How Document AI Services are Transforming Enterprise Data Entry Automation

Intelligent document processing (IDP) software enables organizations to collect all kinds of data and make sense of it in meaningful ways, and artificial intelligence is taking this software to even higher levels of advanced automation. Here’s how companies of all kinds can leverage this technology to realize significant productivity gains.

As I think back to my childhood, growing up in the 80s and 90s with cassette tapes, CDs and DVDs, and even a TV with a corded remote control, it blows my mind how much technology has changed over a short span of time, especially now, with multi-model neural modeling and artificial intelligence.

It was 2007 when the first smartphones and devices started to penetrate consumer markets and I, like many other parents and tech geeks, rushed to buy an iPhone. Then in 2010 with the first iPad, we could suddenly leverage smart devices capable of taking notes and synching with cloud services, allowing web forms to replace paper forms with wet signatures.

Within a decade we went from smart devices to generative AI, where neural models can learn, predict, create, and complete tasks previously handled by people. With generative AI, the rate of change has accelerated well beyond the already fast pace we had grown accustomed to. 

AI has changed everything

As a document automation specialist, I have focused on helping businesses leverage AI with intelligent document processing software — this includes big players like Microsoft, Google and Amazon. I have been surprised by how many IT pros don’t know that Microsoft, for instance, offers document AI services.

Microsoft is already on version 3.1 of its AI Document Intelligence data set, which was previously divided into two groups: Cognitive Services and Applied AI. Microsoft recently added neural language expansion to hundreds of different language sets, eliminating the need to configure Azure to the language you need the document processed with — it all happens automatically.

As of July 2023, Microsoft added customized document classification for automated contextual classification and separation of multi-page documents, plus high-density, high-definition OCR, which allows small print to be recognized on large format drawings. When it comes to AI innovations, no company will likely keep pace with Microsoft, or Google and Amazon, in terms of capabilities and speed to market.

Today, any software company that’s limited to embedded, stale OCR technology will struggle to compete in the realm of document intelligence and automation solutions.

Not long ago, document separation always required barcodes, patch codes, and blank page separators. Now with neural modeling, common document types like invoices, receipts, and tax documents can be separated without any manual prepping. Before, documents like HCFA and UB04 medical claims had to be manually separated into batches, but now you can capture multiple different document types and automatically separate them for processing. Data extraction was always zonal, meaning you had to tell the OCR engine certain pixels or x/y coordinates of the data you wanted extracted. Tabular data extraction, or in some instances where tables roll over to following pages for use cases such as invoices, explanation of benefits (EOB), and bills of lading (BOL), required lots of manual human intervention or expensive IDP solutions like Paradatec or Captiva InputAccel made most lower volume customer scenarios unaffordable. Leave it to the big three tech companies to democratize a once expensive, hard-to-acquire technology and make it available to the masses.

Making the most of the opportunity in AI today

Document AI is driving businesses to do more with less, helping to make our lives easier and serve our customers more efficiently. We are still in the infancy of AI, but the technology is already delivering significant improvements to businesses in terms of the automation of data entry, validation, and matching, and exception handling.

For instance, we see federal government customers using Microsoft Azure and Google to perform ID verification to scan, identify, and extract domestic and international IDs and then validate the IDs against web services such as ID.now. Reducing the most expensive cost, human labor, is just the beginning of the revolution in automation. For example, when you enter a McDonald’s today, you are greeted by an ordering portal, a digital display with a full menu, eliminating much of the low-wage hourly workforce. These job roles see a lot of turnover, and the costs of training and retraining are significant. Automated kiosks are now performing the work at a fraction of the cost, and they never call in sick or leave early for an appointment.

Some may say that technology is replacing people and jobs and I agree that’s a concern. But most people don’t really want to work those types of jobs anyway, and technology has simply filled the gaps that businesses are challenged by.

Based on the McDonald’s example, consider the benefits a digital kiosk is providing:

• No handling of cash minimizes errors and risk: People make mistakes, giving back the wrong change or possibly stealing. Automating the transaction speeds up the process without risk of error.
• Staffing, training, and turnover issues related to low-wage workers are eliminated.
• The ability to track consumer purchase data allows AI algorithms to offer new meal options, deliver loyalty rewards, and send notifications directly to users.

The McDonald’s example strays from the theme of document AI, but it serves as a reminder that consumer behavior often aligns with business user behavior. Companies that process paper invoices have been automating that process for years because it involves a high volume of paper and electronic images with lots of variability and layout. It’s also a common process across the world and there is no data standard like we see in healthcare. In addition, the process is labor-intensive, which means a high probability of errors.

Back in 2001, the concept of IDP was new and largely untapped. Products were expensive, difficult to implement and just didn’t automate enough in some cases to justify the spend. Line-item extraction was a problem; it was inaccurate and took longer to adjudicate automated extraction of rows of data than manually entering data. Broadband internet access was hobbled by, at best, T1 speeds (remember when 1.5 MBps was considered fast?) Furthermore, there was no auto separation, meaning humans would still have to separate multi-page invoices with separator sheets, and ultimately the technology wasn’t mature enough to make the impact it has today.

It was the same thing when plasma TVs came out; they were very expensive and the tech was error-prone with image burn issues and fading pictures. It took several years for LED and OLED models to fix the issues of the past and make the economics of buying a $400 70” inch flat screen feasible.

Technology inflection points throughout history drive business efficiency and progress. My 10-year-old self would not have believed that things like generative AI could be possible, let alone part of mainstream culture. But the fact is neural machine learning, the unthinkable genie in the bottle, is now at our fingertips and available to the masses. As a technologist and solution designer, I am thrilled by the tools we have at our disposal to solve real-world business problems. The barriers to automation of the past are quickly being eliminated to the point that almost anything is possible. The limit is only our imagination. Everything has changed.