How often does it get things wrong?

It depends on document quality. What matters is that nothing is filed unless it passes validation — a misread becomes an exception, not a wrong invoice in your accounts.

Do we need to prepare documents in a particular way?

No. We work with what you already receive. If one supplier consistently sends something unreadable we will tell you — it is usually cheaper to ask them for a different format than to automate around it.

Where does the extracted data go?

Into the system you already use. If it has an API we write directly; if not, we automate the interface.

Does our data go to an external model?

Only if you choose that. It can also run entirely locally, with OCR and models you host — more expensive to set up and slower to run, but nothing leaves your network.

At what volume is it worth it?

As a rule of thumb: under 50 documents a month rarely justifies it. Over 200 almost always does. In between it depends on how expensive the time of whoever types them today is.

Automated document reading, including the ugly ones

Reads invoices, forms and PDFs and turns them into data

Nobody retypes documents any more

Demos always use clean invoices. Yours are not clean.

What you get

Remove the typing

Data lands straight in your system with no manual step.

Handle mixed formats

Every supplier has a different layout. We do not depend on fixed positions.

Check before writing

Totals and suppliers are checked against existing records before filing.

Surface the exceptions

Anything that fails goes to a person with the original attached.

How it works

The flow, step by step

1
1. Intake
Documents are collected from email, WhatsApp or a shared folder.
- n8n
2
2. Reading
OCR turns the image into text, then a language model identifies each field.
- OCR
- GPT-4.1
3
3. Checking
Every extracted field is compared against data you already hold.
- PostgreSQL

Document received · WhatsApp · 22:14

I sent the invoice as a photo, hope it is readable.

Read it: supplier, total and VAT match the order. Posting it now.

An unreadable photo is never posted. It goes to a person with the original attached.

Where it works well

Invoices and delivery notes from recurring suppliers
Structured forms, even scanned crooked
PDFs generated by other software
High volumes of similar documents

Where it does not

Messy handwriting — we read it, but with a human check
Very blurry or cropped photos land in exceptions often
One-off documents seen a few times a year — the cost exceeds the gain

What we use

n8n
PostgreSQL
e-Factura

Where we use it

Frequently asked questions

How often does it get things wrong?: It depends on document quality. What matters is that nothing is filed unless it passes validation — a misread becomes an exception, not a wrong invoice in your accounts.
Do we need to prepare documents in a particular way?: No. We work with what you already receive. If one supplier consistently sends something unreadable we will tell you — it is usually cheaper to ask them for a different format than to automate around it.
Where does the extracted data go?: Into the system you already use. If it has an API we write directly; if not, we automate the interface.
Does our data go to an external model?: Only if you choose that. It can also run entirely locally, with OCR and models you host — more expensive to set up and slower to run, but nothing leaves your network.
At what volume is it worth it?: As a rule of thumb: under 50 documents a month rarely justifies it. Over 200 almost always does. In between it depends on how expensive the time of whoever types them today is.

Want to see whether it works on your documents?

Claim a free audit