AI-Powered Invoice Processing: from Email to Database & Chat Notifications

Automatically process PDF invoices directly from your email inbox. This workflow uses AI to extract key data, saves it to a PostgreSQL database, and instantly notifies you about the new document in your preferred chat application.

The workflow listens for new emails, fetches PDF attachments, and then passes their content to a Large Language Model (LLM) for intelligent recognition and data extraction. Finally, the information is securely archived in the database, and a summary of the invoice is sent as a notification.

📝 This workflow is highly customizable.
It uses PostgreSQL, OpenAI (GPT), and Discord by default, but you can easily swap these components.
Feel free to use a different database like MySQL or Airtable, another AI model provider, or send notifications to Slack, MS Teams, or any other chat platform.

⚠️ Note: If the workflow fails to extract data correctly from invoices issued by certain companies, you may need to adjust the prompt used in the Basic LLM Chain node to improve parsing accuracy.

Use Case

• Automating accounts payable for small businesses and freelancers
• Centralizing financial documents without manual data entry
• Creating a searchable database of all incoming invoices
• Receiving real-time notifications for new financial commitments

Features

• 📧 Email Trigger (IMAP): Monitors a dedicated email inbox for new messages with attachments
• 📄 PDF Filtering: Automatically identifies and processes only PDF attachments
• 🤖 AI-Powered Data Extraction: Uses an LLM (e.g., GPT-4o-mini) to extract invoice number, buyer/seller details, amounts, currency, and due dates
• ⚙️ Structured Data Output: Converts AI output to standardized JSON
• 🔍 Database Write Logic: Prevents duplicates by checking invoice/company combo
• 🗄️ PostgreSQL Integration: Stores extracted data into company and invoice tables
• 💬 Chat Notifications: Sends invoice summary as message to a designated channel

Setup Instructions

⚠️ API Access & Costs

To use the AI extraction feature, you need an API key from a provider like OpenAI.
Most providers charge for access to language models. You'll likely need a billing account.

1. PostgreSQL Database Configuration

Ensure your database has the following tables:

-- Table for companies (invoice issuers)
CREATE TABLE company (
    id SERIAL PRIMARY KEY,
    tax_number VARCHAR(255) UNIQUE NOT NULL,
    name VARCHAR(255),
    address TEXT,
    created_at TIMESTAMP WITH TIME ZONE DEFAULT CURRENT_TIMESTAMP
);

-- Table for invoices
CREATE TABLE invoice (
    id SERIAL PRIMARY KEY,
    company_id INTEGER REFERENCES company(id),
    invoice_number VARCHAR(255) NOT NULL,
    -- Add other fields: total_to_pay, currency, due_date
    created_at TIMESTAMP WITH TIME ZONE DEFAULT CURRENT_TIMESTAMP,
    UNIQUE(company_id, invoice_number)
);

Then, in n8n, create a credential for your PostgreSQL DB.

2. Email (IMAP) Configuration

In n8n, add credentials for the email account that receives invoices:

• IMAP host
• IMAP port
• Username
• Password

3. AI Provider Configuration

• Log in to OpenAI (or similar provider)
• Generate API key
• In n8n, create credentials and paste the key

4. Chat Notification (Discord)

• Go to Discord > Server Settings > Integrations > Webhooks > New Webhook
• Select channel
• Copy Webhook URL
• In n8n, paste URL into the Discord node

Placeholders and Fields to Fill

Testing the Workflow

1. Send a test invoice to the inbox as a PDF attachment
2. Run the workflow manually in n8n and check if the IMAP node fetches the message
3. Verify AI Extraction – inspect the LLM output (e.g., GPT node) and confirm structured JSON
4. Check the DB – ensure new rows appear in company and invoice
5. Check the chat – verify the invoice summary appears in the chosen channel

Customization Tips

• Change the DB: Use MySQL, Airtable, or Google Sheets instead of PostgreSQL
• Other notifications: Swap Discord for Slack, MS Teams, Telegram, etc.
• Expand AI logic: Extract line items, prices, etc. by customizing the prompt
• Add payment logic: Allow marking invoices as paid via emoji or a separate webhook