Invoice OCR: a reader that does not lie

“Invoice OCR”, “OCR reader”, “invoices to Excel”, “AI OCR”. The use case is boring, which is why it works: a PDF or a photo arrives; you need issuer, tax ID, number, date, bases, VAT rates, tax and total — without typing them into the ERP again.

We have an invoice OCR reader for that. It reads the document, proposes fields, applies rules (base + VAT ≈ total, tax-ID checksum, duplicate) and leaves a queue. A human confirms. We do not post blind. We do not train a public model on your books.

What it is not: a Verifactu system, homologated certified-digitisation software, or a licence to shred the archive. Anyone selling all three in one slogan is mixing RD 1007/2023, Order EHA/962/2007 and a text reader.

What an OCR reader does (and does not)

OCR turns image or PDF into text. On top, a model or templates try to label fields. Useful accuracy is not “99 % in the demo”. It is: of twenty of your invoices this week, how many pass the rules untouched, how many go to doubt, how many to alarm.

The reader does not decide VAT. It does not onboard a new supplier. It does not replace the ledger. It does not issue your sales invoice. If you need to issue under a SIF / VERI*FACTU, that is outbound software. OCR lives inbound.

“OCR to Excel” is an honest first month when the ERP has no draft tray. Fixed columns, original PDF beside the row, a status (ok / doubt / alarm). After the queue holds, you wire the ERP. Writing straight into the VAT book is how you dirty the quarter.

Free, price, the sample that matters

Free one-shot PDF OCR exists. It tests a file. It is not a habit: no queue, no rules, no processor, the file ends in a personal account.

Serious price is implanting: twenty of your documents (not the vendor’s), including crooked photos, credit notes, exempt lines and a locked PDF. A bar with fifty invoices a month does not need a group pack. A workshop with three hundred and foreign suppliers needs different rules.

Kit Digital (process management, or processes with AI) can pay to implant that flow if there is a deliverable, training and a habit. It does not pay for a loose OCR licence with no queue and no owner.

Paper, certified digitisation, the myth of throwing everything away

Digitising is not mandatory in 2026 just because OCR exists. Keeping invoices still follows VAT and commercial rules. If you want the digital copy to replace paper and destroy the original, you need certified digitisation: AEAT-homologated software (Order EHA/962/2007), a faithful signed image, a documentary database with ledger fields. Text OCR is not a homologation criterion. Pulling a tax ID is not that digitisation.

Ranking educational pages lead with that legal split, then the product. So do we. The reader stops retyping. Paper retention or certified digitisation is another line — another vendor if needed.

Privacy

An invoice carries company data and sometimes people. Processor, residency, no-training clause, deletion. The pilot uses this week’s flow, not five years “so the model learns”. If data must not leave the network, we can read on our own server. Decide that before the first PDF.

How it sits next to Verifactu

If you issue with a SIF, Verifactu (or non-VERI*FACTU mode) is about that issuance. OCR will not get you out of Word. It will not get you into VERI*FACTU. They can coexist: you issue on one side; on the other, you stop retyping inbound bills. Two flows.

See what Verifactu is and assisted billing.

A week-one setup that survives Monday

Pick one mailbox or one folder as the inbox. Ten invoices, not a hundred. Run the reader. Count ok / doubt / alarm. Sit with the person who actually posts VAT. If they do not trust the ok pile, tighten the rules — do not raise the “accuracy” slogan.

Keep the original PDF next to the extracted row. If a field is wrong, they correct it in the queue, not in the ledger first. After five quiet days, then talk ERP. The firms that skip that week spend a quarter cleaning duplicates.

Frequently asked questions

Do you have an invoice OCR reader?

Yes. It extracts fields, applies rules, and leaves exceptions to a human. It is not a Verifactu SIF and not certified digitisation.

Can I send invoices to Excel?

Yes, as the first-month tray. Fixed columns, original PDF, status. Not as the eternal VAT book.

Is the OCR free?

One-shot PDF tests exist. The habit (queue, rules, processor) is not a free one-click site.

Does it work on receipts?

Sometimes, with worse accuracy. We measure national supplier invoices first. Receipts and foreign invoices come later, not mixed into the pilot’s “success”.

If admin lives on the keyboard, request an assessment with ten PDFs from this week (no extra personal data). That beats a 99 % slogan. Related: certified digitisation (OCR is not a licence to shred).