What intelligent document processing is
Intelligent document processing uses AI to read invoices, delivery notes, forms and other PDFs, pull out the fields that matter, and post them into Xero or your own system. Each field gets a confidence check, and anything doubtful goes to a person. We build it for UK businesses still keying documents in by hand.
Documents we read, and where they go
| Document | What gets pulled out | Where it lands |
|---|---|---|
| Supplier invoices | Supplier, invoice number, dates, line items, VAT, totals | Xero or Sage as a draft bill, matched to the purchase order |
| Delivery notes | Items, quantities, order reference, signature present or not | Your stock or job system, checked against what was ordered |
| Application and order forms | Names, addresses, choices ticked, handwritten notes | Your CRM or case system as a new record |
| Certificates and compliance PDFs | Issuer, reference, expiry date | The supplier or asset record, with a reminder before expiry |
What happens to each document
Arrives
From a mailbox, a scanner folder, an upload page or a supplier portal. The original file is stored untouched.
Read
The AI reads the page, including scans and photos, and returns each field with how sure it is.
Checked
Simple rules run next: totals add up, VAT is right, the supplier exists, the order number is real.
Posted or queued
Clean documents go straight into the system. Anything doubtful waits in a review queue with the field highlighted.
Logged
Every value is recorded with its source page and who approved it, so an auditor can trace it back. Postings can be reversed.
You may not need AI for this
Plenty of document work is already handled by tools you pay for.
When AI extraction is worth building
- Documents come from dozens of suppliers, each with its own layout.
- The data has to go into your own system, which no off-the-shelf capture tool supports.
- There are handwritten notes, stamps or poor scans to deal with.
- Each document needs checking against an order, a job or a contract before posting.
Use Xero's capture tool or plain OCR
If it's mainly supplier bills going into Xero, its own bill capture or a tool such as Dext will handle most of it. If every document uses the same template, plain OCR with fixed positions works. We'll point you there first.
What moves the price
How many different layouts arrive, how poor the scans are, and how much checking each document needs against your other records. The target system matters too. Posting a bill to Xero is simpler than writing into an old database with no API.
We count the volumes and the keying time in the Workflow Audit, then quote each phase at a fixed price. If the target system is the real blocker, a discovery spike checks we can write to it first.
Questions about document processing
How accurate is AI at reading invoices?
It depends on the documents. Clean digital PDFs read well. Poor scans and handwriting read less well. That is why every field carries a confidence score, the totals are checked by plain rules, and anything doubtful goes to a person. We measure it on your own documents before you rely on it.
Can it post invoices into Xero?
Yes. It creates draft bills in Xero through its API, with the supplier, lines and VAT filled in and the PDF attached. Someone in accounts approves them as they do now.
Where do our documents go when the AI reads them?
It can all run inside your own cloud account, so documents never leave it. The original files, the extracted data and the audit log are stored there too. Nothing is used to train a public model. See [secure AI platforms](/services/secure-ai-platform).
What happens when it reads a field wrong?
A person corrects it in the review queue, and the correction is logged against the document. Anything already posted can be reversed. Repeated mistakes on one supplier's layout tell us where to adjust the checks.
Send us a sample of the pile
On a free call, show us a handful of the documents and where they end up. If Xero's own capture or a standard tool would do the job, we'll say so.