Automated Invoice Processing & ERP Entry

The Goal: Zero-Touch Invoice Processing

Processing PDFs manually is a massive operational bottleneck. By leveraging multimodal AI models (like GPT-4o) and an orchestration tool (like Make.com), we can build an autonomous pipeline that reads incoming emails, extracts structured data from messy PDF invoices, and pushes it directly into your accounting software.

Here is the exact step-by-step blueprint to build this agent.

---

Step 1: The Email Webhook (Trigger)

Create a new scenario in Make.com. Your first module will be an Email / Gmail Custom Webhook.

• Set up a dedicated email address (e.g., invoices@yourcompany.com).
• Configure the module to trigger only when an email contains a .pdf attachment.
• Pro-tip: Add a router here to filter out internal company emails to prevent spam from triggering the agent.

Step 2: AI Vision Extraction (The Brain)

Add the OpenAI (ChatGPT) module and select the Create a Chat Completion action. We will use the gpt-4o model because of its superior vision capabilities. Map the downloaded PDF attachment from Step 1 into the image/document input field.

You must force the AI to return Valid JSON. Use this exact System Prompt:

System Prompt: You are a highly precise accounting data extraction assistant. Your job is to read the provided invoice document and extract the following information into a strict JSON format.

Do not include markdown formatting or conversational text. Output ONLY valid JSON.

Required fields:

• "vendor_name": (string) The name of the company issuing the invoice.
• "invoice_number": (string) The unique invoice ID.
• "date_issued": (YYYY-MM-DD)
• "total_amount": (number) The final total including tax.
• "line_items": (array of objects containing "description", "quantity", and "price")

Step 3: JSON Parsing & Validation

Add a JSON -> Parse JSON module in Make.com. Map the exact output from the OpenAI module into this parser.

Add a Make.com Filter after this module to act as our first guardrail:

• Condition 1: total_amount exists.
• Condition 2: vendor_name is not empty.

If the AI hallucinates or fails to read the PDF properly, the filter stops the workflow and routes the invoice to a human for manual review.

Step 4: Push to ERP (The Action)

Add the QuickBooks (or Xero) module to your scenario. Select the Create a Bill/Invoice action.

• Map the parsed vendor_name to the Vendor ID field (you may need a search module right before this to match the name to an existing internal ID).
• Map the total_amount and map the array of line_items directly into the accounting software's line-item arrays.

---

📥 1-Click Import (Make.com Blueprint)

Don't want to build this from scratch? Download our pre-configured Make.com JSON file below.

1. Download the invoice-extraction-v1.json file.
2. Open Make.com, create a new scenario, click the three dots at the bottom, and select "Import Blueprint".
3. Connect your OpenAI and QuickBooks API keys, and you are ready to process invoices!

Download Blueprint .JSON