AI document processing
From incoming files to records your team can use.
AI document processing takes the retyping out of invoices, orders, forms and emails. We read the files you receive, extract the fields you need, check them against your business rules and send verified records to your accounting system, CRM or database. Anything unclear goes to a person together with the original document.
What AI document processing includes
Document samples
We collect typical layouts, scans and known exceptions, and agree which fields are required and which values are acceptable.
Reading and extraction
We choose the right mix of text recognition (OCR) and AI for your file types. Each extracted value stays linked to the place in the document it came from.
Checks and review
Rules check completeness, formats and duplicates. Uncertain or incomplete records go to a person together with the original document.
Delivery and quality
Approved records go to your database, CRM or accounting system. We track field accuracy, manual corrections and recurring errors.
What you get after the pilot
- A list of document types and fields, with rules for exceptions.
- A working process for reading, checking and reviewing your documents.
- A connection to the system where records should go, and a quality report from the pilot.
Intelligent document processing versus plain OCR
OCR turns a scan into text. It does not know which number is the total, which date is the due date or that a supplier has moved the VAT line elsewhere. Template tools solve this for one fixed layout and break when a new supplier sends another.
Intelligent document processing recognises the document type, finds fields wherever they are, reads tables and checks results against rules such as “line items must add up to the total”. It copes with dozens of layouts without a template for each. It still makes mistakes on poor scans and unusual documents, so a doubtful field is never written to your system silently.
Invoice data extraction and other common document types
Supplier invoices
Supplier, numbers, dates, line items, VAT and totals, matched against your supplier list and purchase orders.
Orders and requests
Customer orders in PDF or email become draft orders with products, quantities and delivery details.
Applications and forms
Applications and onboarding forms, scans included, with checks for missing signatures or fields.
Contracts and letters
Key terms, dates and parties in a summary card. Legal judgement stays with your specialist.
What this can look like
An operations team receives order forms in several layouts. The process recognises the document type, extracts the fields, flags missing information and prepares a draft order for review.
Who AI document automation is for
A good fit
- Teams that retype the same fields from invoices, orders, applications or contracts every day.
- Companies receiving documents in many layouts from different suppliers or customers.
- Processes where the extracted data goes on to a system: accounting, a CRM, an ERP or a database.
Not the right fit
- A handful of documents a month: manual entry is cheaper than setting up and checking automation.
- Documents that need a legal or expert judgement on every page. AI can prepare a summary, but a specialist still decides.
What affects the cost of AI document processing
We quote cost and timing in EUR after testing extraction on your samples. These factors matter most:
Document types and layouts
One invoice type from a few suppliers is a smaller job than invoices, orders and contracts from hundreds of senders.
File quality
Clean digital PDFs are easier than phone photos, faded scans or handwriting.
Rules and matching
Checking that fields are filled is simple; matching against orders or price lists adds logic and tests.
Where records go
A spreadsheet is quick; an accounting system or ERP needs an integration and agreed error handling.
We’ll quote cost and timing in EUR after a short brief.Discuss document automation
How accurate AI document processing is, and how we measure it
We do not promise an accuracy figure before seeing your documents. We measure it on them:
- accuracy for each field, because a supplier name and an invoice total carry different risks;
- the share of documents that pass all checks without a person;
- reviewer time on the rest, compared with manual entry;
- recurring errors, so we fix their cause rather than each case.
These pilot numbers decide whether to widen the process or keep manual entry for some document types.
How we move from sample documents to daily use
Each stage ends with something you can check before we continue.
Brief and examples
We collect sample documents, the list of fields and the rules your team uses today, including the cases that cause most trouble.
Test on real cases
We run extraction on your samples and measure field-by-field accuracy before anything is connected to a live system.
Pilot in your tools
The process handles a share of real incoming documents, with uncertain records sent to a reviewer.
Review and next step
We compare accuracy and review time with manual entry and decide which document types to add next.
What we need from you
A few dozen anonymised documents of each type (clear and poor-quality ones), the list of fields you need, the rules for checking them, and access or a test account for the system where records should go.
We also need the person who processes these documents today and knows which errors are costly. The pilot is measured against their checks.
AI makes mistakes, so we don’t promise full automation without review. Where an error is costly, a person approves the result, and confidential data goes only to services your company has approved.
AI document automation: common questions
Can the process handle tables and line items?
Yes. Line items, quantities and prices are extracted as a table and checked against the document totals. If they do not add up, the document goes to review instead of creating a wrong record.
Can you process scanned documents?
Usually yes. We check the scan quality, layout variety and languages on your samples before choosing the method, and agree the accuracy target for each field.
Will every record be sent automatically?
No, only records that pass all checks. Incomplete, contradictory or unusual documents go to a person for review.
Which document type should we start with?
One recurring type, such as invoices, orders, contracts or applications, with real examples from your team, including the awkward ones. Once that type works reliably, we add the next.
Which document formats can you work with?
PDFs, scans, photos, emails and spreadsheets. If a supplier can send data in a structured format, we use that instead of reading the document.
What about personal data in the documents?
Invoices, forms and emails often contain names, addresses and bank details. We agree which fields are extracted and stored, mask what the task doesn’t need, keep test sets anonymised and keep a log of who corrected what.
Read before you start
Short practical guides on the same subject.
- Since 2006
- Written scope before work starts
- One contact person
- Reply within 24 hours
Which documents does your team retype?
Bring a non-confidential sample and tell us where the data should go. We’ll reply with questions and a suggested first step.
The first conversation is free, with no obligation.