSP infrastructure is recommended for maximum control and multi-tenant capability.

# Option A: Self-hosted on MSP Linux server (Ubuntu 22.04 LTS recommended)
sudo apt update && sudo apt upgrade -y
sudo apt install -y docker.io docker-compose
sudo systemctl enable docker && sudo systemctl start docker

# Create n8n directory structure
sudo mkdir -p /opt/n8n/data
sudo chown -R 1000:1000 /opt/n8n/data

# Create docker-compose.yml
cat << 'EOF' > /opt/n8n/docker-compose.yml
version: '3.8'
services:
  n8n:
    image: n8nio/n8n:latest
    restart: always
    ports:
      - '5678:5678'
    environment:
      - N8N_BASIC_AUTH_ACTIVE=true
      - N8N_BASIC_AUTH_USER=mspadmin
      - N8N_BASIC_AUTH_PASSWORD=${N8N_PASSWORD}
      - N8N_HOST=n8n.yourmsp.com
      - N8N_PORT=5678
      - N8N_PROTOCOL=https
      - WEBHOOK_URL=https://n8n.yourmsp.com/
      - N8N_ENCRYPTION_KEY=${ENCRYPTION_KEY}
      - DB_TYPE=postgresdb
      - DB_POSTGRESDB_HOST=postgres
      - DB_POSTGRESDB_PORT=5432
      - DB_POSTGRESDB_DATABASE=n8n
      - DB_POSTGRESDB_USER=n8n
      - DB_POSTGRESDB_PASSWORD=${POSTGRES_PASSWORD}
    volumes:
      - /opt/n8n/data:/home/node/.n8n
    depends_on:
      - postgres
  postgres:
    image: postgres:15
    restart: always
    environment:
      - POSTGRES_USER=n8n
      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD}
      - POSTGRES_DB=n8n
    volumes:
      - /opt/n8n/postgres-data:/var/lib/postgresql/data
EOF

# Create .env file with secrets
cat << EOF > /opt/n8n/.env
N8N_PASSWORD=$(openssl rand -base64 32)
ENCRYPTION_KEY=$(openssl rand -hex 32)
POSTGRES_PASSWORD=$(openssl rand -base64 32)
EOF

# Start n8n
cd /opt/n8n && sudo docker-compose up -d

# Verify running
sudo docker-compose ps

# Option B: n8n Cloud (simpler, $20/mo)
# Sign up at https://n8n.io/cloud/ and skip the Docker setup above

Step 4: Configure Dealer Specialties StoryBuilder for Vehicle Listings

Deploy and configure Dealer Specialties StoryBuilder as the primary vehicle listing description generator. This SaaS platform handles the high-volume, VIN-decoded vehicle description writing using AI. It integrates directly with the dealership's DMS and inventory management platform to automatically generate descriptions as vehicles are stocked in.

# 1. Contact Dealer Specialties sales: 1-800-808-4898 or dealerspecialties.com
# 2. Request StoryBuilder activation for dealership account
# 3. Provide DMS integration credentials:
#    - CDK Global: Dealer code + Fortellis API subscription
#    - Reynolds: Partner integration request (may require 2-4 week approval)
#    - Tekion: REST API key from Tekion Admin Portal

# 4. Configure StoryBuilder settings in the admin portal:
#    a. Dealership Profile:
#       - Dealership name, address, phone
#       - Brand voice: [Professional/Friendly/Luxury/Value] (match dealer preference)
#       - Excluded phrases: Configure any words/phrases the dealer wants avoided
#    b. Content Settings:
#       - Description length: 150-300 words (optimal for SEO and readability)
#       - Feature highlight count: Top 5 features per vehicle
#       - Auto-regeneration: Enable (every 5-7 days for SEO freshness)
#    c. Compliance Disclaimers:
#       - Footer text: 'Pricing excludes tax, title, license, and dealer fees.'
#       - State-specific disclosures: Add per dealership location
#    d. Feed Distribution:
#       - Enable push to: Dealer website, AutoTrader, Cars.com, CarGurus
#       - Set update frequency: Every 4 hours

Step 5: Build VIN Decode Data Enrichment Pipeline

Create the data enrichment pipeline in n8n that takes raw vehicle data from the DMS feed and enriches it with detailed VIN-decoded information from NHTSA and DataOne. This enriched data feeds into the custom AI content generation prompts for trade-in summaries and financing explanations.

