AI document processing: from PDFs and emails to clean data
Invoices, quotes, contracts, application forms, certificates: if your team reads documents and types what they find into another system, AI can do most of that work. I build document processing that extracts the right fields, checks them, and only asks a person when something is unclear.
Background
Experience: document AI at Herald
Extracting data from files and emails is part of what Herald offers brokerages and carriers, and I lead the AI behind it. Insurance paperwork is long, varied and unforgiving, which makes it a good test: the same kind of engineering works for any document-heavy business.
- Works on PDFs, scans, spreadsheets and email bodies
- Structured output checked against rules before it is saved
- Uncertain fields sent to a person, not guessed
- Source references so anyone can verify a value
Documents I automate
Invoices and receipts
Supplier, amounts, VAT, line items and due dates straight into your accounting software.
Quotes and purchase orders
Products, quantities and prices matched against your catalogue and turned into orders.
Contracts and policies
Parties, key terms, dates and obligations pulled out and summarised for review.
Application and intake forms
Client or customer details from any form layout, checked for missing information.
Certificates and compliance
Expiry dates, coverage and references tracked, with reminders before anything lapses.
Emails and attachments
Requests and documents that arrive by email, read and processed without opening them by hand.
How it works
1. Read
Text and layout are read from any format, including scans and photos.
2. Extract
A language model fills a structured schema with exactly the fields your system needs.
3. Check
Values are validated: totals add up, dates make sense, IDs exist in your system.
4. Review and save
Anything uncertain goes to a person with the document and the suggested value side by side. Clean data goes into your system with a link back to the source.
How I usually work
- 1
Understand the work
Look at how the task is done today, how often, by whom, and what a good result looks like. Often a simple automation is enough and no AI is needed.
- 2
Pick one workflow
Choose the task with the most repetitive hours and a clear definition of done, and measure it before changing anything.
- 3
Build and test on real cases
Build the workflow on real data and tools. At first a person approves the AI’s work, so nothing goes out unchecked.
- 4
Measure, then expand
Compare before and after. Let it run where it has proven reliable, then move on to the next workflow.
Frequently asked questions
What is AI document processing?
AI document processing, also called intelligent document processing, uses AI to read documents like PDFs, scans and emails, extract the information you need into structured data, and send it to your systems. Unlike old OCR templates, it handles documents in any layout.
How accurate is it?
It depends on your documents, which is why every project starts by testing on a sample of your real documents. Fields the AI is unsure about go to a person instead of being guessed, so errors are caught before they reach your systems.
When is it worth it?
When documents are processed every day. Running costs are typically cents per document, so the real question is how many hours of typing it removes, which is worth measuring first.
Does it work with scans and photos?
Yes, as long as a person can read them. Very poor scans or handwriting may need a person to check more often.
Where does the extracted data go?
Into your accounting software, CRM, ERP, database or spreadsheet, or any tool with an API. If a tool has no API, the data can be prepared for import.
Is my data safe? What about GDPR?
It should be designed with GDPR in mind from the start: data kept in the EU where possible, access limited to what the AI needs, and a person approving sensitive actions. The setup also needs documenting for your privacy records.
Related articles
What building AI agents in insurance taught me about AI workflows
Article · From the field · 4 min read
AI workflow examples: 12 workflows small businesses automate first
Article · Use cases · 4 min read
Thinking about AI for your company, or working on something similar? I’m happy to talk it through. You can send me an email or book a call.