Typing numbers from a PDF into a spreadsheet is a hundred-year-old job that somehow survived into your evenings. It is exact, repetitive, and completely rule-bound, which makes it perfect robot food.
What it does
You email a receipt to one address. The workflow pulls the attachment, AI reads it like a bookkeeper (vendor, date, total, tax, category), and writes one clean row into your bookkeeping sheet. Then it files the email away under a processed label. PDFs and photos of paper receipts both work.
The unglamorous rule that makes it trustworthy: when a field is unreadable, the workflow writes nothing rather than guessing. A blank cell gets your attention; a confidently wrong number gets into your tax filing.
How it works
A Gmail trigger watches the receipts label. The attachment gets extracted (text from PDFs, the image itself for photos) and handed to the AI step with strict instructions to return structured fields or nulls, never inventions. The row lands in the sheet, the email gets relabeled, and month-end becomes a review job instead of a typing job.
Setup
- Import the workflow file into your n8n instance.
- Create a Gmail label or alias for receipts and set the filter rule the workflow's note describes.
- Connect an AI provider key (a vision-capable model if you want photo receipts, not just PDFs).
- Connect Google Sheets and pick your bookkeeping sheet.
- Forward one clean PDF and one crumpled photo receipt and check both rows.
The template runs on n8n, any recent version, cloud or self-hosted. Every place that needs one of your accounts is marked with a note inside the workflow.
Get the workflow
Rather have this running in your business without touching a single node? That is the actual job. Write to hello@nurho.net or start with the ROI calculator to see if it is even worth automating.