• In n8n, create a new workflow: 'VIN Data Enrichment Pipeline'
• Node 1: Webhook Trigger (receives inventory data from DMS feed or manual trigger) — Type: Webhook | Method: POST | Path: /vin-enrich | Authentication: Header Auth (X-API-Key)
• Node 2: NHTSA VIN Decode (free) — Type: HTTP Request | Method: GET | URL: https://vpic.nhtsa.dot.gov/api/vehicles/DecodeVinValues/{{$json.vin}}?format=json | Response: Parse JSON
• Node 3: Extract NHTSA Fields — Type: Set | Fields to extract: make, model, year, trim, engine, transmission, drivetrain, bodyClass, fuelType
• Node 4: DataOne VIN Decode (optional, for richer data) — Type: HTTP Request | Method: GET | URL: https://api.dataonesoftware.com/webservices/vindecoder/decode | Query Params: vin={{$json.vin}}&client_id=[DATAONE_ID]&authorization_code=[DATAONE_KEY] | Note: DataOne costs ~$0.05-0.15/decode
• Node 5: Merge NHTSA + DataOne Data — Type: Merge | Mode: Combine by position
• Node 6: Format Enriched Vehicle Object — Type: Code (JavaScript)

// n8n Code node (JavaScript)

const vehicle = {
  vin: $input.first().json.vin,
  stockNumber: $input.first().json.stockNumber,
  year: $input.first().json.year,
  make: $input.first().json.make,
  model: $input.first().json.model,
  trim: $input.first().json.trim,
  mileage: $input.first().json.mileage,
  askingPrice: $input.first().json.askingPrice,
  engine: $input.first().json.engine,
  transmission: $input.first().json.transmission,
  drivetrain: $input.first().json.drivetrain,
  bodyClass: $input.first().json.bodyClass,
  fuelType: $input.first().json.fuelType,
  exteriorColor: $input.first().json.exteriorColor,
  interiorColor: $input.first().json.interiorColor,
  features: $input.first().json.features || [],
  condition: $input.first().json.condition || 'Used',
  tradeInValue: $input.first().json.tradeInValue || null,
  retailValue: $input.first().json.retailValue || null
};
return [{ json: vehicle }];

Step 6: Build Trade-In Appraisal Summary Generator

Create the n8n workflow that generates professional trade-in appraisal summary documents using OpenAI GPT-5.4 mini. These summaries are sent to customers who have received a trade-in value assessment, explaining the valuation in clear, professional language that builds trust and transparency.

Step 7: Build Financing Option Explanation Generator

Create the n8n workflow that generates clear, compliant explanations of financing options for customers. This is the highest-compliance-risk component of the project. All output templates must be reviewed by the dealership's legal counsel before production use. The system generates plain-language explanations of financing terms, not binding offers.

// Code node (JavaScript) that maps stateCode from webhook payload to the
// appropriate state-level financing disclaimer

const stateDisclaimers = {
  'CA': 'California customers: See dealer for complete CCPA disclosures.',
  'NY': 'New York customers: All financing subject to NY Banking Law requirements.',
  'TX': 'Texas customers: Advertised rates subject to approved credit through participating lenders.',
  'FL': 'Florida customers: Electronic fee and tag agency fees may apply.',
  'DEFAULT': 'See dealer for complete details. All financing subject to credit approval.'
};
const disclaimer = stateDisclaimers[$input.first().json.stateCode] || stateDisclaimers['DEFAULT'];
return [{ json: { ...$input.first().json, stateDisclaimer: disclaimer } }];

Step 8: Configure DMS Data Feed Integration

Connect the dealership's DMS to the n8n automation server to enable automatic content generation when vehicles are added to inventory or when trade-in appraisals are completed. The integration method varies by DMS vendor.

