Still typing data from PDFs into your system? What an AI document assistant actually does

Somewhere in most businesses there's a person with two windows open. On the left, a PDF. On the right, the system it needs to go into. They read a number, type it, read the next one, type that. Order forms from customers. Delivery dockets from suppliers. Timesheets from subcontractors. Test certificates, purchase orders, booking requests that arrive as attachments.

Nobody hired them to do that. It just grew. And because it's spread across the week, nobody has ever added it up.

This is one of the jobs AI is now reliably good at. Not "AI runs your business". Just: read this document, pull out these fields, put them there, and let a person check it.

The test: one week of copy and paste

Before talking about tools, find out what you actually have. For one ordinary week, ask whoever does the data entry to keep a tally.

1. What documents come in? List each type: supplier bills, customer orders, dockets, forms, certificates.
2. How many of each? A rough count is fine.
3. Where does each one end up? Xero, a job system, a spreadsheet, a register.
4. How many minutes does each take? Opening, reading, typing, checking, filing.
5. How often is something wrong later? A wrong quantity, a missed PO number, a date typed the American way.

Then look at the answers and ask three questions:

• Is it the same kind of document over and over? Fifty order forms from a dozen customers is a good fit. Fifty completely different letters is not.
• Are the fields clear? "Customer, PO number, items, quantities, delivery date" is easy to define. "Work out what this person wants" is harder.
• Does a mistake matter? If a typo means the wrong delivery goes out, a consistent process with a check step is worth more than speed.

If you've got a few hours a week of the same few document types going into the same place, you've found a good candidate.

What an AI document assistant does

An AI document and data assistant is a small tool built around one job. A typical one works like this.

Documents arrive where they already arrive. Usually a shared inbox. The assistant watches it, or staff forward attachments to a dedicated address, or upload a photo from their phone. Nobody has to change how customers or suppliers send things.

It reads each document and pulls out the fields you care about. Customer, order number, line items, quantities, dates, totals. It works on PDFs, scanned pages, photos of paper dockets and the body of an email. It handles the fact that every supplier lays their paperwork out differently, which is where older template-based scanning tools tend to fall over.

A person checks before anything is saved. The review screen shows the original document on one side and the extracted values on the other, with anything uncertain highlighted. Your staff member confirms or fixes it in seconds instead of typing it from scratch. You choose how strict this is: every document checked, or only the ones flagged as unsure.

The data goes where it belongs. Into your job or order system, a database, a register, or Xero as a draft bill. The original document is filed against the record so anyone can find it later.

It keeps a record. What came in, what was extracted, who approved it, and when. That makes it easy to answer "did we ever get that docket?" without searching three inboxes.

What stays the same

Your suppliers keep sending what they send. Your customers keep ordering the way they order. Your accounting system stays where it is. The people who know the work still make the calls; they just stop doing the typing.

The assistant also doesn't decide anything important on its own. If the job is "work out whether this order should be accepted", that stays with a person. What it can do is put the order in front of them already filled in.

Is it safe to put our documents through AI?

It's a fair question, and the answer depends on how the tool is set up. Ours use commercial AI services with terms that rule out training on your data, and your documents are stored in Australian hosting with access limited to the people who need it. If you want the longer version of what that means in practice, our piece on what "your data stays in Australia" actually means covers it.

What we'd suggest either way: start with one document type that isn't your most sensitive, see how it performs on your real paperwork, and widen it from there.

Where it fits with what you already use

Plenty of businesses already use a receipt or bill-capture app linked to their accounting software, and for standard supplier bills those work well. Keep using it.

The gap is usually everything that isn't a standard bill: customer order forms, delivery dockets that need matching to jobs, subcontractor timesheets, compliance certificates that need logging against a staff member or site. Those don't go into the accounts. They go into the part of the business that runs on spreadsheets and inboxes, and that's where a custom assistant earns its place. Often it's the first piece of a job or workflow tracker, because once orders arrive as clean data, the rest of the process can follow.

If the data currently lands in a spreadsheet that only one person understands, our piece on replacing that spreadsheet is worth reading alongside this one.

When it isn't worth it

• The volume is small. If it's a dozen documents a week and they take two minutes each, keep doing it by hand.
• Every document is different. If each one needs reading and judgement, AI can summarise it for you, but it won't remove the work.
• An existing app already does it. If your bill-capture app or job platform already reads the documents you care about, check what you're already paying for first.
• The documents are the problem. If the real issue is that orders arrive incomplete, the better fix might be a proper order form on your website so the data is clean from the start.
• Nobody would check it. A review step only works if someone owns it. If that's not realistic yet, sort out who does before building anything.

What working with us looks like

We start with a 30-minute call and look at a handful of your real documents and where the data needs to go. You get a written scope and a fixed price. Then we build a free working prototype of the core feature: your documents going in, the fields coming out, a review screen to check them. You'll see how accurate it is on your own paperwork, not on a demo. If it doesn't do what we agreed, you walk away. If cost is the first question, our pricing page explains how we work it out.

You see it working before you pay.

Send us a few sample documents and we'll tell you honestly whether an assistant is worth building.