POSTED: November 6, 2020
Using machine learning to read and interpret documents
Does your organization still rely heavily on documents supplied by customers for key business processes? Read on to understand how some of the tools for reading and interpreting these documents are continuing to mature.
Change is in the air
We all know that there is an abundance of change right now. If you’ve been in a restaurant lately, you’ll likely have noticed that QR codes have made a resurgence after years of being declared dead. Will we ever go back to paper menus? I’ve had to help more than one parent understand there is no longer a dedicated app required for scanning QR codes with your phone.
The fields of Artificial Intelligence (AI) and Machine Learning (ML) are not immune to this change. The release of GPT-3 over the summer has produced a lot of interest with its ability to generate human-like text.
Though not nearly as big of an announcement, but still interesting, was Google’s announcement last week of its release of Lending DocAI.
There is still a lot of paper in the world of mortgage applications
If you’ve recently purchased a home or refinanced with these historically low interest rates, you’ll know that the mortgage process is often still paper intensive. Lending DocAI is intended to process a borrower’s income and asset documents, in order to decrease the time it takes for a lender to review a loan application.
What Lending DocAI does and why it is different
In April 2019, Google released Document Understanding AI as a “canned” cognitive service in beta. It was originally described as an “API to parse forms or tables from PDF, TIFF, or GIF documents.”
The major cloud providers have long offered “canned” cognitive services related to image, text, classification, etc. Here are examples:
- Amazon Textract
- Microsoft Form Extractor
- Google Vision API
These offerings are vertical independent. They are general tools that can be utilized in a wide variety of applications. However, Google’s Lending DocAI is the first example in the Google Cloud Platform (GCP) of a vertical specific offering.
Lending DocAI offers the same functionality as Document AI but is being marketed directly to the mortgage industry. It utilizes Optical Character Recognition (OCR) and a Form Parser to extract data from the documents and return it in several forms such as key-value pairs or a body of text.
Let’s take a look at Lending DocAI’s capabilities
Here’s a quick example:
We can see it picked out the right name:
It even identified which box I selected:
It did pick up that a signature was present but got the spelling wrong. We can easily blame that on my penmanship!
State-of-the-art data extraction capabilities
This is a good demonstration of the state of the art in extracting information from documents. If your organization spends a lot of time manually extracting information out of documents, you should do some proof-of-concept work to understand if a tool like this could be beneficial. Success and value will depend on factors such as:
- Velocity of incoming documents
- Quality of handwriting, if appropriate
- Number of variations of documents
We’re planning to keep a close eye on Lending DocAI. It is still in beta and that is unlikely to change soon, as Google has a history of keeping services in beta for a long time. If you are interested in learning more, check out the video from Google at the bottom of this web page: https://cloud.google.com/solutions/lending-doc-ai
Want to learn more about how the latest OCR technologies can help your organization? Through a road mapping exercise or a proof of concept, Covail can help show you how to take advantage of these technologies and create Intelligent Operations for your organization. Contact an expert today.