• CDK Global (via Fortellis): Log into Fortellis Marketplace: https://marketplace.fortellis.io
• CDK Global (via Fortellis): Subscribe to 'Inventory Events' API (pay-as-you-go)
• CDK Global (via Fortellis): Configure webhook endpoint: https://n8n.yourmsp.com/webhook/vin-enrich
• CDK Global (via Fortellis): Set event types: vehicle.stocked, vehicle.updated, tradein.appraised
• CDK Global (via Fortellis): Test with sample webhook delivery
• Tekion: Access Tekion Admin Portal > API Settings
• Tekion: Generate REST API key with inventory read scope
• Tekion: In n8n, create a Schedule Trigger node (every 30 minutes)
• Tekion: Route new vehicles to VIN Enrichment pipeline
• Reynolds & Reynolds — Option 1 (Scheduled CSV/XML export): Configure Reynolds to export inventory feed every 4 hours
• Reynolds & Reynolds — Option 1 (Scheduled CSV/XML export): Set up SFTP server on MSP infrastructure
• Reynolds & Reynolds — Option 1 (Scheduled CSV/XML export): n8n FTP Trigger node monitors for new files
• Reynolds & Reynolds — Option 2 (Dealerslink middleware): Dealerslink pulls from Reynolds and exposes API for n8n
• Fallback (Manual CSV Upload): For dealers with restricted DMS access, create an n8n Form Trigger node with fields for VIN, stock#, price, mileage, and condition — route to VIN Enrichment pipeline

# Poll inventory endpoint for vehicles updated in the last 30 minutes

# Poll Tekion inventory endpoint for new/updated vehicles:
GET https://api.tekion.com/v1/inventory?updated_since={{$now.minus(30,'minutes').toISO()}}

curl -X POST https://n8n.yourmsp.com/webhook/vin-enrich \
  -H 'Content-Type: application/json' \
  -H 'X-API-Key: [YOUR_API_KEY]' \
  -d '{"vin":"1HGBH41JXMN109186","stockNumber":"A12345","mileage":45000,"askingPrice":22995,"condition":"Used","exteriorColor":"Silver","interiorColor":"Black"}'

Step 9: Configure Content Distribution to Website and Marketplaces

Set up the output side of the pipeline so AI-generated content flows automatically to the dealer's website, third-party marketplaces (AutoTrader, Cars.com, CarGurus), and CRM. StoryBuilder handles vehicle listing syndication directly; the custom pipeline handles trade-in and financing content routing.

# === Website Integration (for custom AI content) ===

# Dealer.com (Cox Automotive):
# 1. Contact Dealer.com support to obtain inventory feed API credentials
# 2. Use their Inventory API to update vehicle description fields:
# PUT https://api.dealer.com/inventory/{stockNumber}
# Body: { "comments": "[AI-generated description]" }

# Dealer Inspire (Cars Commerce):
# 1. Obtain SFTP credentials from Dealer Inspire rep
# 2. Push updated inventory CSV with description column to SFTP
# 3. Dealer Inspire processes on 4-hour cycle

# DealerOn:
# 1. Use DealerOn's Inventory Manager API
# 2. PATCH /api/v1/vehicles/{vin}/description

# === CRM Integration (for trade-in summaries & financing explanations) ===

# DriveCentric:
# POST https://api.drivecentric.com/v1/communications/email
# Body: { "contactId": "[ID]", "subject": "Your Trade-In Appraisal Summary",
#         "htmlBody": "[AI-generated HTML content]" }

# AutoRaptor:
# POST https://app.autoraptor.com/api/v1/leads/{leadId}/emails
# Body: { "subject": "...", "body": "..." }

# VinSolutions:
# Use VinSolutions Workflow API to create follow-up action with email content

# === Compliance Audit Log ===
# All generated content is also saved to a compliance log:
# In n8n, add a Google Sheets or database write node at end of each workflow
# Columns: timestamp, vin/stockNumber, contentType, promptVersion,
#          aiModel, outputText, reviewedBy, approvedTimestamp
# Retain for minimum 24 months per FTC CARS Rule

Step 10: Configure Photo Enhancement Pipeline

Set up Spyne AI for automated vehicle photography enhancement. Photos captured using the Ricoh Theta SC2 (interiors) and iPhone (exteriors) are uploaded to Spyne, which removes backgrounds, enhances quality, and produces marketplace-ready images that pair with AI-generated descriptions.

