Invoice Processing for Enterprise telecom IT Solutions provider
89%
reduction in exceptions handling
94%
Straight-through processing
98%
Accuracy
Client
Enterprise IT Solutions provider
Goal
Accurately extract key data from telecom invoices

CHALlENGE Overview
Invoices come in different forms depending on vendor. Client is looking for a flexible and scalable AI solution for telecom invoice processing.
Need to automate the repetitive and error-prone manual process of reviewing and correcting hundreds of telecom invoices sent in different formats.
The complexity and frequency of changes to these telecom invoices made traditional RPA template-based automation ineffective.
The client provided us with less than 200 data samples across seven data variations, and asked us to train a robust model to extract the following eight fields:
- Vendor name
- Account number
- Invoice number
- Billing date
- Received date
- Due date
- Amount
- Currency
WHAT WE DID
The model is able to index and classify documents at both the document level and the field level, solving the problems of different providers using different labels for the same category.
- Classify and extract data from invoice documents, routing exceptions to staff. Each manual intervention improves the completeness and confidence of AYR’s machine learning models, which makes the platform smarter and more accurate.
- Model training with 98% accuracy in 4 days to extract key data from telecom invoices.
- Built a model that can extract data from multiple tables on a single page, sum the line item amounts, check results and do the same across all pages of these multi-page invoices.
Impact
- Increased straight through processing from 30% to 94%
- 98% accuracy
- $1 million + in savings
- Dramatic reduction in late fees and credits needed from carriers
- 89% reduction in exceptions handling
- Full integration with finance and accounting platform
- Production-ready in four days.