In our this article of the Intelligent Capture Stack series, we take on a topic that is often neglected when it comes to evaluating technology: ease of use. More than just the end user experience, tackling ease of use unleashes document automation for every organization as long as capability isn’t the trade-off for simplicity.
The end user experience — when it comes to intelligent capture — typically refers to the person reviewing the output data and ensuring it is correct. The ultimate goal, however, is to remove that “experience” as much as possible so the focus must shift to another critical end user aspect. The other end user whose experience is often overlooked, due to focus on features and functionality, is the person configuring and optimizing a system.
Most of the time, instead of evaluating how easy it is to configure a document classification project, many organizations focus on only the enabling capabilities. Quantifiable evaluations such as presence of automated classification and the ability to separate documents without use of barcodes or separator sheets, etc. Rarely is there the ability to actually observe and rank systems by how easy they are to configure without a significant investment.
This focus on features over ease-of-use has led to some pretty relaxed focus of vendors on the same topic. The result, in most cases, is a GUI that is the equivalent of a 777 cockpit.
Achieving Capability and Simplicity
Where vendors have more recently claimed ease of use is on the availability of pre-trained capabilities (or “skills” in Robotic Process Automation parlance) that can classify and extract data on an invoice or bank statement for instance. While pre-trained capabilities do indeed simplify the configuration process, they often do so at the expense of overall capability.
For instance, the pre-built invoice skill might do really well on invoices of a certain type and for specific data fields such as invoice number, date and total. However, it is a rare case where an organization can use something strictly out-of-the-box. The result is either an organization stops short of satisfying their real need or they end up engaging in expensive and time-consuming configuration to achieve the needed capabilities. If PO Number is needed and not already supported, they must use the complex GUI of solutions to build and optimize this capability.
The current approach in the world of intelligent capture is to offer one of two paths: take pre-built skills as-is or engage in more expensive customization. This shouldn’t be.
How to Avoid the Trade-off
The real promise of machine learning is to remove work. This “work” should be defined as both pre-production as well as production work. Time and money expended to configure and refine a system should also be in the scope of machine learning — not just document processing work. Staff shouldn’t need to learn the software. The software should be able to learn their requirements.
The ultimate goal is to use machine learning so that any organization can realize their own out-of-the-box capabilities or take existing skills and easily modify them, training the system to add new capabilities. The result is that systems can be configured in a matter of hours or days to deliver reliable, high-performance automation without the need to oversee a complex configuration effort.
In a world where machine learning is applied to the entire user experience, organizations can literally have their cake and eat it, too.
Past articles focused on the Intelligent Capture Stack may be of interest to you as well as these useful eBooks on What Is Accuracy? and Data Science with Advanced Capture.