Step 11: Compliance Review & Template Approval

Conduct a formal compliance review of all AI-generated content templates with the dealership's legal counsel or compliance officer. This is a mandatory step before production deployment—particularly for financing explanations. Document approvals and create a compliance sign-off record.

curl -X POST https://n8n.yourmsp.com/webhook/trade-in-summary \
  -H 'Content-Type: application/json' \
  -H 'X-API-Key: [KEY]' \
  -d '{"customerFirstName":"Sample","vehicleYear":"2020","vehicleMake":"Toyota","vehicleModel":"Camry","vehicleTrim":"SE","mileage":45000,"condition":"Good","tradeInValue":18500,"marketComparables":"Similar vehicles in the area range from $17,800-$19,200","conditionNotes":"Minor cosmetic wear on front bumper. All mechanical systems functioning properly. Tires at 60% tread life.","dealerName":"Sample Motors"}'

curl -X POST https://n8n.yourmsp.com/webhook/financing-explanation \
  -H 'Content-Type: application/json' \
  -H 'X-API-Key: [KEY]' \
  -d '{"customerFirstName":"Sample","vehicleYear":"2024","vehicleMake":"Honda","vehicleModel":"CR-V","vehiclePrice":35000,"downPayment":5000,"loanTermMonths":60,"aprRange":"4.9%-7.9%","monthlyPaymentEstimate":"$565-$605","lenderName":"Honda Financial Services","dealerName":"Sample Motors","stateCode":"TX"}'

Step 12: Production Deployment & Monitoring Setup

Deploy all workflows to production, enable real-time monitoring, set up alerting, and establish the ongoing operational rhythm. This includes creating a dashboard for the dealership to track AI content generation metrics and setting up MSP monitoring for system health.

Custom AI Components

Vehicle Listing Description Prompt

Type: prompt System and user prompt pair for generating vehicle listing descriptions using GPT-5.4 mini. This is a backup/supplement to StoryBuilder for cases where the dealer wants custom descriptions with a different voice or for vehicles that StoryBuilder doesn't fully capture (e.g., specialty vehicles, modified trucks, classic cars).

Implementation:

MODEL SETTINGS:
- Model: gpt-5.4-mini
- Temperature: 0.6
- Max Tokens: 500
- Top P: 1.0
- Frequency Penalty: 0.3 (reduces repetitive phrasing across listings)
- Presence Penalty: 0.1

Trade-In Appraisal Summary Prompt

Type: prompt System and user prompt pair for generating professional trade-in appraisal summary narratives. These summaries are customer-facing documents that explain how the dealership arrived at a trade-in value, building transparency and trust. The prompt is designed to NEVER include customer PII beyond first name.

MODEL SETTINGS:
- Model: gpt-5.4-mini
- Temperature: 0.4
- Max Tokens: 800
- Top P: 1.0
- Frequency Penalty: 0.2
- Presence Penalty: 0.1

Financing Option Explanation Prompt

Type: prompt System and user prompt pair for generating clear, TILA-aware explanations of financing options. This is the highest-compliance-risk component. The prompt embeds mandatory disclaimer language and is designed to produce educational explanations, NOT binding financing offers. All output requires human review before customer delivery.

MODEL SETTINGS:
- Model: gpt-5.4-mini
- Temperature: 0.3 (lowest setting for maximum consistency on financial content)
- Max Tokens: 1000
- Top P: 0.9
- Frequency Penalty: 0.1
- Presence Penalty: 0.1

PII Firewall Middleware

Type: integration A data sanitization layer that runs before any data is sent to external AI APIs. It strips Nonpublic Personal Information (NPI) as defined by the FTC Safeguards Rule, ensuring compliance with GLBA requirements. This runs as an n8n Code node inserted before every OpenAI API call.

Implementation:

// n8n Code Node. Insert before every OpenAI API call node in all workflows.

// PII Firewall - n8n Code Node (JavaScript)
// Insert this node BEFORE every OpenAI API call node in all workflows
// Purpose: Strip NPI/PII before data leaves the dealership's control boundary

const input = $input.first().json;

// Define fields that must NEVER be sent to external AI APIs
const BLOCKED_FIELDS = [
  'ssn', 'socialSecurityNumber', 'social_security',
  'driverLicense', 'drivers_license', 'dlNumber',
  'creditScore', 'credit_score', 'ficoScore', 'fico',
  'income', 'annualIncome', 'monthlyIncome',
  'employer', 'employerName', 'employerAddress',
  'bankAccount', 'accountNumber', 'routingNumber',
  'fullAddress', 'streetAddress', 'homeAddress',
  'lastName', 'last_name', 'fullName', 'full_name',
  'email', 'emailAddress', 'email_address',
  'phone', 'phoneNumber', 'cellPhone', 'homePhone',
  'dateOfBirth', 'dob', 'birthDate',
  'coSignerName', 'coBuyerName',
  'bankStatements', 'payStubs', 'taxReturns'
];

