Bills of Lading processing for tier 1 Commercial Bank
1
Business Analyst
20
Documents
4
days
98.46%
Accuracy
Client
Tier 1 Commercial Bank
Goal
Create a trained model to accurately extract data using a small number of sample documents.
CHALlENGE Overview
All previous attempts to automate the extraction of key data from these semi-structured documents had failed to meet the levels of accuracy required. Our client had only achieved 70% accuracy after four months, five hundred documents and the work of four data scientists
The client provided us with twenty data samples across five data variations and asked us to train a robust model to extract twelve fields.
WHAT WE DID
- One business analyst was trained on the platform
- Three day period to build model, label fields, and train model
- Twenty documents were labeled
Business value
- 98.46% accuracy of extraction achieved in less than ninety minutes training time
- Post-processing and Human-in-the-Loop (HITL) further improved results
- Results achieved in three days far exceeded the results from a competitor's solution, even after four months of training
