Last week I started a conversation on what template-less means and where it came from. There’s a lot of hype around IDP and the ability to address document automation without requiring the creation of templates that dictate exactly where needed data is found on a page. The reality is that templates if applied correctly, provide the best level of performance because of all of the guesswork regarding locating data. Instead of using error-prone keywords or other rules, you instruct the software exactly where the data resides. The real problem is the amount of effort it takes to use a template. For instance, if I want to reliably locate and extract data on highly variable and complex health remittances (otherwise known as EOP or EOBs), I would simply tell the software where to find the information. But to do that, I would need to create a template for every variation and that is time-consuming to do; I would potentially need to create and maintain hundreds of templates. And even after all that effort, there are situations that I cannot really account for. Perhaps a template for one health insurer only accounts for a single-page remittance but then a multi-page remittance pops up. My template won’t work on that one.
So there is nothing inherently wrong with templates, they just take a lot of effort. But a template-less approach won’t significantly improve things. Using a rules-based approach, it might take less time to construct a set of rules for EOP/EOB extraction, but what I can in time saved, I lose in an increase in errors and complexity. And let’s get real, to construct a proper set of rules takes a significant amount of time to analyze all of the varieties of EOPs and this rarely gets done. So an IDP solution that claims to be “template-less” isn’t giving organizations a significant improvement; only a trade-off. That sucks.
The real solution isn’t pitching template vs. template-less approaches. Rather, it is through the application of machine learning that can analyze a large volume of data and select the right approach for the problems to solve. With machine learning, it doesn’t really matter if a template is used or some more complex set of logic. The benefit is that the algorithms do all of the heavy lifting both for the initial configuration and the ongoing upkeep. In this scenario, the real objective is to get as close to the precision of templates as possible. After all, the real goal of IDP is to get the maximum amount of information out of your documents at the highest level of accuracy with the lowest amount of effort.
At Parascript, we have invested significantly in a system we call Smart Learning that takes all the guesswork and complexity out of deploying an optimized IDP solution. After analysis of data, if it discovers that the best approach is through automated creation of a template, it will do it. If a problem requires more sophisticated models that require data analysis, it will do that. And if the answer lies somewhere in-between, you guessed it, Smart Learning will figure it out.
Here is our Document Processing Fundamentals eBook to get you started!