// Create sanitized copy
const sanitized = { ...input };
let strippedFields = [];

for (const field of BLOCKED_FIELDS) {
  if (sanitized[field] !== undefined) {
    strippedFields.push(field);
    delete sanitized[field];
  }
  // Also check nested objects one level deep
  for (const key of Object.keys(sanitized)) {
    if (typeof sanitized[key] === 'object' && sanitized[key] !== null) {
      if (sanitized[key][field] !== undefined) {
        strippedFields.push(`${key}.${field}`);
        delete sanitized[key][field];
      }
    }
  }
}

// Detect and redact SSN patterns (XXX-XX-XXXX) in any string field
const ssnPattern = /\b\d{3}[-\s]?\d{2}[-\s]?\d{4}\b/g;
for (const key of Object.keys(sanitized)) {
  if (typeof sanitized[key] === 'string' && ssnPattern.test(sanitized[key])) {
    sanitized[key] = sanitized[key].replace(ssnPattern, '[REDACTED]');
    strippedFields.push(`${key} (SSN pattern)`);
  }
}

// Detect and redact phone patterns in free text
const phonePattern = /\b(?:\+?1[-.])? \(?\d{3}\)?[-.] ?\d{3}[-.] ?\d{4}\b/g;
for (const key of Object.keys(sanitized)) {
  if (typeof sanitized[key] === 'string' && phonePattern.test(sanitized[key])) {
    sanitized[key] = sanitized[key].replace(phonePattern, '[REDACTED]');
    strippedFields.push(`${key} (phone pattern)`);
  }
}

// Detect and redact email patterns in free text
const emailPattern = /[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}/g;
for (const key of Object.keys(sanitized)) {
  if (typeof sanitized[key] === 'string' && emailPattern.test(sanitized[key])) {
    sanitized[key] = sanitized[key].replace(emailPattern, '[REDACTED]');
    strippedFields.push(`${key} (email pattern)`);
  }
}

// Log what was stripped for audit trail
sanitized._piiFirewall = {
  timestamp: new Date().toISOString(),
  fieldsStripped: strippedFields,
  totalStripped: strippedFields.length,
  status: strippedFields.length > 0 ? 'FIELDS_REMOVED' : 'CLEAN'
};

return [{ json: sanitized }];

Compliance Audit Logger

Type: workflow An n8n sub-workflow that logs every AI-generated content item to a compliance audit trail. Required for FTC CARS Rule 2-year record retention. Captures the full content, prompt version, AI model used, generation timestamp, and human reviewer information.

Implementation:

// Compliance Audit Logger - n8n Workflow Specification
// Trigger: Called as sub-workflow from all content generation workflows

// INPUT SCHEMA:
// {
//   contentType: 'vehicle_listing' | 'trade_in_summary' | 'financing_explanation',
//   vehicleVin: string,
//   stockNumber: string,
//   promptVersion: string (e.g., 'v1.2'),
//   aiModel: string (e.g., 'gpt-5.4-mini'),
//   temperature: number,
//   inputPrompt: string (the full prompt sent to AI),
//   outputContent: string (the full AI-generated text),
//   dealershipId: string,
//   dealershipName: string,
//   generatedAt: ISO timestamp,
//   requiresHumanReview: boolean,
//   reviewedBy: string | null,
//   reviewedAt: ISO timestamp | null,
//   approved: boolean | null,
//   modifications: string | null (description of any human edits),
//   distributionChannels: string[] (e.g., ['website', 'autotrader', 'crm_email']),
//   piiFirewallStatus: string
// }

// WORKFLOW NODES:

// Node 1: Input Validation
// Type: Code (JavaScript)
// Validates all required fields are present

// Node 2: Write to PostgreSQL
// Type: Postgres node
// Table: content_audit_log
// SQL:
// INSERT INTO content_audit_log (
//   id, content_type, vehicle_vin, stock_number, prompt_version,
//   ai_model, temperature, input_prompt, output_content,
//   dealership_id, dealership_name, generated_at,
//   requires_human_review, reviewed_by, reviewed_at,
//   approved, modifications, distribution_channels,
//   pii_firewall_status, created_at
// ) VALUES (
//   gen_random_uuid(), $1, $2, $3, $4, $5, $6, $7, $8,
//   $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, NOW()
// )

