Why Robotic Process Automation Falls Short for Insurance Claims Processing
October 27, 2020 / Insurance, Intelligent Process Automation, Robotic Process Automation, Use Case
In our last post we discussed how companies often find some automation solutions enable them to address rather simple use cases involving structured data, but hit the wall when it comes to more complex processes. Using robotic process automation for insurance claims is one such instance where that’s often the case.
It’s an issue because many insurance companies are embarking on digital transformation journeys, looking to transform their businesses by digitizing many processes that are now mainly analog and labor-intensive. In many cases, they’ve started down that path with RPA tools and likely got some quick wins.
Limitations of RPA in insurance claim processing
Perhaps they automated processing of some highly structured documents such as online applications or used RPA to do some screen-scraping, both of which do indeed save time for humans. The key to success with RPA, as well as templated approaches to automation, is to work with documents that are highly structured and a process that’s deterministic.
But that’s really the tip of the iceberg when you’re talking about true digital transformation. Consider the insurance claims process, for example. It can involve a slew of different documents and images, including an adjuster’s free-form notes and, depending on the type of claim, estimates from auto body shops or contractors and medical reports. Virtually none of these documents are structured, so RPA and templated automation approaches will be of little to no help.
Related Article: How Intelligent Process Automation Complements RPA: A Use Case
AI is required for unstructured data
Dealing with this unstructured data is where the real opportunity lies in terms of achieving true digital transformation. What’s required is process automation technology that enables companies to scale human decision-making and apply it to use cases that involve hundreds of thousands of complex documents consisting of long-form text, tables, images and more.
Achieving that kind of process automation requires artificial intelligence technology. Many vendors claim to have added AI to their existing product, allowing them to address a broader set of use cases involving semi-structured and unstructured content. But when customers try to implement these products they often find the tools are glorified template and rule engines that cannot adequately address use cases such as claims processing that involve complex documents. (For more on this topic, see our previous post outlining the shortcomings of templated approaches and another on how to identify AI imposters.)
The specific AI technologies required to automate claims processing include natural language processing (NLP), which is the ability for the AI program to read and comprehend a document much like a human would. Another is transfer learning, which is a type of machine learning that enables lessons learned while addressing one problem to be applied to another, different problem.
Effective automation for insurance claims
In the case of automating insurance claims processing, AI technology could be applied to myriad types of documents involved in the process, including images. An intelligent process automation tool would “read” the documents (thanks to its NLP capabilities), perhaps classify them by category and pull out salient data – customer name, account number, nature of claim, dollar amounts and so on – for input into the company’s downstream claims processing system.
Over time, the automated process would become increasingly more effective, because the transfer learning component would enable it to apply knowledge gained to different areas. Maybe initially the program can only identify photographs that show damage to a car, while later on it can differentiate a car from a truck.
The real key to effective insurance claims process automation, though, is to enable the folks on the front lines who actually carry out the process, and know it inside out, to train the process automation tool. Indico’s Intelligent Process Automation platform, for example, is intended to be used by business people; no data scientists are required. Business people create automation models by using the tool to mark up maybe a few dozen documents, teaching it which data points are most important.
From there, the tool can use transfer learning to apply what it knows to other documents it’s never seen before and extract the same key value pairs you’re after. There’s no need to train it on every single variation of a document you may encounter – a huge differentiator from templated approaches and RPA.
IPA is the kind of technology that can bring transformative change to an insurance company, or any company that deals with document-intensive processes. An 85% reduction in process cycle times and 4x increase in process capacity are not at all uncommon outcomes.