Data entry ops · Production

Nanonets AI extracts data from 140,000+ handwritten historical documents with 95% accuracy for SciencePo researcher

The problem

Florian Oswald needed to extract land usage and value data from over 140,000 handwritten historical documents with non-standard, intersecting table formats. Traditional OCR tools could not handle such formats, and manual processing would have taken months.

First attempt

Traditional OCR tools were unable to capture data from the non-standard table format where rows and columns intersected.

Workflow diagram · grounded in source
1
Large-scale document extraction needed
trigger
“He had to extract data like land usage and value from 140,000+ hand-written documents”
2
AI extracts from unstructured docs
ai_action
“Nanonets AI has been trained with millions of documents and can extract data from unstructured document types. With some training, the AI could differentiate between intersecting rows and columns.”
3
Post-processing rules validate output
validation
“With Automated Workflows, they could place post-processing rules to identify errors and reduce validation time”
4
High-accuracy data delivered
output
“I was able to extract data with over 95% accuracy in just 2 hours”
Reported outcome

The researcher extracted data with over 95% accuracy in just 2 hours using Nanonets AI, and was able to get started within a day.

Reported metrics
Extraction accuracyover 95%
Time to extract data2 hours
Documents to process140,000+
Time to get startedin a day
Show all 5 reported metrics
extraction accuracyover 95%
time to extract data2 hours
documents to process140,000+
time to get startedin a day
manual processing time (avoided)months to process manually
Reported stack
Nanonets AIAutomated Workflows
Source
https://nanonets.com/customer-success-story/extracting-handwritten-documents-using-nanonets
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The researcher extracted data with over 95% accuracy in just 2 hours using Nanonets AI, and was able to get started within a day.

What tools did this team use?

Nanonets AI, Automated Workflows.

What results were reported?

Extraction accuracy: over 95%; Time to extract data: 2 hours; Documents to process: 140,000+; Time to get started: in a day (source-reported, not independently verified).

What failed first in this deployment?

Traditional OCR tools were unable to capture data from the non-standard table format where rows and columns intersected.

How is this data entry ops AI workflow structured?

Large-scale document extraction needed → AI extracts from unstructured docs → Post-processing rules validate output → High-accuracy data delivered.

Related data entry ops cases