// Node 3: Google Sheets Backup (redundant storage)
// Type: Google Sheets node
// Spreadsheet: '[DealerName] AI Content Audit Log'
// Sheet: Monthly tab (auto-created)
// Append row with all fields

// Node 4: Retention Check (runs weekly via separate Schedule trigger)
// Type: Postgres node
// Purge records older than 24 months:
// DELETE FROM content_audit_log WHERE created_at < NOW() - INTERVAL '24 months'
// (Keep 24 months per FTC CARS Rule, with 30-day buffer)

// DATABASE SCHEMA (run once during setup):
// CREATE TABLE content_audit_log (
//   id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
//   content_type VARCHAR(50) NOT NULL,
//   vehicle_vin VARCHAR(17),
//   stock_number VARCHAR(20),
//   prompt_version VARCHAR(10) NOT NULL,
//   ai_model VARCHAR(50) NOT NULL,
//   temperature DECIMAL(3,2),
//   input_prompt TEXT NOT NULL,
//   output_content TEXT NOT NULL,
//   dealership_id VARCHAR(50) NOT NULL,
//   dealership_name VARCHAR(200) NOT NULL,
//   generated_at TIMESTAMPTZ NOT NULL,
//   requires_human_review BOOLEAN DEFAULT false,
//   reviewed_by VARCHAR(100),
//   reviewed_at TIMESTAMPTZ,
//   approved BOOLEAN,
//   modifications TEXT,
//   distribution_channels TEXT[],
//   pii_firewall_status VARCHAR(20),
//   created_at TIMESTAMPTZ DEFAULT NOW()
// );
// CREATE INDEX idx_audit_dealership ON content_audit_log(dealership_id);
// CREATE INDEX idx_audit_content_type ON content_audit_log(content_type);
// CREATE INDEX idx_audit_created ON content_audit_log(created_at);
// CREATE INDEX idx_audit_vin ON content_audit_log(vehicle_vin);

Content Quality Scoring Agent

Type: agent An automated quality assurance agent that evaluates AI-generated content before it is published. It checks for compliance issues, factual consistency with VIN data, readability, and brand voice alignment. Runs as a secondary AI call that scores the primary content and flags issues for human review.

Implementation:

// MODEL SETTINGS:
// Model: gpt-5.4 (use the more capable model for QA)
// Temperature: 0.1 (near-deterministic for consistent scoring)
// Max Tokens: 500
// Response Format: JSON mode enabled

// n8n WORKFLOW INTEGRATION:
// 1. After content generation node, add this as a subsequent OpenAI call
// 2. Parse the JSON response
// 3. If recommendation === 'PUBLISH': auto-distribute to channels
// 4. If recommendation === 'REVIEW': route to human review queue with issues highlighted
// 5. If recommendation === 'REJECT': log to error queue, regenerate with modified prompt
// 6. Log QA score to compliance audit trail regardless of recommendation

// COST NOTE: GPT-5.4 at ~$2.50/1M input + $10/1M output tokens
// For 300 vehicles/month: ~$1-3/month additional cost for QA scoring
// This is negligible compared to the compliance risk mitigation value

