In data automation, there has always been a trade-off between speed and accuracy. Even IBM’s Automated Data Preparation touts the ability to toggle between greater speed (useful for getting quick answers) and greater accuracy (useful for getting precise answers) as a feature. However, advances in technology make that dichotomy obsolete. Now, you can have both more speed as well as higher accuracy—the best of both worlds.
Vidado is an innovative pioneer in digital automation for the enterprise and is pleased to introduce Vidado READ, an AI-powered cognitive document automation solution that surpasses humans at reading handwriting. With out-of-the-box accuracy levels of greater than 90% on the most difficult documents, Vidado READ substantially exceeds the levels of accuracy achieved from existing optical character recognition (OCR) solutions. Vidado’s READ engine is a cognitive document automation solution that goes beyond just a “handwriting OCR” and promises to make error-prone, highly inefficient and slow manual paper workflows—common in financial services, healthcare, government, and nonprofits—a thing of the past.
Why it works
Why does Vidado’s READ engine work so well? READ’s machine vision and natural language processing models have been trained using tens of millions of data points collected from text transcription tasks that Vidado has processed over the past five years. These tasks, which number over one billion, represent the world’s largest handwriting training data set, and they enable READ to read handwriting with greater than human accuracy right out of the box.
How does this compare with other AI engines? To put it in perspective, the READ engine was trained with over 10 million images—twice the number of images used by Facebook’s DeepFace which has achieved better-than-human accuracy levels at facial recognition.
Who it is built for
Vidado’s READ engine was purpose-built for high-risk, highly regulated, data-sensitive enterprises such as financial, insurance and healthcare institutions. With accuracy, speed, and cost-efficiency built into the system from the ground up, READ is enabling organizations in these industries to join in on the digital wave transforming the world of business today and do it without ripping and replacing legacy systems which cost millions and takes years to implement. With Vidado, businesses can integrate a cognitive document automation solution for their documents with minimal custom training and business rules, unlike other solutions that require considerable training, the re-engineering of hundreds of manual processes, and development of custom business rules to work seamlessly with existing systems and infrastructures.
How it was built
Designed as a self-improving machine learning engine, Vidado harnessed the power of crowdsourcing to deliver valuable data to customers while simultaneously creating training data. Each successive generation of the READ engine has been improved by feeding data about where its predecessors had weaknesses. Over the last five years, the iterations have enabled Vidado to finally release the current version of the READ engine, which no longer relies on the crowd but uses continuous machine learning and neural network encoding to deliver results that are better than the most highly competent humans.
How well it works
A few data points on speed, accuracy, uptime and straight-through processing will help put things into perspective.
- Accuracy. READ outperforms humans with accuracy levels greater than 90% out of the box on the most difficult documents. For the sake of comparison, human data accuracy levels for single-entry indexing generally fall between 85 and 95 percent. POC results from the largest enterprise in insurance and financial services have only proved these data points time and time again.
- Cost. With READ, paper and digital workflows can be merged, enabling enterprises to eliminate manual workflows with a solution that is quick to deploy and cost-efficient.
- Speed and Throughput. READ is 1,000 times faster than a human and one READ engine can do the work of 80 full-time employees. With average processing turnaround times in minutes, READ far surpasses humans in throughput and speed and is still better than legacy automation systems.
- Scalability. The number of engines working in concert can scale up or down as needed to meet the throughput demands of even the largest enterprise workflows. Higher accuracy and throughput rates and the ability to run around the clock and all year long mean no more paper backlogs or the need to hire temporary staff to deal with seasonal fluctuations.
- Straight-Through Processing. READ delivers up to 50 percent increases in straight-through processing and can help reduce manual staff effort by 75 percent.
Interested in learning more? Take the Vidado Challenge to see whether or not you are able to surpass READ in reading speed and accuracy. You can also request a personalized demo and we’d be happy to have one of our solution architects talk to you about your current cognitive document automation needs.