The New AI Employee
How-to

Intelligent document processing: how a machine reads your paperwork

Intelligent document processing reads invoices and forms, checks the fields and passes them on. How it works in plain words, and what runs today.

The New AI Employee4 min read
A scanned delivery note under a magnifying lens, with highlighted fields lifting off the page into a clean table.
Short answer

Intelligent document processing reads a business document the way a clerk would. It pulls out the fields that matter, checks them against what you already hold, and passes them on as a draft. Anything it can't read exactly comes back to you.

What is intelligent document processing, in plain words?

Intelligent document processing reads your paperwork the way a clerk would, picks out the details that matter and hands them on.

Say a small wholesaler gets supplier bills all week. A clerk would read each one, find the supplier, the date, the number and the total, and type them in. A machine does the same reading, then checks the numbers add up before anything goes further.

People also call it AI document processing or document processing automation. The idea is the same. Nobody is hunting for the number at the bottom of page two any more.

Here's what it looks like for a few kinds of paper.

DocumentWhat gets checkedStatus
Supplier invoiceSupplier is on your list, lines add up to the total, number not used beforeRunning today
Credit noteMatches the original invoice and its open balanceRunning today
Expense receiptBelongs to the right person and periodBuilt to order
Client or supplier formEverything required is there and nothing has expiredBuilt to order

How does a machine read a document?

There are three ways: templates for layouts that never change, scan reading for images, and an AI model for everything else.

Think of a plumber who buys from five regular suppliers. Their invoices look the same every month, so a template knows where the total sits. A photo of a paper bill needs scan reading first, which turns the picture into text. A bill from a supplier you use once a year goes to an AI model, which finds the details on a page it has never seen.

This is why automation document processing works better than a single trick. Each document gets the simplest method that reads it reliably.

The simple methods are fast and easy to check. The AI model is kept for the awkward ones. That keeps things predictable, and it keeps the odd surprises small.

  • Templates: for a sender whose layout never changes, the system knows where each field sits.
  • Scan reading: turns a scan or a photo into text, the first step for any image.
  • AI model: reads the document like a person and labels the fields, even on a layout it hasn't seen.
  • Rules after reading: whichever method read it, fixed rules decide what happens next.

What checks catch the mistakes?

The checks after reading matter more than the reading itself, because they decide whether a field can be trusted.

Say a bakery's flour bill is read as the wrong total because a decimal point was blurry. A check notices the lines don't add up to the total. The bill is set aside for you, not entered.

The usual checks: does the supplier match one on your list, do the lines add up, has this invoice number been entered before, is it clearly an invoice and not a statement or a quote. A document that fails any check comes back to you with a note.

It's like a careful colleague who says, 'I'm not sure about this one, can you look?' instead of quietly typing something that might be wrong. That small habit is what makes automatic reading safe to rely on.

Interactive

Is this kind of document ready for automatic reading?

Pick one kind of document your business handles, such as supplier bills or delivery notes. Tick every statement that is true for it today.

Where does the data go?

Read data should land where your team already works, usually as a draft: a bill in the accounts, a row in a register or a line in a sheet.

That's what AI document automation means in practice. Not a pile of text, but a finished draft in the place you'd have typed it. Here's the route for one invoice.

  1. The invoice arrives in the mailbox or a shared folder.
  2. The system decides what kind of document it is: an invoice, a credit note, a statement or something else.
  3. It reads the fields, by template, scan reading or model.
  4. It checks each field against what is already known.
  5. If every check passes, it makes a draft bill with the PDF attached and files the PDF.
  6. If any check fails, it sends the document to you with a note of what was unclear.
  7. It writes every action to the log, and the result is ready.

What about bad scans and photos?

Poor scans and phone photos can usually be read, but they get a closer look and go back to you when they can't be read.

Picture a site manager sending a crumpled receipt photo taken in the rain. The system tries. If a figure is unclear, it doesn't guess. It asks.

A good habit is to ask people on site to photograph documents flat, in good light, with all four edges showing. It costs them ten seconds and it saves you a lot of hand-typing later.

Handwriting is harder. If a document is mostly handwritten, expect more of them to come back to you.

How do you start?

Start with the one document type that arrives most often and can be checked against something you hold. For most businesses that's supplier invoices.

Use the checklist above to test one type. Then count how many arrive each month and how long each one takes by hand.

Our AI employees run on our servers against your own accounts, today with Google Workspace mail, Google Drive, Google Sheets and Xero. Every step starts with your approval, then you choose when it goes automatic. One command stops it, and the records stay yours. You rent it by the month, with one price per workflow in the quote.

Questions

What is intelligent document processing?

Software that reads a business document, pulls out the fields, checks them and passes them on as a draft.

What is the difference between scan reading and intelligent document processing?

Scan reading turns an image into text. Intelligent document processing goes further: it finds the fields, checks them and prepares the entry.

How accurate is AI document processing?

Every field is checked against what you hold, and anything unclear comes back to you. You review the drafts.

Can it read handwriting?

Sometimes, but handwriting is harder, so more of those come back to you.

Where does the extracted data go?

Into your own tools, usually as a draft bill, a register row or a sheet line, with the original file kept.

Tell us your workflow. We will build it around you.

Tell us which document you handle most. Write to hello@thenewaiemployee.com and we'll say how it could be read.

Talk to us Related service: Documents and filing

Keep reading

Write to us

Tell us which hours you want back.

Three lines are enough: what you do, what takes your time, how many hours a week it costs. Write to hello@thenewaiemployee.com.

Offices in Sheridan (Wyoming), Panama City, London, Bucharest and Milan.