Insurance companies have long sought to automate the underwriting processes involved in issuing property and casualty policies, and with good reason: effective underwriting can mean the difference between profit and loss.
Looking at data covering three decades, researchers at McKinsey found operating results had the greatest impact on a commercial property and casualty insurer’s overall financial performance. “And within operating results, loss ratio generates much more variability than expense: when comparing top- and bottom-quintile performers in both the United States and the United Kingdom, loss ratio varies by up to 28 percentage points, whereas expenses vary by just 2 to 4 points,” McKinsey found.
A good loss ratio, of course, is an indicator of sound underwriting practices and quality risk assessments. Today, effective risk assessments requires using modern tools and technologies such as artificial intelligence to analyze as much data as you can gather on a particular market segment.
Getting to data-driven underwriting
But, as the McKinsey report notes, many commercial insurance companies, especially, are hindered by legacy technology. “Based on our observations, anywhere from 30 to 40 percent of underwriting’s time is spent on administrative tasks, such as rekeying data or manually executing analyses,” the report says.
With technology evolving rapidly, McKinsey sees a brighter future ahead, with intelligent automation in insurance. Application programming interfaces (APIs) will enable connections to all manner of data and tools, ultimately resulting in a “fully digital workflow” where underwriters have easy access to data needed to make sound risk assessments. Getting there is not without its challenges, however.
“One challenge has been the translation of business requirements into practical technical specifications that are fit for purpose and accommodate the complexity inherent in commercial underwriting,” McKinsey says. “Another challenge has been sourcing the required talent and capabilities to fuel transformation.”
Intelligent automation for insurance underwriting
Translating business requirements into practical technical specifications is indeed a challenge. It’s why many AI projects fail to deliver, because things get lost in the translation from the business people on the front lines who understand how their processes work and the data scientists who try to automate them. And the data science talent shortage is a topic we’ve written about in the past.
The solution to both issues is an insurance process automation solution that can be implemented by the business folks who actually perform the insurance underwriting process, and therefore understand it inside and out. These are the people who know where those administrative tasks lie, the ones that are chewing up 30 to 40 percent of their time – the sort of processes that are ripe for automation.
Another key is to have an automation solution that can handle the varied document types that may be involved in a property and casualty underwriting process, including property assessments, credit reports, Dun & Bradstreet reports, motor vehicle records and more. It’s unlikely a robotic process automation (RPA) or templated tool using optical character recognition (OCR) will be able to deal with the variation inherent in these documents. What’s required is a tool that can deal with unstructured content, meaning documents that don’t necessarily follow any given pattern or structure.
Indico creates citizen data scientists
Indico’s Intelligent Process Automation platform delivers on all of these fronts. It’s based on a database of some 500,000 labeled data points which, along with natural language processing capabilities, enables the platform to understand the context of documents and “read” them much like a human would. Then it’s a simple matter for the tool to extract required data and input it into a downstream platform, such as one that insurance underwriters use to assess risk.
The Indico intelligent automation platform is also intended to be used by business people, turning them into “citizen data scientists.” They use the platform to train models by labeling actual documents involved in the underwriting process, denoting the fields that should be extracted. Labeling just a few dozen documents will produce a model that performs with extremely high accuracy.
Such tools will remove many of the mundane tasks involved in underwriting, leaving more time for the analysis that is where the real value – and insurance company profit – comes from.
This article has been published from the source link without modifications to the text. Only the headline has been changed.
