Client:
Financial Services Provider
A Financial Services Provider based in Paris has built a platform that provides financial services to their clients. Customers outsource their task of Accounts Payable processing to this client which can help them focus on their core business. Our Client leverages technology and expertise to increase efficiency, which enables their customers to reduce the cost of Accounts Payable processing and the turnaround time of payments.
The Challenge
- Our client provides a range of financial services which generates different types of invoices - card receipts, tax declarations and payment receipts from 100s of its customers. They are a fast growing company with more than 20,000 invoices to process per month.
- The company was processing 100s of different invoice formats which they received as images, these images could include a lot of junk information or have multiple invoices in a page.
- They attempted to enforce a uniform standard of invoices, but that presented friction in vendor onboarding.
Manually processing these documents was no longer an option with their scale and rapid growth. Their team needed a robust platform powered by artificial intelligence that could take on the load of identifying the documents and extracting specific fields from the invoices, if they were to scale without adding additional resources to invoice management.
They tried to work with other OCR solutions such as Amazon Textract and Docparser, but encountered either accuracy issues or lack of flexibility in working with new formats of document
The Solution
Nanonets AI solution uses a three-step workflow to provide a robust solution to their problem:
1. Object Detection - Detect invoices from images with multiple items
Object Detection will detect the presence of 3 invoices in the image below. The object detection model has been re-trained with the clients images and has an accuracy score of ~96%.
2. Image Classification - Classify invoices into different categories ie receipts, declarations etc
Image Classification will identify the different types of documents as present in the image below. The image classification model has been re-trained on the clients documents and has an accuracy score of ~98%.
3. OCR - Extract the data fields relevant to the invoice type
Nanonets OCR AI will capture the required data fields from each document, based on the type of document as detected in the classification model above.
The AI has been pre-trained with 1000s of invoices and re-trained with the client’s invoices to provide a high level of accuracy. Additional checks are set up at each step of the workflow, as a fail-safe mechanism to ensure highest veracity before data is exported to their internal systems.
