Invoice OCR: reading purchase invoices and receipts automatically

A restaurant receives invoices every week from the wholesaler, the butcher, the drinks supplier and the cleaners. An installer collects receipts from the builders’ merchant in the van. All that paper and all those PDFs have to go into the books: supplier, date, amount and VAT retyped and the invoice stored. Invoice OCR takes over the retyping, not the thinking.

This article is about invoices you receive. For the invoices you send yourself, see automating invoicing from invoice to payment.

What OCR is and what AI adds

OCR stands for optical character recognition: software that recognises text in a scan, photo or PDF. Classic OCR only returns loose text. The hard part is knowing which number is the total, which is the VAT amount and which date is the invoice date. That is where AI helps: the model recognises the structure of an invoice, even when every supplier uses a different layout.

The result is a suggestion: a completed purchase entry with supplier, invoice number, date, amounts and VAT. Someone reviews that suggestion and approves it.

How the process runs

1. Arriving: invoices come in through a dedicated email address, or you take a photo of a receipt with your phone.
2. Reading: the software extracts supplier, invoice number, date, amounts, VAT and IBAN from the document.
3. Recognising: the supplier is linked to an existing contact, with the ledger account you used last time.
4. Checking: calculation rules and comparisons with earlier invoices look for anything unusual.
5. Approving: a member of staff reviews the suggestion next to the original and approves or corrects it.
6. Passing on: the entry goes to your accounting package and the document is stored.

What goes well

• Digital PDFs from regular suppliers: clean documents with a fixed layout are usually read well.
• Recurring invoices: rent, phone and software arrive in the same form every month.
• Recognising suppliers: after a few invoices the system knows which contact and which account belong to them.
• Searching: every document is searchable in one place, instead of in a shoebox or a mail folder.

What does not go well

This is the difference between a demo and real life. Expect these awkward cases:

• Receipts: crumpled, faded or photographed at an angle. Thermal paper also fades over time.
• Several VAT rates: a wholesale receipt with products at different rates has to be split correctly.
• Credit notes: a negative amount read as a normal invoice turns your books upside down.
• Foreign invoices: different date formats, currencies and reverse-charged VAT.
• Several invoices in one PDF: or an invoice with a delivery note behind it.
• Handwritten notes: such as a discount the supplier added in pen.

Checks you build in

A read invoice is a suggestion, not a fact. Good checks catch most mistakes before they reach your books. More on checking AI output in general in verifying AI output.

Calculation rules

• Totals: do the lines and the VAT add up to the total?
• VAT rate: does the rate match what this supplier normally charges?
• VAT number: does the number have a valid structure?

Comparing with earlier invoices

• Duplicate invoices: the same invoice number from the same supplier does not go through twice.
• Changed IBAN: if the bank account differs from earlier invoices, always have someone verify it by phone. This is a well-known form of invoice fraud.
• Unusual amount: an invoice much higher than normal gets extra attention.

When the system is unsure, that has to be visible. A field the software is not confident about should be highlighted, so the member of staff knows where to look.

Is it worth it?

Work it out with your own figures. Count for a month how many purchase invoices and receipts you process and how much time each document takes. Set that against the time for checking, because that stays. The more documents from the same suppliers, the bigger the difference.

In Vorlyo’s software, reading documents is a task for the AI assistant: it makes the suggestion, you approve. It connects to the same suppliers and projects as the rest of your administration. An overview of the industries we build software for is on software by industry.