Automating Invoice Data Extraction with OCR, AI, and Google Sheets

The Manual Process

Processing invoices manually is a repetitive workflow: find the document, open it, read the relevant information, copy the values, and enter them into a spreadsheet. When invoices arrive regularly, even a simple process like this can consume a significant amount of time.

Document Collection and OCR

I built an internal n8n workflow to automate this process from end to end. The workflow starts by locating a dedicated Google Drive folder used for payment invoices. It then finds the subfolder for the current month and retrieves PDF and image files stored there.

Each document is downloaded from Google Drive as binary data and sent to OCR.Space for text recognition. The OCR configuration is optimized for Japanese documents and enables table recognition, scaling, and the OCR Engine 2 parser.

Cleaning and AI Extraction

The extracted text is then cleaned before being passed to an AI model. The cleanup step removes unnecessary whitespace, normalizes line breaks, and standardizes quotation marks. This gives the AI a cleaner input and reduces noise from the OCR output.

The AI extraction step uses GPT-5.4-nano to convert the unstructured invoice text into structured JSON. The expected data includes the invoice date, due date, client name, subtotal, tax, total, bank information, invoice items, and additional notes.

Google Sheets Integration

The structured response is parsed into JSON and passed directly to Google Sheets. The workflow appends key invoice fields such as invoice date, due date, client name, subtotal, tax, total, bank name, and account number as a new spreadsheet row.

The workflow can be started manually during development or through a webhook, making the same processing pipeline available for automated execution.

The Complete Pipeline

The important part of this approach is that each stage has a clearly defined responsibility: Google Drive handles document discovery and storage, OCR converts documents into text, AI transforms that text into structured data, and Google Sheets becomes the final destination.

This eliminates most of the repetitive copy-and-paste work involved in invoice processing while keeping the workflow relatively simple to maintain. It also provides a practical example of combining traditional automation with OCR and AI rather than relying on AI alone.