I have a scenario where data needs to be extracted from dynamic documents such as passports and government-issued IDs from different countries. Please advise on the most suitable solution for this use case.
Additionally, I would like your opinion on using Azure OCR and Azure OpenAI for this requirement. One limitation I observed with Azure OCR is that when documents are unclear or of poor quality, the service does not notify us or provide a confidence score to indicate extraction reliability.
Recommended UiPath solution for passports & government IDs:
1.UiPath Document Understanding
Use Intelligent OCR (Google / Microsoft / OmniPage)
Use Pretrained ML Extractor or ML Models for IDs
2. Image quality checks
Validate resolution, blur, and completeness before extraction
3. Validation rules
MRZ checksum, date logic, field length, country rules
4. Confidence handling
Use field confidence scores from DU
Set thresholds → auto-approve vs manual review
5. Human-in-the-Loop
Route low-confidence cases to Action Center
I think the suitable option is Document Understanding using the ID/Passport prebuilt model or a custom ML Extractor because it handles variable layouts and provides confidence scores with validation.
Azure OCR and Azure OpenAI can support extraction, but Azure OCR does not give low‑quality warnings or confidence scoring, so reliability checks are limited compared to UiPath DU.
For more:
OR
Happy Automation
What is the pricing of Document Understanding to check the feasibility. Or from where i will get the pricing details
@rizvana.mohammed
UiPath does not publish fixed Document Understanding pricing publicly; it depends on the licensing model so
I suggest Pls check with UiPath Sales team for exact pricing:
Document Understanding - Metering and charging logic (Unified Pricing)
If helpful, mark as solution. Happy automation with UiPath
For passports and government IDs from multiple countries, a dedicated ID/document OCR solution is the best fit. These tools are designed for dynamic layouts and usually provide field-level confidence and validation.
Azure OCR is good for basic text extraction but not ideal for structured ID data and it doesn’t clearly flag poor-quality documents. Using Azure OCR with Azure OpenAI can help interpret or format extracted text, but it won’t solve accuracy or confidence issues. You’d still need to build your own quality checks.
use a specialized ID OCR engine; Azure OCR + OpenAI works only as a supporting or fallback option.
Does it mean like for each new format we need to train the ML model, to extract name and other details
No, you don’t need to train a model for every new format. Most ID/document OCR engines handle multiple formats out of the box. Training is only needed for unusual or unsupported document types.
Okay. So Do you think that Document understanding is the most reliable solution
How does this Validation Station (Human-in-the-loop) stage work in unattended runs. Do we need to validate each document or just one’s with less confidence score
Only one time you have to validate and verify , after that bot will train and work independently for all.
Yes, Document Understanding can be a reliable solution, especially if you combine it with a dedicated OCR engine for IDs and passports. It’s flexible it lets you handle multiple document types, apply pre-trained ML models, and implement validation rules.
This topic was automatically closed 3 days after the last reply. New replies are no longer allowed.