Document extraction
Document extraction: scanned PDFs into structured data
We turn scans, PDFs, forms and photos into the fields and tables your systems need, in the format you already use, with OCR from the cloud you already run or entirely on your own servers.
Can AI turn scanned PDFs and images into structured data?
Yes. A pipeline reads the document, finds the fields you care about (parties, dates, amounts, reference numbers, line items) and writes them into your table, database or system, sorted by your rules.
Which OCR does it use?
The one that matches the cloud you already run, not the one with the best benchmark: AWS Textract, Azure AI Document Intelligence or Google Document AI. Published benchmarks are run on clean test sets, so we first test the provider you already use on your own sample scans. Staying in your own cloud saves transfer costs and keeps access control in one place.
When documents can’t leave your infrastructure at all, the whole pipeline runs locally.
Do you need an LLM to extract data from documents?
Often not. Pulling fields out of documents is a narrow, repeatable job. When the layout is fixed, rules are enough. When the layout varies but the job stays the same, a small model trained for extraction usually does it faster and cheaper than a large language model, without sending the text to a model vendor. We use a large model where the documents need real reasoning.
How do you know it’s accurate?
Before we build anything, we agree the pass mark with you: which fields, what accuracy, measured on which of your documents. The extraction is tested against that set before launch, and it does not go live until it clears.
When don’t you need it?
When every document comes from one template, a rules-based parser is cheaper. When it’s a one-off batch of a few hundred pages, manual entry may cost less than a build. We’ll tell you which on the call.
Questions about this service
Which documents can it read?
Scanned and digital PDFs, images and photos of documents, forms and tables. Send one real sample to the call and we’ll tell you what’s realistic.
Can it run without sending documents to the cloud?
Yes. OCR and extraction can run entirely on your own servers, with small local models instead of a cloud API.
What format does the data come out in?
Yours: a spreadsheet layout, a database table or a record in your system through its API.
What does it cost?
Extraction projects depend on the documents, so we quote them after seeing a sample. The quote is fixed before we start.
Bring us one workflow. We’ll tell you what it needs.
A 30-minute call. Bring one real process or one sample document. We’ll tell you whether it needs an LLM, a small model, or just a script, and what it would take. No deck.