Data entry ops · Production

Nanonets automates passport and document data extraction for Dutch aviation HR firm

The problem

A Dutch aviation staffing company employed 60 data keyers to manually process 10,000 documents per month—passports, flight licenses, and resumes—at high cost, with low accuracy and significant inefficiency, compounded by varied image orientations, low-resolution scans, and multi-language content.

First attempt

Traditional OCR tools including Amazon Textract, Abby, and Google Vision were tried but proved insufficient: they required extensive pre- and post-processing, could not handle multi-language documents or low-resolution images simultaneously, and did not support the client's custom data requirements.

Workflow diagram · grounded in source
1
Document upload trigger
Trigger
The user uploads a PDF or image file to initiate the automated pipeline.
▾ source quote
“the holistic solution allowed the user to upload a PDF / image file”
2
Orientation detection and correction
Ai action
A deep learning architecture identifies the orientation of the image and automatically corrects it.
▾ source quote
“Our deep learning team built additional architectures to identify the orientation of the image and to correct it”
3
Page and area identification
Ai action
The system identifies the page of interest from numerous passport pages and captures relevant information from the area of interest.
▾ source quote
“We also built a crop feature to identify the page of interest from numerous passport pages and then capture the relevant information from the area of interest”
4
Multi-model OCR extraction
Ai action
A combination of multiple machine learning models is deployed in real time to extract data from the document.
▾ source quote
“a combination of multiple machine learning models were deployed in real time to tackle the challenges in the large variety of data”
5
Database integration
Integration
The extracted data is integrated seamlessly with the client's existing database.
▾ source quote
“Can be integrated seamlessly with their existing database”
6
Error-driven model retraining
Feedback loop
Data where models made errors is re-collected and new models are rebuilt and redeployed so the system learns over time.
▾ source quote
“We re-collected the data where the models made an error and built model building processes to redeploy the new models. The system learns over time and accuracy of the system keeps improving”
Reported outcome

The Nanonets OCR system fully automated document data entry with no human effort required, and the system learns over time so accuracy keeps improving.

Reported metrics
Manual data keyers employed60
Monthly documents processed manually10,000 documents a month
Human effort required after automationNo human effort was required
System accuracy trajectoryaccuracy of the system keeps improving
Reported stack
Optical Character Recognition (OCR)docker
◆ Does this fit your context?

Compare to your context

Tell us your scale, team, and constraints. We'll show what changes at your size, what fails at your scale, and whether this case is a fit, needs adaptation, or won't scale to you. Free demo, no signup.

Compare to your context →
~30 seconds · free
Source
https://nanonets.com/customer-success-story/ocr-passport
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The Nanonets OCR system fully automated document data entry with no human effort required, and the system learns over time so accuracy keeps improving.

What tools did this team use?

Optical Character Recognition (OCR), docker.

What results were reported?

Manual data keyers employed: 60; Monthly documents processed manually: 10,000 documents a month; Human effort required after automation: No human effort was required; System accuracy trajectory: accuracy of the system keeps improving (source-reported, not independently verified).

What failed first in this deployment?

Traditional OCR tools including Amazon Textract, Abby, and Google Vision were tried but proved insufficient: they required extensive pre- and post-processing, could not handle multi-language documents or low-resolutio…

How is this data entry ops AI workflow structured?

Document upload trigger → Orientation detection and correction → Page and area identification → Multi-model OCR extraction → Database integration → Error-driven model retraining.

Related data entry ops cases

WHAT TO DO WITH THIS

Now compare it to your context

This case is one data point. Whether its pattern fits you depends on your volumes, your stack, and your exception load — that comparison is the step no case study can do for you.

Browse the Data entry ops cases →