aiprocesai.lt
Process review
LT|EN

Document OCR automation

AI document OCR: invoices and documents into your ERP without re-typing

OCR is only the first step. The system recognises the document, extracts the fields, checks them against your rules and passes the approved record to your ERP or accounting system. People handle only the exceptions.

Rivilė, Directo, B1.lt or your other system stays the system of record.

Book a process reviewHow it works
  • PDFs, scans, photos
  • invoices, orders, CMRs
  • ERP entry after checks
Illustration: processing a supplier invoice
Invoice.pdf · received by emailPDF

Document type: supplier invoice

  • 6 fields extracted
  • Supplier found
  • Order matched
Needs reviewVAT does not match the line items
The responsible employee approves or corrects the field.
NEXTReady for ERP

Which documents

What the system can take in

Typical documents we automate with OCR:

  • Supplier invoices

    PDFs, scans and e-invoice attachments

  • Purchase orders

    Orders with line items and quantities

  • Transport documents and CMRs

    Carrier documents, shipment data

  • Delivery notes

    Delivery and goods-received documents

  • Excel order forms

    Customer and supplier spreadsheets

  • Customer and supplier PDFs

    Orders, specifications, quotes

  • Photos and scans

    Documents photographed on a phone

  • Email attachments

    Documents arriving in a shared inbox

Each document type and format is configured during implementation, based on your examples – not out of the box.

OCR and document automation

What OCR is – and why it is not enough

OCR (optical character recognition) turns an image of a document into text. For a business that is not enough: the text still has to be understood, checked and entered into a system. Example – invoice OCR into Rivilė or another ERP:

Today

By hand

  1. Email with an invoice
  2. The attachment is opened
  3. Fields are re-typed by hand
  4. Checked against a spreadsheet or order
  5. Entered into the ERP
  6. Mistakes are fixed later

Every document means the same steps again.

OCR only

Recognised text

  1. PDF or scan
  2. Text is recognised
  3. The fields still have to be found and checked
  4. Still entered into the ERP by hand

Faster to read, but the work remains.

With the system

Document automation

  1. The document arrives
  2. The document type is recognised
  3. The required fields are extracted
  4. Checked against rules and data
  5. Exceptions go to a person
  6. The approved record goes to the ERP or accounting
  7. Every action is logged

People handle exceptions instead of re-typing documents.

Checks and human review

A person sees only what is unclear

Every field gets a state. Clear fields are accepted, unclear ones wait for a person’s decision, and without required data no record is created in the ERP.

Illustrative example
Document review

Incoming document

Invoice.pdf

Received by email · 1 page

Extracted fields

  • Document typeSupplier invoiceField accepted automatically
  • SupplierSupplier LtdField accepted automatically
  • Invoice no.INV-000123Field accepted automatically
  • Date28 Sep 2026Field accepted automatically
  • Amount, euros1,240.00Field accepted automatically
  • VAT, euros254.20Review required

    Does not match the line items (260.40).

    The responsible employee approves or corrects the field, or rejects the document.
  • Order no.PO-0458Blocked

    Order not found in the ERP – no record is created until this is resolved.

  • ERP destinationPurchase ledgerField accepted automatically
Ready for ERPonce the flagged fields are resolved · every action is logged

What the system checks

Example rules – the actual ones are set for your process.

  • The supplier exists in your ERP
  • The invoice number is not a duplicate
  • Totals reconcile with the lines
  • The order exists
  • Customer or supplier matches the order
  • Required fields are present
  • Your business rule is met

Integration

Your ERP or accounting system stays the system of record

The system does not replace your ERP – it removes the manual work before the data reaches it.

  1. 01

    Incoming documents

    • Email
    • PDF
    • Excel
    • Photo or scan
    • Portal

  2. 02

    Document processing

    • Type recognition
    • Field extraction
    • Checks

  3. 03

    Control

    • Business rules
    • Human review
    • Exception queue

  4. 04

    Your system

    • ERP
    • Accounting
    • CRM
    • TMS
    • Internal system

    System of record – unchanged

Your data stays yours.

We use business-grade APIs that don’t train models on your data.

Hosting per the contract.

For sensitive processes the system can run on your own or an EU server, and data processing is agreed in the contract.

Limited permissions.

The system writes only where you allow it, and only the records agreed.

Review before writing.

You can require every ERP entry to wait for a person’s approval.

Action log.

What was done with each document, when and why is stored.

System and support per the contract.

Handover, maintenance and monitoring are set out in the proposal and agreement.

