Many businesses use Optical Character Recognition (OCR) solutions for their financial and administrative processes. The technology can convert text images from scanned/photographed documents into characters, using a process called dematerialization.
OCR can help automate invoice processing by aiding data extraction. Without this technology, digitalization is limited, because it still requires manual effort, therefore higher resource costs. I.e. accounts payable could send a digital invoice via email to approvers, but processing would still require manual data entry. OCR solved this problem. While OCR invoice processing has transformed the AP workflow and digitized invoice processing, OCR on its own has limitations.
Why OCR Isn’t Enough
Retrieving data from an invoice is one thing, placing data into the correct fields within an accounts payable system is another, especially when payees and vendors use different formats and software. OCR can extract data but it won’t intuitively know what it means. You need a system that can dependably transfer the data from the invoice to the software system with no oversight.
To resolve that, companies may try to use an OCR provider with high recognition success rates. The truth, however, is that no provider is good enough on its own – the confidence scores come back too low to rely on them for even a semi-automated flow. The only way to solve this is to combine the power of multiple OCR solutions and use them on top of each other to improve the confidence score.
That is exactly what Monite did – we combine many OCR engines in one API. When an invoice comes in, we first use one engine to recognize and post-process data and assess the confidence score. If the score comes back too low, we use other engines one after another until we get to a high confidence score. Through machine learning, we achieve near 100% accuracy.
What is a document confidence score?
What is a document confidence score?
The document confidence score reflects the probability that the key invoice fields – e.g. numbers, addresses, amount due – were correctly captured.
To maximize recognition quality, we also use machine learning-based preprocessing – documents are cleaned up and brightened to remove unnecessary artifacts from the get-go. We also offer an extension with the ‘human check’ for complex cases. If the confidence score is low after all the engines, the invoice can be automatically sent for a manual review.
Monite’s solution automates many tasks that previously required manual processing, including transferring invoice information for payment processing, matching general ledger codes, or sending the invoice to the correct approver.
Monite for Invoice OCR API
Building multiple OCR integrations and other elements in-house is costly and time-consuming. That is why Monite created one API through which you can get the full functionality and value you need without spending a lot of time on development.
