15 hours
a month retyping
300 invoices at 3 minutes each, before any checking.
Technology: OCR and data extraction
Invoices and documents read automatically, with nothing retyped.
Invoices, delivery notes and shipping statements arrive as PDFs, sometimes scanned. Someone opens them again and retypes the data. The software reads them, extracts the fields you need and checks them before they are used.
“He understood the assignment, he asked good relevant questions and delivered excellent work that did not need correcting.” Verified client, data extraction from PDFs
At a glance
- Used for
- Reading invoices, delivery notes, statements and orders without retyping
- Not needed for
- Structured e-invoices (XML): the data is already in fields
- Where it runs
- On your own servers if you like: documents stay in the company
- The key point
- Every extracted value is checked, never taken on trust
How it works
From PDF to ready data, in 4 steps.
The problem
The data is already there, printed in a PDF.
A supplier sends the invoice as a PDF, the courier sends the shipping statement, the warehouse receives the delivery note. The figures are all there, but to use them someone has to retype them: into a spreadsheet, into the ERP, into a check.
With a few documents it is manageable. With hundreds a month it becomes a full-time job, and every manual copy brings a few mistakes with it.
In detail
What happens at each step.
There is no single OCR that fits everything. The first step is knowing which document you are looking at.
- 01
The document
Structured e-invoice, PDF with text inside, or a scan: three different cases, three different paths. In a structured e-invoice the data is already in fields and OCR is not needed.
- 02
The reading
In a PDF with text the data is read directly; only scans need OCR. Then the fields are extracted, with one template per supplier or courier.
- 03
The check
Lines must add up to the total, tax must match, codes must exist. Whatever does not add up does not pass: it goes to a person.
- 04
The data, ready
The data goes where it is needed: the ERP, a spreadsheet, a comparison with order or contract.
When it pays off
It pays off when documents are many and look alike.
It makes sense if
- You receive hundreds of documents a month, or a few dozen very long ones
- Layouts repeat: the same suppliers, the same couriers
- The data then has to be compared or loaded into an ERP
- Today someone retypes it by hand
You do not need it if
- You only receive structured e-invoices and your ERP already reads them
- You get only a few documents a month
- Every document is different and a person has to read it anyway
An example with numbers
300 invoices a month: where the time goes.
Reference figures, to redo with your own.
30 invoices
to look at, afterwards
If one in ten has something that does not add up, the person opens only those.
8 lines
wrong out of 400
An OCR that is 98% right still gets 8 lines out of 400 wrong. That is why every line is checked.
We already do it. In the shipping statements project the program reads the PDFs of several couriers, each with its own template, hundreds of lines per invoice, and recalculates every shipment against the contract. Read the full case
The limits
What OCR does not do on its own.
It makes mistakes on scans
A 1 read as a 7, a 0 as an 8. That is why numbers are checked against totals, never taken on trust.
Every layout has to be learned
A new supplier or a changed layout has to be added. We plan for it at the start instead of finding out later.
It does not understand context
It reads “10% discount”, but whether that discount was due is in the contract, not the document.
Doubts stay with a person
When a value fails the checks, the system flags it instead of guessing.
Frequently asked
Questions on this topic.
Do structured e-invoices need OCR?
No. An XML e-invoice, like the Italian SdI format or a Peppol UBL invoice, already has its data in structured fields: it is read directly, with no OCR. OCR is for scanned PDFs and documents that arrive only as images.
How accurate is it?
On a PDF with text inside, reading is exact. On scans it depends on quality: that is why every extracted value goes through checks on totals and codes, and whatever does not add up goes to a person.
Do the documents leave the company?
Not necessarily. The software can run on your own servers without sending documents to outside services.
Is artificial intelligence needed?
Not always. For recurring layouts, templates and rules are enough. AI helps when layouts change often, but the result still has to be checked.
From how many documents does it pay off?
Usually from a few hundred a month. Sooner, if the data then has to be compared with orders, price lists or contracts.
Start with the problem
Send us a sample of your documents.
An invoice, a delivery note or a shipping statement: we will tell you whether it can be read automatically, and how.
- The first check is free and commits you to nothing
- Every message gets read, there is no call centre
- Fixed price and payment by milestone
- Italian and English, VAT invoices
- No phone call unless you want one
- If nothing should be built, we will say so
Prefer to write directly? [email protected]
Message received.