AI Invoice Data Extractor: Gmail to Google Sheets via LlamaParse & OpenAI

Unlock Automated Invoice Processing with this AI Agent

This AI-driven workflow streamlines your accounts payable process by automatically fetching PDF invoices from your Gmail, intelligently parsing them, extracting key information, and organizing it neatly into a Google Sheet. It's designed to handle complex PDFs, including those with tables and figures, ensuring accurate data capture without manual intervention.

This agent acts as a dedicated AI for invoice data extraction. It first monitors your Gmail for new invoices based on your filters. When an invoice PDF is detected and not yet processed (checked via a 'invoice synced' label), it's sent to LlamaParse for robust conversion into a machine-readable format (markdown). Then, an OpenAI LLM (e.g., GPT-3.5-turbo) is prompted to extract specific fields like invoice number, dates, supplier details, line items, and totals. The extracted data is structured according to a predefined schema using a Langchain output parser and then appended to your designated Google Sheet. Finally, the processed email is labeled in Gmail to prevent duplicate entries.

Key Features & Benefits

• Automated Email Monitoring: Watches your Gmail for incoming invoices with PDF attachments based on custom filters.
• Advanced PDF Parsing: Leverages LlamaParse to accurately convert complex PDFs, including embedded tables and figures, into markdown text, crucial for structured data extraction.
• AI-Powered Data Extraction: Uses OpenAI (GPT models via Langchain) to intelligently identify and extract predefined data fields from the parsed invoice content.
• Structured Data Output: Employs a Langchain structured output parser with a customizable JSON schema to ensure data is consistently formatted for direct use in spreadsheets or other systems.
• Google Sheets Integration: Automatically appends extracted invoice data to a specified Google Sheet for reconciliation, reporting, or further analysis.
• Duplicate Prevention: Labels processed emails in Gmail (e.g., 'invoice synced') to avoid redundant operations and maintain data integrity.
• Customizable Extraction Logic: Easily adapt the data fields to be extracted by modifying the JSON schema in the 'Structured Output Parser' node and tailoring the prompt in the 'Apply Data Extraction Rules' (LLM Chain) node.
• Handles Processing Delays: Includes logic to wait and check LlamaParse job status, ensuring the workflow proceeds only when parsing is complete.

• An n8n instance (Cloud or self-hosted).
• Gmail account credentials (OAuth2 recommended for secure connection).
• LlamaParse API Key from LlamaIndex Cloud. You can sign up at https://cloud.llamaindex.ai/. The free tier typically allows a generous number of PDF parses per day.
• OpenAI API Key with access to a model like gpt-3.5-turbo-1106 or newer.
• Google Sheets API credentials (OAuth2) to allow writing to your spreadsheet.
• A label named invoice synced (or your preferred name) pre-created in your Gmail account to mark processed emails.

1. Download the n8n workflow JSON file (ai-invoice-data-extractor-v1.0.0.json) and import it into your n8n instance.
2. Configure Gmail Trigger ('Receiving Invoices'):
   • Connect your Gmail account using OAuth2 credentials in the 'Receiving Invoices' node.
   • Adjust the sender filter (e.g., invoices@example.com) or other query parameters (q) as needed.
   • Ensure the downloadAttachments option is enabled.
   • The workflow checks for emails NOT having the invoice synced label. Make sure this label exists in your Gmail.
3. Configure LlamaParse HTTP Request Nodes:
   • In the 'Upload to LlamaParse' node, select or create 'HttpHeaderAuth' credentials. Name it (e.g., 'LlamaIndex API'). Set Header Name to Authorization and Value to Bearer YOUR_LLAMAPARSE_API_KEY (replace YOUR_LLAMAPARSE_API_KEY with your actual key).
   • Verify the URLs in 'Get Processing Status' and 'Get Parsed Invoice Data' nodes match current LlamaParse API endpoints if they ever change.
4. Configure OpenAI & Langchain Nodes:
   • In the 'OpenAI Model' node, select your OpenAI credential and choose a suitable model (the template uses gpt-3.5-turbo-1106). Adjust temperature if needed (0 is good for factual extraction).
   • In the 'Structured Output Parser' node, review and customize the jsonSchema. This defines the exact data points the AI will attempt to extract (e.g., 'Invoice date', 'invoice number', 'Line items'). Modify this to match your specific needs.
   • In the 'Apply Data Extraction Rules' (LLM Chain) node, critically review and refine the prompt in the text parameter. This prompt guides the LLM on how to extract information from the markdown content (provided by LlamaParse) based on your schema. The existing prompt is a good starting point.
5. Configure Google Sheets Node ('Append to Reconciliation Sheet'):
   • Connect your Google Sheets account using OAuth2 credentials.
   • In the documentId field, select your target Google Spreadsheet.
   • In the sheetName field, select the specific sheet (tab) where data should be appended.
   • The 'Map Output' node prepares data based on the 'Structured Output Parser'. Ensure your Google Sheet has columns corresponding to the keys in your jsonSchema for seamless data entry.
6. Configure Gmail Action Node ('Add "invoice synced" Label'):
   • Select your Gmail account.
   • Ensure the correct labelIds for your 'invoice synced' label is selected to mark emails after successful processing. You can find this ID by listing labels or using the picker if available.
7. Test Thoroughly: Send a test email with a PDF invoice that matches your Gmail trigger filters. Manually execute the workflow or wait for the trigger to pick it up. Check each step for errors and verify data in Google Sheets.
8. Activate the workflow to begin fully automated invoice processing.