Process examples

Where document automation works in practice

Illustrative process examples – not specific client deployments

Finance

Supplier invoice

  1. The invoice arrives by email
  2. Supplier, totals and VAT are extracted
  3. Checked against the order
  4. Record in Rivilė or another ERP

Totals don’t match – to the bookkeeper

Purchasing

Purchase order

  1. Order as PDF or Excel
  2. Line items and quantities are extracted
  3. Products are matched to ERP codes
  4. The order is entered into the ERP

Unknown product code – to the manager

Logistics

CMRs and carrier documents

  1. Scan or photo
  2. Shipment data and dates are extracted
  3. Checked against the order
  4. Record in the TMS or ERP

Missing date or signature – to the dispatcher

Wholesale

Customer order PDF

  1. Customer order as a PDF
  2. Customer and products are matched
  3. Stock is checked
  4. The order is created in the ERP

Price differs from the contract – to the manager

Accounting

Expense documents

  1. Receipt or invoice in a photo
  2. Amount, date and seller are extracted
  3. An accounting category is assigned
  4. Record after approval

Unclear category – to the bookkeeper

Custom document

Your company form

  1. Company PDF or Excel form
  2. Fields are mapped to your structure
  3. Checked against the rules
  4. Record in the internal system

New form version – for review

Is it a fit

When document automation fits – and when it doesn’t

A good fit when

  • Many repeated, similar documents
  • Meaningful document volume
  • Data is re-typed by hand today
  • A clear target system for the data
  • Predictable business rules
  • Recurring exceptions that can be defined

Not a fit when

  • Only a few documents a month
  • Every document is fundamentally different
  • No stable process
  • Source quality is unusable for recognition
  • Automation would cost more than it saves

Not sure? We assess this in the process review. How the process review works →

Implementation

Document OCR automation: from process review to launch

The first three stages don’t commit you to anything. Pricing is provided after the process review and scope assessment; after that we work under the proposal and the agreement.

How we price →Our services →Examples from analysis →

  1. Process review

    You show the document flow and a few typical documents on your screen. We listen and ask the right follow-up questions.

    Video call · 30–45 min

  2. Process and scope description

    We describe the document types, fields, check rules, exceptions and the first-phase scope in writing.

    In writing

  3. Proposal

    If there is a fit, we provide a proposal for the first phase. If the system would not pay back, we say so in writing.

    No commitment

  4. Implementation

    After the contract, a technical validation tests access, connections and recognition on your documents. Then we build the first working phase.

    Usually 6–10 weeks

  5. Testing

    The system first runs in test mode on your real documents. Records are switched to write into your ERP or accounting system only after review.

    Together with your team

  6. Launch and support

    We launch the system and train your team. Support and monitoring after launch are defined separately in the proposal and agreement.

    Per agreement

Questions

Frequently asked questions about document OCR

What is OCR, and how is it different from document automation?

OCR (optical character recognition) turns an image of a document into text. Document automation goes further: it recognises the document type, extracts the required fields, checks them against your rules and data, passes unclear cases to a person and sends the approved record to your ERP or accounting system. Unlike traditional zonal OCR, which looks for text at fixed coordinates, Vision-AI understands the document’s context, so it also works with non-standard layouts. Accuracy is measured during the technical validation, on data you approve.

Which ERP systems do you integrate with?

Rivilė, Directo, B1.lt, Xero, QuickBooks and other systems with an API or an import path. Whether a record is written automatically or waits for a person’s approval is set for your process. Your ERP stays the system of record.

Which document formats are supported?

PDFs (standard and scanned), Word/DOCX, Excel, invoices in the email body and photos (JPG/PNG): invoices, orders, CMRs, delivery notes and company forms. Each document type is configured from your examples, and source quality is assessed during the technical validation.

What happens when a field is unclear?

The field is flagged for review, and the responsible employee approves or corrects it, or rejects the document. If required data is missing – for example, the order is not found in the ERP – no record is created until this is resolved. Every action is logged.

How much time does it really save?

We assess the business case for the specific process – how much manual work, error correction and planned hiring the system would take over. Accuracy is measured during the technical validation, on data you approve.

Is the data handling secure?

Yes. We use business-grade APIs that don’t train models on your data. For sensitive processes the system can run on your own or an EU server, and data processing is agreed in the contract.

How much does an OCR automation project cost?

Pricing is provided after the process review and scope assessment. It depends on the process, integrations and implementation scope.

How long does deployment take?

The first working phase is usually planned for 6–10 weeks, and the exact timeline goes into the proposal after the process review. The first version always launches in test mode.

Start with one document flow

In a 30–45 minute process review you show a few typical documents on your screen, and afterwards we assess in writing what is worth automating and where a person needs to check.

Book a process review