Unleashing the Power of Unstructured Submission Data

Artificial Intelligence - September 22 2021

The lack of data standardization across commercial insurance documents is an issue that costs the industry an enormous amount of time, effort, and money, and adds friction to underwriting and placement processes. Forrester states that approximately 85% of enterprise data is not structured. This means that valuable data is locked away in emails, tables, document attachments, policies, quotes, binders, etc. The difficulty of harnessing unstructured data in submission documents is a major challenge as the industry looks to make smarter use of data in commercial lines underwriting. Today, the submission intake process lacks automation and efficiency as human knowledge workers are bogged down manually harvesting data from piles of paperwork, broker emails, spreadsheets, and PDF files with no standard structure or format. Today, carriers often take a “first in, first out” approach to processing submissions in their overflowing submission queue, rather than focusing on the best submissions to quote.

“Data is the lifeblood of insurance companies. Insurance companies compete on analytics and algorithms, and they compete on experiences. This is our equivalent of that physical capital that Tesla has to invest in. Part of it is making an appropriate investment, but part of it is also recognizing that we were capturing a lot of information before, but we were using only a fraction of it,” explained Rob Galbraith, author of The End of Insurance As We Know It during our recent webinar titled 2021 Commercial lines Underwriting Priorities and Trends. “If you think about an 80-20 rule, we were probably capturing 80% of the data that we really needed but only using about 20% of it. Particularly, unstructured data, adjusters’ notes, inspection notes, images, videos, typed notes or handwritten notes.”

“The data sources out there just continue to explode, and they’re morphing every single day. The issue around data and data integrity is something that requires an infrastructure. We need to build an infrastructure that actually supports the right underwriting inputs to that process – how do you package that data and how do you refresh it?” stated Art Borden, VP – Product Services, CNA insurance. “It is an ongoing challenge. You have data scientists who want to play, you have underwriting talent that want to play, you have operations folks who are clearly interested in not having to rekey information if they can avoid it. What you build to manage that data pipeline is changing dramatically.”

According to projections from Gartner, white collar workers spend 30–40% of their time searching, finding, and assessing unstructured data. In the case of mid-market and enterprise commercial lines insurance, underwriting teams and administrators spend countless hours combing through complex, lengthy documents to manually mine key data points needed to make risk, pricing, and placement decisions. Submissions from brokers are often submitted on paper forms or in PDF documents for underwriters to read and harvest the information needed and then manually rekey the data into multiple transaction systems. This manual processing increases the likelihood for human error, is costly, time consuming, and quite often prevents the insurer from responding to the application in a timely manner.

A recent InsTech London report identified data extraction and ingestion as one of the major themes that will be driving change in insurance in the next decade. The report takes a detailed look at the organizations working to overcome the frictional cost of unstructured insurance data and manual data extraction and organization. 

“Carriers need to adopt technology solutions that both provide solutions for underwriters’ most pressing current concerns and prepare them for a future in which AI and machine learning play an increasingly larger role,” writes Rob Whitton, Vice President of Business Development, Decision Research Corporation in the Commercial Lines Underwriting Priorities eBook. 

According to a new study conducted by Celent and commissioned by Equisoft, insurers said they gain the most value from using data to build actuarial pricing models, automate underwriting decisions, and assess customer satisfaction.

AI Converts Unstructured Data to Meaningful Data for Underwriters

AI helps insurers escape the operational bottlenecks associated with manually extracting and ingesting data during the submission intake process. Machine learning and natural language understanding technology automatically mines semi-structured and unstructured text in a variety of data sources to extract key information required to assess the risk, eliminating the need for underwriters and administrators to manually sift through documents. AI solutions purpose-built for insurance can be trained to extract and ingest data from a variety of digital insurance documents such as submission emails and attachments, new business applications, forms, endorsements, broker presentations, loss run reports, statements of value, policies, binders, quotes, and more. During the submission process unstructured data is extracted into useable formats such as XML and JSON which can be easily consumed by downstream systems.

