An Intelligent Approach to Document Processing

Ahead of Intelligent Automation Week Chicago, HyperScience’s Chief Customer Officer, Tim Kalimov, wrote a post on intelligent document processing. Check it out below.

Documents are messy and vary in layout, quality and complexity. They can contain handwriting or cursive, can be faxed, mailed into a central mailroom, or photographed by a smartphone and uploaded to a portal. They are often low resolution, skewed images with blank pages or crossed out lines, which makes the process of identifying that document, locating and extracting the relevant information, and getting it into the correct system highly manual.

This is bad for customers and bad for businesses. Whether it’s adding a beneficiary to an account, processing invoices, or accelerating claims payouts, people and businesses expect results faster than ever before. Simply put, manual entry and outdated, legacy data capture products can’t compete, leading to inaccurate data, slow processing times, missed SLAs and poor customer experience.

Where Automation Can Help

Put another way, the information needed to efficiently and accurately process the myriad of documents that exist today is not fully contained within those documents. Additional context and comprehension are required, which is part of the reason that enterprises have continued to rely on manual data entry operations for so long.

Financial services, healthcare, insurance and government organizations are dealing with high-value transactions, and faced with the trade-off between accuracy and automation, they have relied on humans, with their context and understanding/experience of the world, to manually extract data from the plethora of incoming documents.

Fortunately, technologies are available today to narrow this chasm. Advances in Machine Learning are bridging the gap between human understanding and machine processing, transforming business operations in the process.

Whereas older technologies rely on explicit rules, the beauty of Machine Learning is that it trains on real-world data and continues to learn and adjust itself in response to the data it’s exposed to. Machine Learning has unlocked capabilities that weren’t possible before, taking us from a place where we couldn’t possibly write software to reflect the world – including the diverse document types and text inputs it contains – to one that can train and teach itself. When it comes to processing the vast volume of documents generated by and moving between businesses and customers, Machine Learning (when done correctly – more on that later) is the ideal solution.

By training on a proprietary dataset that is representative of the real world, HyperScience, for example, is able to read handwritten and machine-printed text. In addition, by definition, Machine Learning solutions get better over time, incorporating humans-in-the-loop to review and resolve fields it’s not sure about, which improve the underlying models.

Investing in Software: What to Consider

That doesn’t mean that all solutions are created equal, however. With the rise of ML, AI and IA, it can be difficult to navigate the evolving marketplace and select a solution that adds value at your organization. Achieving intelligent automation requires a thoughtful, strategic approach, and there are several factors that must be considered when evaluating enterprise software solutions.

Here are two key factors to keep in mind as you navigate the automation landscape:

  • Recognize that people will be involved in the process.

People aren’t perfect, and neither are machines. To solve challenging, real-world problems (such as document processing and data extraction), machines need supervision. Once you accept that there’s no silver bullet automation solution, the key is to investigate how a solution involves humans in the loop. Invest in a system that knows when it might make a mistake and when to ask for help and send exception cases to data entry teams to review. Make sure that the work it asks a person to do is minimal to maintain the highest level of automation possible. Even better if a solution has built-in tools or reporting to help you manage how your people are doing (e.g. operational and performance metrics).

  • Prioritize easy-to-use but beware a holy grail of false promises.

It used to be that enterprise software required significant developer resources to set up, implement and maintain. Even today, legacy data capture products can take months and hundreds of hours of development to get new use cases up-and-running, which makes bringing on new business lines challenging and can negatively impact revenue growth. It’s important to note that enterprise software and easy-to-use are no longer mutually exclusive. Ask what is required to deploy the solution and get it fully operational. Is there an API? What kind of hardware is required? How will it fit within existing workflows, and once it’s live, how challenging is it for business users to use? Look for modern-day simple solutions but be skeptical of anyone that promises results without setup. Even for the most advanced enterprise AI solutions out there, upfront setup is still required.

By leveraging the latest technologies, organizations can streamline complex document processing workflows and extract valuable information that can be fed into downstream systems to drive better business outcomes. We look forward to meeting you in Chicago to discuss how intelligent automation solutions can decrease costs, improve response times, enhance customer service and drive revenue at your organization.

Credits

Copy: Tim Kalimov, Annie Christian

Design: James Rivas

Back to Articles