A longstanding problem that has plagued artificial intelligence projects, including document process automation, is complexity. Essentially, the issue is there’s a disconnect between the business people who understand the process and the data scientists charged with automating it. Trying to translate between the two becomes, to put it politely, challenging.
In many cases, companies are successful with a proof of concept (POC) project involving a modest subset of relatively simple documents. But when it comes time to scale the project, introducing lots more documents of various types, including unstructured content, things tend to break down. The data science teams find it difficult to keep up with requirements, or perhaps even completely understand them. Business process owners know what they want, but struggle to translate those requirements to the data scientists.
It’s nobody’s fault, really. Document process automation involves technologies that are inherently complex, including natural language processing, machine learning and transfer learning. Business people can’t be expected to understand them, yet they are the ones who best understand how their processes work and what aspects of them will benefit most from automation.
Related Article: How Intelligent Process Automation Addresses the AI Data Problem
It’s an issue that screams out for document process automation tools that are simple enough for business people who are the process experts to use on their own, with no help from data scientists or anyone in IT, in order to automate redundant, boring tasks. It’s an idea known as citizen-led development.
One approach to citizen-led development is to use an intelligent process automation (IPA) tool with an extensive knowledge base built in and a user interface (UI) that’s friendly enough for business people to use.
Indico’s IPA platform, for example, includes a base model that consists of more than 500 million data points, which is enough to enable it to understand human language and context. Business users can then customize the model to take on whatever task they’re trying to automate. To date, it’s been used for financial services automation, as well as healthcare and intelligent automation in insurance, to streamline processes involving corporate in-boxes, insurance claims, SEC 10-Q and 10-K forms, and much more.
With the point-and-click UI, users can quickly label the sorts of data they want to pull out of a document, whether dollar amounts, contract language, images – virtually anything. This is far different from the rules-based approach used by robotic process automation and other automation tools that take a templated approach to automation.
Produce working models in hours, not weeks
With IPA, a business process expert need only label a few hundred documents before the tool can build a working production ready model. And the model will continually learn on its own as it sees more documents, becoming better over time – a far cry from templated approaches that require ongoing model maintenance.
And the best part is, it’s the business people who fully understand the process who are using the tool; no data scientists are required.