AI solutions can learn the structure of complex insurance documents just like a human would and extract data from the documents, dramatically reducing the time it takes to access key data needed to process submissions. Carriers today are overwhelmed by the sheer volume of email submissions; often they are only able to quote less than 50% of the submissions received. The submission process today is also riddled with human error as underwriters and administrators manually rekey data into clearance and registration systems, CRM, rating engines and other downstream systems. A manual underwriting process with a single data error results in an estimated $6.5 billion in annual premium leakage in the first year and over four years total leakage exceeds $ 22.3 billion. A simple misclassification can be extremely costly. “Underwriters spend an enormous amount of time digging up data or – worse – asking an agent or the insured to give them the data. It’s tremendously inefficient and unnecessary,” explains Bob Frady, CEO and Co-Founder, HazardHub, now part of Guidewire, during a recent interview

Leveraging AI and machine learning alleviates the burden of administrative tasks, and reduces the risk of human error by eliminating the need to manually rekey data into a myriad of systems. The result is that underwriters are empowered by technology to make swifter and smarter underwriting decisions, quantifying risk and accelerating speed to quote. “A human can only look at so many applications or inspection reports throughout the day and cannot necessarily aggregate all those same reports that hundreds of underwriters across the organization are also looking at to find some hidden patterns and factors. Those are the possibilities that are open today. It takes a different mindset re-evaluating where we are from a technology standpoint, from a data standpoint, and just rethinking our process with a data lens to start as a priority,” adds Rob Galbraith.

“Now, if we think about AI solutions designed for insurance, which is a very document-intensive industry, we can design a system that gets more useful with the number and variety of documents that are fed into the system,” explains Colin Toal, Chief Technology Officer, Chisel AI while discussing The Powers of the AI Learning Effect. “The more policy documents or submissions or binders or quotes that are fed into the system, the more the system becomes familiar with the document format and data. The system is then able to extract, understand and summarize the data patterns, providing more value to the user.”

Studies have shown that on average, enterprises leverage about 35% of their structured data for insights and decision-making, but only 25% of their unstructured enterprise data. At Chisel AI, we use supervised learning, a “human-in-the-loop” approach to “teach” the machine how to read and extract unstructured data from commercial lines digital insurance documents, and automatically execute key processes on these extracted values such as triaging and prioritizing submissions. Leveraging AI to extract and ingest data across downstream systems, enables human knowledge workers to focus on high-value work. In the case of underwriters and underwriting teams, it enables them to focus on strategic judgement work, risk assessment, and nurturing distribution partner and broker relationships. During a recent interview, Susan Penwarden, Chief Technical Underwriter, Aviva Canada confirmed that “beyond efficiency, leveraging the power of data analytics and machine learning to drive better pricing and underwriting decisions, manage performance and improve our understanding of risk is continuing to increase in importance.”

“The data has to, at some point, turn into information that is friendly to how a human interacts with the data. It’s got to be meaningful, and it can’t be, "Here's a nine-hour video of a drone footage." That’s not valuable for a person,” explains Nathan Root, Head of Product Management in Technical Underwriting, Zurich North America. “I think there’s a lot of decisions to be made and a lot of work underway. It’s exploding right now as to how to best consume that information that’s available and package it in such a way, into information that can support the underwriting decisions. It’s not going to be a rules engine. It’s going to be some AI approach or machine learning type of approach. Understanding how those decisions will need to be made and how those models will need to be constructed is the work ahead for the industry.”

Read the Carrier’s Guide to AI-Powered Submission Intake

To learn how to unlock accurate, actionable risk data from submission documents to identify the best submissions to quote faster, drive premium growth, and reduce loss ratios, download A Carrier’s Guide to AI-Powered Submission Intake.

Submission Intake Carrier Guide Cover

Read the Guide


Browse different topics

Recent Posts