Testing & Validation

• VIN Decode Accuracy Test: Submit 10 known VINs (mix of domestic, import, luxury, truck) to the NHTSA vPIC API endpoint via the n8n webhook and verify that year, make, model, trim, engine, and transmission are correctly extracted. Compare against actual window sticker data. Expected: 100% accuracy on core fields.
• StoryBuilder Integration Test: Verify that 5 newly stocked vehicles in the DMS appear in Dealer Specialties StoryBuilder within 4 hours and have AI-generated descriptions. Check that descriptions include correct vehicle specifications, top features, and dealership-configured disclaimer footer text.
• Trade-In Summary Generation Test: Submit a test trade-in appraisal request via the n8n webhook with known vehicle data (2020 Toyota Camry SE, 45,000 miles, Good condition, $18,500 value). Verify the output: (a) addresses customer by first name only, (b) states the correct trade-in value, (c) references mileage and condition, (d) includes the 7-day validity notice, (e) contains no PII, (f) is 200-350 words.
• Financing Explanation Compliance Test: Submit a test financing explanation request with trigger terms ($35,000 vehicle, $5,000 down, 60 months, 4.9%-7.9% APR). Verify the output: (a) presents APR as a range/estimate, (b) includes 'subject to credit approval' language, (c) mentions customer's right to choose their own lender, (d) includes the mandatory static disclosure footer, (e) does NOT contain any of the forbidden phrases ('guaranteed approval', 'everyone is approved', etc.), (f) is flagged for human review (not auto-sent).
• PII Firewall Test: Submit a test payload that intentionally includes blocked fields (SSN: 123-45-6789, email: test@test.com, phone: 555-123-4567, creditScore: 720, income: 75000) and verify that ALL PII fields are stripped before the data reaches the OpenAI API call. Check the _piiFirewall audit log shows all fields were detected and removed.
• Content Quality Scoring Agent Test: Submit a deliberately poor-quality generated description (wrong vehicle year, includes fake testimonial language like 'Customer John says this is the best car ever', and missing compliance disclaimers) to the QA agent. Verify it returns a REJECT recommendation with specific issues cited for accuracy and compliance.
• End-to-End Vehicle Listing Pipeline Test: Add a new vehicle to the DMS (or simulate via webhook) and track it through the complete pipeline: DMS → VIN decode → StoryBuilder description generation → website feed update → marketplace syndication. Verify the complete listing appears on the dealer's website within 8 hours with accurate AI-generated description and enhanced photos.
• Website Feed Integration Test: Verify that AI-generated descriptions appear correctly on the dealer's website by checking 5 random vehicle detail pages (VDPs). Confirm: description is present, no HTML rendering issues, disclaimer footer visible, no truncation of text, and photos are paired correctly.
• CRM Email Delivery Test: Trigger a test trade-in summary and financing explanation through the CRM email integration. Verify the email arrives in a test inbox with: correct formatting, dealership branding, all disclosure text visible, and no broken links or images.
• Load Test: Submit 50 simultaneous vehicle content generation requests to the n8n pipeline and verify: (a) all 50 complete successfully, (b) average response time is under 10 seconds, (c) no API rate limit errors from OpenAI, (d) all 50 entries appear in the compliance audit log. This simulates a bulk inventory refresh scenario.
• Compliance Audit Log Test: After running all tests above, query the PostgreSQL content_audit_log table and verify: (a) all generated content items have entries, (b) timestamps are accurate, (c) prompt versions are recorded, (d) the QA score is captured, (e) the human review status is correctly tracked for financing content.
• Failover Test: Temporarily disable the OpenAI API key and submit a content generation request. Verify that: (a) the n8n error workflow triggers, (b) an alert email is sent to the MSP monitoring address, (c) the error is logged, (d) the system does not crash or expose error details to the dealer interface.

Client Handoff

Conduct a 2-hour on-site or video training session with the dealership's key stakeholders. Training should cover three audiences:

Internet Sales Manager / Content Manager (Primary User - 60 minutes)

• How to review and approve AI-generated vehicle descriptions in StoryBuilder
• How to trigger manual content regeneration for specific vehicles
• How to review trade-in appraisal summaries in the CRM before delivery
• How to review and approve financing explanations (human-in-the-loop workflow)
• How to edit AI-generated content while preserving required disclaimers
• How to flag content quality issues to the MSP
• Walkthrough of the dealer-facing metrics dashboard

F&I Manager (Compliance Reviewer - 30 minutes)

• Understanding the financing explanation review queue
• What compliance elements to verify before approving (APR ranges, disclaimer presence, no forbidden phrases)
• How to request prompt template changes through the MSP
• Quarterly compliance audit process overview

General Manager / Dealer Principal (Executive Overview - 30 minutes)

• ROI metrics: time saved per listing, content consistency improvements, SEO impact
• Monthly reporting walkthrough
• Escalation paths for issues
• Service agreement review: what's included in the monthly managed service

Documentation to Leave Behind

Success Criteria to Review Together

Maintenance

• Weekly MSP Tasks (15-30 minutes): Review n8n workflow execution logs for errors or failures
• Check OpenAI API usage and costs against baseline
• Verify StoryBuilder auto-regeneration is cycling (descriptions refreshing every 5-7 days)
• Review any content flagged as REJECT by the Quality Scoring Agent
• Confirm DMS data feed is running on schedule

