Automated Invoice Processing
Banking & Financial ServicesBusiness Impacts
Increased Efficiency
20%
Improvement in Model Accuracy
Reduced Time & Effort
Customer Key Facts
- Location : United Kingdom
- Industry : Accounting
Problem Context
The customer, one of the largest professional services firms in the world, used a third-party tool for processing customer invoices. It only worked on previously-seen invoice templates and provided very low accuracy of ~50 percent. They wanted to build a template-independent solution for efficient invoice processing.
Challenges
- Accurately extracting entities from completely unseen template invoices is challenging
- A huge quantity of labeled data is needed to get good results
Technologies Used
AWS Comprehend
AWS Textract
AWS S3
AWS S3 Glacier
TensorFlow
Convolutional Neural Networks
Siamese Networks
Custom Named Entity Recognition Model
Developing an Automated Template-free Invoice Processing Solution For a Large Accounting Firm
Solution
Leveraging best-in-class deep learning and OCR techniques, Quantiphi developed a template-independent invoice processing solution that achieved 20% improvement in accuracy as compared to the customer’s existing model. The solution helped the customer minimize time and effort, which in turn improved cost effectiveness, efficiency gain, and ease of operation with great accuracy and lower error rates.
Result
- Automated invoice processing via a deep learning based template-independent OCR solution
- High model accuracy in relevant entity extraction
