Maybe you’ve read the marketing brochure: Vendor XYZ claims “using advanced artificial intelligence, provides Intelligent Document Processing software that is used to automate a lot of different tasks associated with document-based information.”
And if that sounds ambiguous, wait until you see the AI-based software in action!
“Our product has everything you need to solve pretty much any document-related task such as identifying document types and taking unstructured data within a document and turning it into useful, structured, actionable information. And we have NLP too!”
“In fact, using our no-code GUI, our machine-learning software provides over 400 different parameters for configuring image perfection, document classification, separation, data location, and extraction. Our software provides over 50 parameters for recognition alone! And this is for just one recognizer! With all of these functions available to you, you’re only limited by your imagination…”
…And perhaps your inability to possibly understand and learn every single feature so that you can get something into production.
Sounds complex right? And yet, this is the state of the “intelligent capture” (or the intelligent document processing) market: solutions that are focused on delivering more and more “knobs and dials” to address a wide range of use cases. Unfortunately, very few of those solutions enable real focus on what is most important: reliable and precise output.
IDP and Its Value
Intelligent Document Processing is unlike any other enterprise solution in that the entire value hinges on the ability to process as much document-based information as possible at the highest accuracy levels possible. Remember, we’re trying to remove the need for us intelligent humans to do these tedious tasks. To do so, the software has to mimic human intelligence.
As Richard Feynman famously said, “I think I can safely say that nobody understands quantum mechanics.” What Dr. Feynman meant was that (at least at the time and more than likely true today) the study of quantum physics dealt mostly with observing the behaviors of quantum mechanics vs. attempting to understand why or how it behaved in a specific way. In another lecture, he explained the idea of deciding which was correct of two very different theories that produced the exact same outcome. He said that you cannot make a determination with science.
What exactly does all of this have to do with intelligent document processing? Well first, we as users, often have to spend too much time understanding the how and why instead of what is most important, which is the results. Most of us, I can confidently say, don’t really understand Intelligent Document Processing; it’s just too costly to take the time to develop the specialized skills sets.
Solving Automation of Document-oriented Tasks
Second, there are many different ways to solve automation of document-oriented tasks from creating rules manually, to creating systems that build knowledgebases, to using complex deep learning neural networks to learn how to accomplish tasks. And in each case, they may all produce the exact same results. And remember, getting results is all that matters. So why have we been distracted with creating and understanding the mechanisms rather than focusing on exactly what is most important?
Fortunately the answer of how to select between different options that produce the same results is fairly easy in the IDP realm. You select the option that delivers results in the quickest amount of time, with the least cost and with the least disruption to your current processes. Note: you do not select an option based upon features.
Increasingly the option with the most promise is based upon self-configuration, also known as software-defined systems. These systems, rather than relying on manual set-up, can configure and optimize themselves using only input data that describes a particular scenario. For IDP software, that input is tagged sample sets. Using this data, the system can analyze hundreds and even thousands of attributes that may not even be identifiable by humans in order to produce the most reliable software designed to solve your specific tasks. Imagine that: software that not only solves your problems on its own, but in a manner which provides a very customized solution without the typical associated costs.
With IDP software that can define itself, the notion of features quickly becomes antiquated.