OCR
We turn PDFs, images and scanned documents into structured data that can be reviewed and used. Reading the document is only the first step.
REQUEST A DIAGNOSTICWhat OCR is
OCR stands for optical character recognition: it identifies text inside an image or a scanned document. It is a first step, not the end of the process. A list of recognized words may still require manual work; a structured invoice lets you review its fields and continue with the process. Recognizing a number does not tell you by itself whether it is a total, a tax or a line item.
OCR, templates and language models
Classic OCR
It converts the image into text. It does not know what each data point means.
Template-based extraction
It is taught where each field is in each document format. It is precise with stable formats and stops working when the layout changes.
Extraction with language models
The model reads the image and returns the requested fields without a template per supplier. It copes better with varied formats and lower-quality scans, but it may accept a doubtful value instead of leaving it empty.
Which documents
Invoices and receipts
Totals, taxes, issuer details and line items.
Forms
Fields filled in on a computer or by hand, checkboxes and signatures.
Contracts
Parties, dates, amounts and clauses that need to be located.
Other documents
PDFs, images and scans with information that someone copies by hand today. What changes from one type to another is the fields that are extracted and how they are reviewed.
An example: Scanera
An invoice can arrive as a PDF, as a photograph or as a scanned image. Receipts also have to be read. Copying their data by hand leaves the information scattered. Scanera accepts PDFs and images of invoices and receipts and follows this flow.
Learn about ScaneraExtract
Information such as totals, customer details and invoice lines is extracted, not a list of loose words.
Review
The fields are presented so they can be checked against the original document. Before using the data, amounts and lines should be compared with the original.
Classify and match
Customer and supplier invoices are organized and matched automatically. An issued invoice and a received one serve different purposes; keeping that distinction lets each record be reviewed in its context.
Export
Reviewed data can be exported to a custom Excel file. The fields and the output format follow how the team works.
Examples of what we have already done
Invoices and receipts
Reading invoices and receipts and passing their fields to accounting software or an ERP.
Forms
Turning paper forms or scanned PDFs into data organized in a database.
Contracts
Extracting parties, dates and amounts from contracts so they can be found and consulted.
Delivery notes and transport
Reading delivery notes and transport documents and recording their data in the logistics system.
Customer onboarding
Reading documents and proofs provided at sign-up and prefilling the fields, with human review.
Historical archive
Digitizing old documents so they can be searched.
When it helps
This flow helps when documents arrive in varied formats and several people need the same fields. It also makes it possible to identify where human judgment is still needed.
Where a person is still needed
- Extraction does not replace checking against the original document.
- Incomplete documents, ambiguous amounts or unexpected data require review even if they were extracted automatically.
- Automatic matching does not remove the need to check doubtful data.
- Image quality affects the result: blurry photos, faded thermal receipts and stamps over the text make it worse.
- Line items and tables are the hardest to extract well.
- Documents often contain personal data. If they are processed with external services, you need to know where they are handled and under what contract.
FREQUENTLY ASKED QUESTIONS
What is OCR?
OCR stands for optical character recognition: it identifies text inside an image or a scanned document. It is a first step, not the end of the process.
What types of documents can be processed with OCR?
Invoices, receipts, forms and contracts, among others, as PDF, image or scan. What changes from one type to another is the fields that are extracted and how they are reviewed.
Does OCR replace human review?
No. Extraction does not replace checking against the original document. Incomplete documents, ambiguous amounts or unexpected data require review even if they were extracted automatically.
What can be done with the extracted data?
Review it field by field and export it with the fields and format the team uses, for example to Excel. For invoices, also match each one to its customer or supplier.
Does it work with phone photos and handwriting?
With phone photos, yes, if the image is legible. Handwriting was the weak point of classic OCR. Language models with vision read it better, although it is still where review matters most.
What accuracy can be expected?
It depends on the document. With computer-generated PDFs it is high; with poor-quality scans, low. What matters is accuracy per field, and that is only known by measuring it on your own documents.
Is OCR still needed with electronic invoicing?
For structured electronic invoices, no: they already carry the data in a machine-readable format. OCR is still useful for receipts, paper documents, PDFs and images.
What is needed to get started?
A sample of real documents, the list of fields you need and where that data has to end up.
From the example to your process
We can work with your team to study which documents you receive, which fields you need and where that data has to end up.