Are your mailroom processes automated but still not getting the desired results? Learn why deep neural networks power the new recognition engines for Digital Mailroom 2.0.
I'd like to take you behind the scenes of one of my recent projects - it's an interesting case, as it highlights the evolution from traditional OCR/ICR technology which powers automated document processing in many mailrooms to the next-gen technology of AI and deep neural networks.
What’s the situation? My client was searching for a better solution for bulk document processing for invoice bundles. An operations team is responsible for managing deliveries and processing packages of incoming invoices. Invoice packages consist of multiple document types: this includes invoices and supporting documentation, including bills of landing, proof of delivery and rate confirmation. The process requires each supporting document to be verified before the invoice can be validated and approved.
Their main problems? Handling and document preparation costs were high, as documents needed to be separated and sorted before scanning. On top of that they had a high cost of data quality control which was managed manually. And traditional OCR/ICR technology wasn't giving them the high level of recognition, processing speed and data quality they needed - too much manual intervention was required for classification, validation and exception handling, reducing operational efficiency. Time to manage invoice packages took too long and cost too much.
Our challenge? To prove that machine learning technologies using deep neural networks could improve data quality and increase efficiency for bulk invoice processing - without adding cost or complexity.
The docBrain platform - powered by deep neural network technology - is the engine underlying the solution. docBrain's platform develops and visualizes data sets: docBrain then trains neural network models with the data sets it creates, composing the foundation for processing the invoice packages as follows:
Automatic Document Reading: the neural network eliminates ALL document preparation - including manual sorting and the separator sheets (usually printed with barcodes) which separate document bundles.
Achieving Automatic Validation: In this part of the process, a second (or several) neural network extracts (reads) data from the different document types.
Validate the invoice packages: Finally based on the outcomes above the role of the neural networks are to validate (or reject) these invoice package documents based on business rules applied to the data provided by the previous process.
With traditional OCR technology, high levels of accuracy for data validation typically involves a lot of programming and custom coding - making it impractical to implement for many processes, as the costs and time outweighs the benefit of high data quality. docBrain's technology is a collection of neural networks to improve performance over traditional OCR technology in each step of the process - from sorting and document classification to data extraction and validation.
How do we verify the quality and how do we improve the results? The platform performs constant quality and assurance measures, conducting spot checks for errors and anomalies and validation of data quality and accuracy. Errors can be minimized to the nth degree: process owners establish thresholds to determine which level of accuracy they want for specific steps of the process. Production dashboards provide views on processing statistics, page accuracy and cost-per page information.
What benefits does this new process deliver?
- There is virtually NO manual set-up or sorting required for scanning invoice packages - bundles can be scanned as-is and the neural networks automatically classify documents.
- It eliminates non-value work
- Processing costs aim to be reduced by minimum 40% due to high levels of data accuracy (up to 99.99%), better data quality, less manual intervention within the process and increased overall efficiency
Now thanks to deep neural networks, straight-through document processing has truly become a reality.
Interested to know more about how deep neural network technology can boost efficiency for document processing? Contact us to schedule a workshop, request a demo or speak with one the members of the project team.