How to Automate Invoice Data Extraction with Make, OpenAI, and Airtable

Typing data from PDFs into a database is a colossal waste of human intelligence. A standard operations team easily loses 15 hours a week just keying invoice amounts into a ledger.

Old-school OCR tools are dead. They break the second a vendor moves a logo or changes a font. Models like GPT-4o read messy documents flawlessly—but only if you force them to output structured data.

By the end of this guide, you will build an automated pipeline that catches emailed invoices, extracts the exact vendor and pricing data, and logs it into Airtable without a single manual keystroke.

Phase 1: Configure the Database

Your database must mirror the exact JSON schema you request from OpenAI. If the names mismatch, the automation breaks at the final step.

1. Step 1. Create the Airtable Base: Build a new table named "Invoices". Add three exact fields: "Invoice ID" (Single line text), "Vendor Name" (Single line text), and "Total Amount" (Currency).

Phase 2: Build the Trigger and File Conversion

We start the pipeline in Make.com. The goal is to filter out the noise, isolate the invoice emails, and convert the PDFs into a format the AI can actually read.

1. Step 2. Set up the Gmail Watcher: Add the Gmail "Watch Emails" module. Select the folder "INBOX". In the "Filter" field, enter exactly: `has:attachment filename:pdf`. Set "Mark as read" to Yes.
2. Step 3. Iterate Attachments: Connect the Gmail "Iterate Attachments" module to the watcher. Map the "Attachments[]" array from Step 2 into this module. This isolates the PDF file data.

1. Step 4. Convert PDF to PNG: Add a CloudConvert "Convert a File" module. Map the file buffer from the Iterator. Set the Input Format to `pdf` and the Output Format to `png`.

Phase 3: The OpenAI Structured Extraction

This is the brain of the operation. Do not use pre-built Make modules for OpenAI. We call the API directly to use Structured Outputs, which guarantees a flawless JSON response.

1. Step 5. Configure the HTTP Request: Add the Make "HTTP - Make a request" module. Set the URL to `https://api.openai.com/v1/chat/completions`. Set Method to `POST`. Add an Authorization header with `Bearer YOUR_API_KEY`.
2. Step 6. Build the Request Body: Change the Body type to "Raw" and Content type to "JSON (application/json)". Paste the configuration below, mapping the base64 image data from CloudConvert into the image_url.

{
  "model": "gpt-4o-2024-08-06",
  "messages": [
    {
      "role": "user",
      "content": [
        {"type": "text", "text": "Extract the invoice data."},
        {
          "type": "image_url",
          "image_url": {"url": "data:image/png;base64,{{1.data}}"}
        }
      ]
    }
  ],
  "response_format": {
    "type": "json_schema",
    "json_schema": {
      "name": "invoice_extraction",
      "strict": true,
      "schema": {
        "type": "object",
        "properties": {
          "vendor_name": {"type": "string"},
          "invoice_id": {"type": "string"},
          "total_amount": {"type": "number"}
        },
        "additionalProperties": false,
        "required": ["vendor_name", "invoice_id", "total_amount"]
      }
    }
  }
}

Phase 4: Parse and Sync

The API returns a stringified JSON object buried inside the choices array. You have to parse it back into variables Airtable can actually read.

1. Step 7. Parse the Output: Add a "Parse JSON" module. Map the `choices[1].message.content` string from the HTTP module into the JSON string field.
2. Step 8. Create the Airtable Record: Add the Airtable "Create a Record" module. Select your "Invoices" base and table. Map the parsed `vendor_name`, `invoice_id`, and `total_amount` variables directly into the matching database fields.
3. Step 9. Configure Error Handling: Machines fail. Right-click the Airtable module and select "Add error route". Connect a Slack "Create a Message" module to your operations channel, and map the Gmail message link for manual review.