aiprocesai.lt

Case study · Wholesale · OCR automation

Invoice OCR automation into Rivilė: 90% less manual work

A wholesale company processing ~200 supplier invoices per month cut bookkeeper time by 90% and brought entry errors close to zero — by integrating AI OCR with their Rivilė accounting system in 10 business days.

Client situation

A wholesale company receives ~200 supplier invoices per month from 40+ vendors. Each invoice arrives by email as a PDF attachment and used to require manual data entry into the Rivilė accounting system.

On average a bookkeeper spent 4–5 minutes per invoice: open the PDF, read the data, find the supplier in Rivilė, enter the total, VAT, date, and invoice number. That adds up to ~15 hours per month on data entry alone — before any error correction time.

Solution: AI OCR + Rivilė API + Make

We built an automated document-processing pipeline that takes an invoice from email arrival to a Rivilė record in 30–60 seconds without human involvement for standard cases. The integration relies on our workflow automation services for the orchestration layer.

1
PDF intake
invoice arrives by email at a dedicated address
2
AI OCR extraction
Azure Document Intelligence structures the data
3
Confidence gating
>95% auto · below — human review
4
Rivilė API write
purchase document created with all fields
5
Slack confirmation
bookkeeper one-click approval

Average time per invoice: < 60 seconds from intake to Rivilė record for standard cases.

What we automated, step by step

1
PDF attachment intake

Make pulls the invoice PDF attachment from a dedicated mailbox and passes it down the pipeline.

2
AI data extraction

Azure Document Intelligence extracts supplier, total, VAT, date, invoice number, and line items.

3
Confidence gating

Confidence above 95% auto-progresses; below the threshold the document is routed to a human for review.

4
Rivilė document creation

REST API automatically creates a purchase document with all fields and the supplier mapping.

5
Audit log + Slack

The bookkeeper receives a Slack notification with a link; a one-click approval finalises the entry.

Before and after the automation

AspectBeforeAfter
Manual data entry~15 hr/month~1.5 hr/month
Entry errors2–3 errors/monthpractically zero
Processing time per invoice4–5 minutes<60 seconds

Security and GDPR

Financial documents require the highest level of care. On this project we used:

  • Azure Document Intelligence (EU data centre, not public models) — data is not used for AI training
  • Invoices transmitted over TLS 1.2+ encrypted channels
  • PDF files automatically deleted 24 hours after processing
  • API access restricted with an IP allowlist
  • NDA signed with the client before access to financial documents
  • Audit log: every invoice tracked with timestamps and processing status

Technical architecture

The integration uses Make (Integromat) as the orchestration layer. Every step is auditable in the Make execution log.

ComponentPurposeAlternatives
Azure Document IntelligenceAI OCR + structured data extractionGoogle Document AI, AWS Textract
Make (Integromat)Orchestration + email trigger + Rivilė API callZapier, n8n
Rivilė REST APIPurchase document creation in the accounting systemDirecto, B1.lt, MS Dynamics
Email gatewayPDF attachment intake from suppliersWebhook, dedicated mailbox
SlackBookkeeper approval notificationsTeams, MS Outlook

Does a similar solution fit you?

Fits when…
You receive 30+ supplier invoices per month in PDF format
A bookkeeper spends 5+ hours/month on manual data entry
You use Rivilė, Directo, B1.lt, or another system with API access
Invoices are structured (PDF / JPEG, not handwritten)
You want to audit and measure accuracy on your real invoices
Better to wait when…
The ERP is still being deployed and has no API access
Invoice volume is low (<20/month) — ROI is insufficient
All invoices arrive in non-PDF formats (fax, handwritten)
Processes are still undefined — standardise first, then automate

Frequently asked questions

Which document formats does the AI OCR support?
PDF (single-page and multi-page), JPEG, PNG, and Word. For structured invoices, accuracy is typically 95–99%.
Does this work with ERPs other than Rivilė?
Yes — Directo, Microsoft Dynamics NAV/BC, and any system with a REST API or CSV import. Rivilė was used here because it was the client's existing system.
Is AI OCR safe for financial documents?
Yes — we use business APIs only (not public models), data is never used for AI training, an NDA is available, and invoices are deleted after processing.

Want a similar OCR pilot?

Standard OCR Pilot — from €500: 1 document type, 1 ERP, 5–10 business days. For the pilot we measure accuracy on your real invoices.

No commitment·Reply within 24 hours·GDPR compliant
📊Calculate ROI