Automate vehicle insurance claim approvals with OpenAI, OCR and PDF API Hub

Description

This workflow automates the full lifecycle of a vehicle insurance claim — from an incoming Gmail email to a signed, watermarked PDF decision letter delivered back to the claimant.

It monitors Gmail for emails with a PDF claim attachment, deduplicates them against a Google Sheet, parses the claim PDF using PDF API Hub's OCR, evaluates the claim with OpenAI GPT-4o-mini (or a built-in mock evaluator for testing), and routes it to one of three paths:

Approved** — generates a signed approval letter, stamps it APPROVED, uploads to Google Drive, and emails the claimant
Rejected** — generates a rejection letter, stamps it REJECTED, uploads to Google Drive, and emails the claimant
Manual Review** — notifies your claims adjuster by email for human review

Every claim is logged and tracked in a Google Sheet.

What you need

Gmail OAuth2** — to monitor the claims inbox and send decision emails
Google Sheets OAuth2** — for claim logging and duplicate detection
Google Drive OAuth2** — to store processed claim PDFs
OpenAI API key** — only required when ai_mode is set to openai
PDF API Hub API key** — get one free at pdfapihub.com (30 free API calls)

Setup

Google Sheets

Create a Google Sheet with a tab named Claims and add these headers in row 1:

claim_id | claimant_name | claimant_email | claim_type | decision | approved_amount | reasoning | risk_flags | summary | processed_at | email_message_id

A sample sheet you can clone is linked in the Sticky Note — Overview node inside the workflow.

After creating the sheet, open every Google Sheets node in the workflow and point it to your sheet and tab.

Google Drive

Create two folders in your Drive: Approved Claims and Rejected Claims. Set each folder in the corresponding Upload node.

Workflow Config node

Edit the Workflow Config Set node to configure:

| Field | Description |
|---|---|
| self_email | Your email — receives duplicate-claim alerts |
| claims_adjuster_email | Email for manual review notifications |
| ai_mode | mock for testing, openai for production |
| mock_decision | APPROVED, REJECTED, or MANUAL_REVIEW (mock mode only) |
| company_name | Used in generated PDF letters |
| support_phone | Used in generated PDF letters |
| appeals_email | Used in generated PDF letters |

How it works

New Claim Uploaded — Gmail trigger polls every minute for emails with subject containing "insurance claim" and a PDF attachment
Check Duplicate in Sheet — looks up the Gmail message ID in the tracking sheet to skip already-processed emails
Extract Claim Data — sends the PDF to PDF API Hub's Parse PDF node and returns the full claim text as JSON
AI Mode Router — routes to OpenAI or Mock evaluator based on ai_mode config
AI Claim Evaluator / Mock AI Evaluator — produces a structured decision with decision, reasoning, approved_amount, risk_flags, and claimant details
Parse AI Decision — extracts the structured JSON from the AI response
Log Claim to Sheet — appends the claim record to Google Sheets
Route by Decision — branches into Approved, Rejected, or Manual Review paths
PDF generation — HTML-to-PDF nodes generate branded decision letters using PDF API Hub
Sign + Watermark — approval letters are digitally signed and stamped APPROVED or REJECTED
Upload + Email — PDFs are uploaded to Drive and the decision is emailed to the claimant

Customisation tips

Edit the AI system prompt in the AI Claim Evaluator node to match your own underwriting rules and risk thresholds
Switch mock_decision in Workflow Config between APPROVED, REJECTED, and MANUAL_REVIEW to test all three paths without making any API calls
Edit the HTML inside the letter-generation nodes to match your company branding
Adjust the >$50,000 → MANUAL_REVIEW threshold directly in the AI prompt

PDF API Hub nodes used

Parse PDF** — extracts claim text from the uploaded PDF attachment
HTML to PDF** ×2 — generates the approval and rejection letter PDFs
Sign PDF** — adds an authorised digital signature to the approval letter
Add Watermark** ×2 — stamps APPROVED (green) or REJECTED (red) across the letter

A note on quality

This is not an AI-generated template. AI was used as a coding assistant for certain pieces, but every path and every input has been manually validated end-to-end — including sending a real claim email, receiving the decision email back, verifying the signed and watermarked PDF was uploaded to the correct Google Drive folder, and confirming the Google Sheet was updated with the claim record. Both the OpenAI path and mock mode were tested manually across all three decision outcomes: APPROVED, REJECTED, and MANUAL_REVIEW.