• Monthly MSP Tasks (1-2 hours): Generate and send monthly report to dealer: content volume, quality scores, API costs, time savings estimate
• Review and update prompt templates if dealer feedback indicates tone or content adjustments needed
• Check for OpenAI model updates or deprecation notices (model versions typically get 6-12 months notice)
• Update n8n and Docker containers
• Verify compliance audit log data integrity and retention policy compliance
• Test PII Firewall with a synthetic payload containing blocked fields

cd /opt/n8n && docker-compose pull && docker-compose up -d

• Quarterly MSP Tasks (2-4 hours): Comprehensive compliance audit: Review 20 random AI-generated content items across all three types against current FTC CARS Rule, TILA, and state-specific requirements
• Update state-specific disclaimers if regulations have changed
• Review prompt templates with dealership F&I Manager for accuracy and effectiveness
• Benchmark AI content quality against competitors' listings
• Evaluate new AI models (e.g., GPT-5.4 mini updates, new Claude versions) for potential quality or cost improvements
• Review and renew vendor contracts (StoryBuilder, Spyne, DataOne)

• Annual MSP Tasks (4-8 hours): Full system audit: security review, API key rotation, credential audit
• Compliance recertification: Have dealership counsel re-review all AI output templates
• Platform evaluation: Assess whether current stack is still optimal or if migration to newer platforms is warranted
• Pricing review: Evaluate managed service pricing against costs and market rates
• Disaster recovery test: Verify backups of n8n workflows, prompt templates, and audit logs can be restored

• Prompt Template Retraining Triggers: FTC or state regulatory changes affecting automotive advertising or financing disclosures
• Dealer rebranding or change in brand voice preference
• New vehicle segment added to inventory (e.g., dealer starts selling EVs, needs EV-specific content)
• Quality Scoring Agent flags more than 10% of content as REVIEW or REJECT in any week
• Dealer feedback indicates recurring content quality issues

• SLA Considerations: Content generation pipeline availability: 99.5% uptime (allows for maintenance windows)
• Response time for pipeline errors: 4-hour acknowledgment, 24-hour resolution during business hours
• Prompt template changes: 48-hour turnaround for non-compliance changes, same-day for compliance-critical changes
• Compliance audit findings: 72-hour remediation for critical issues, 2-week for minor issues

• Escalation Paths: Tier 1 (Dealership staff): Content review issues, approval questions → MSP Help Desk
• Tier 2 (MSP technician): Pipeline failures, integration errors, prompt tuning → Senior engineer
• Tier 3 (MSP senior engineer): API outages, DMS integration failures, security incidents → Vendor support + MSP management
• Compliance Emergency (e.g., FTC inquiry): Immediately pause all AI content generation → MSP management + Dealership legal counsel

Alternatives

...

Full Turnkey SaaS (No Custom Pipeline)

Deploy only an automotive-specific AI content platform like Dealerslink AI Description Builder, VINCUE AI-Enhanced Description Builder, or Fullpath AI Ecosystem. Rely entirely on the vendor's built-in capabilities for all three content types (listings, trade-in summaries, financing explanations) without building any custom API pipelines. The MSP's role is limited to SaaS configuration, DMS integration, and ongoing management.

Fully Custom API Pipeline (No Automotive SaaS)

Build the entire content generation system using OpenAI or Anthropic APIs with n8n automation, without any automotive-specific SaaS platform. The MSP develops custom prompts and workflows for all three content types—vehicle listings, trade-in summaries, and financing explanations. VIN data comes from NHTSA and DataOne APIs. Distribution is handled through direct website and CRM API integrations.

Impel AI Full-Stack Solution

Deploy Impel AI (formerly SpinCar) as a comprehensive AI merchandising and customer engagement platform. Impel covers vehicle photography/360° spins, AI-generated descriptions, AI sales chat, and customer communication—far beyond just content generation. It replaces multiple point solutions with a single platform.

Anthropic Claude API Instead of OpenAI

Replace OpenAI GPT-5.4 mini with Anthropic Claude Sonnet 4.6 or Claude Haiku as the AI model powering the custom trade-in and financing content pipelines. The architecture remains identical (n8n orchestration, same prompts, same integrations), only the AI API provider changes.