Document intake and extraction

What is inside

7 items. Includes video guide, deployment guide, sample data.

• n8n/intake.jsonWatches Gmail and Drive, runs the extraction, files the result.
• schemas/invoice.jsonThe fields to extract from an invoice, with the confidence rule per field.
• schemas/application.jsonThe same for an application form; a template for your own document type.
• prompts/extract.mdThe extraction prompt: fields, evidence, confidence, and the refusal to guess.
• samples/Fifteen fictional documents of three kinds for a dry run.
• GUIDE.mdDefining a document type, the confidence threshold, and where the doubtful ones go.
• walkthrough.mp4Ninety seconds: a PDF arrives, the record appears, one lands in review.

Who it is for

• Someone retypes what is in the PDF into the system every day.
• The documents arrive by email and land in a folder nobody owns.
• You want the uncertain ones to reach a person, not the database.

Requirements

• n8n
• An Anthropic API key
• Gmail or Outlook, Google Drive or SharePoint
• QuickBooks, Airtable or a system with a create endpoint

Install

1. 01

   Import the workflow and connect the mailbox and folder.

2. 02

   Copy a schema and edit the fields to your document type.

3. 03

   Drop the samples in the watched folder and read the review queue.

The recipe this ships

1. trigger

   step

   step

   step

   output

   Tools in the chain: Gmail, then PDF, then Applied AI, then QuickBooks, then Google Drive.

   Paperwork reads itself

   Watches the inbox and the shared folder, reads every invoice, application or submission that arrives, and files the fields into the right system with the original attached.

   Cost per invoice from $9.40 to $2.78; insurance submissions cleared in one minute instead of 32Source: Ardent Partners and FurtherAI

   OperationsSee the build

Where this sits

1. This system

   Document intake and extraction

   In the library, running.

2. Pairs with

   Installed and tuned

   For a document type of your own, the installed rung: we build the schema against your real paperwork and tune it for a week.

3. In your stack

   Deployed on an engagement

   The audit finds which system earns its place first; we install and tune it around how you work.

Questions

How accurate is it?

The card's number is the published one: cost per invoice from $9.40 to $2.78 in Ardent Partners' data. On your documents, the confidence threshold decides what a person sees; the guide shows how to set it.

What document types are included?

Invoices and applications, with the schema format documented so you can add your own.

Where do the doubtful ones go?

A review folder and a Slack or email note, with the field it doubted named.

What if it does not fit our stack?

Then we say so, and the guide names the swap. Every pack documents exactly what it needs to run.

Why is this the most expensive pack?

Two schemas, a review queue and the most wiring. It is also the pack that removes the most typing.

Changelog

1. August 28, 2026In the library. Version 1.0 runs in our own operation.