Data has long been crucial to the insurance industry, informing risk assessment, pricing and claims processing.
However, while some data used in insurance is structured and organized into standardized formats, most data leveraged across the sector is unorganized and highly changeable, locked within reams of paper, email and attachments. This information falls under the category of unstructured data.
According to multiple analysts,
In insurance, complex documents – loss run reports, statements of value (SOV), underwriting supplements, police reports, legal demand letters and medical records – contain information critical to driving maximum operational efficiency and better outcomes for agents, brokers and their customers.
Mastering insurance's unstructured data challenge
Insurance businesses that master unstructured data challenges at scale with minimal disruption will develop a substantial competitive advantage.
Many insurers have attempted to convert unstructured natural language data into structured data with technology solutions. This approach, however, has met with mixed results due in part to insurance's unique document processing demands.
Even with established standardization formats (e.g., ACORD) in the mix, most documents related to insurance underwriting and claims contain multiple pages with elements in various formats, including email attachments, handwritten notes, faxes and others.
Up to this point, technology has been unable to keep up with the explosion of data around insurance, largely because the tools used by underwriting and claims teams rely heavily on structured data to operate. As a result, these teams often process documents by hand.
The opportunity
Ultimately, insurance businesses succeed by providing outstanding protection and service to policyholders. Solving our industry's unstructured data challenge will be a major step toward that goal.
Breakthrough technologies, like Generative AI, excel at natural-language-based tasks. These models can understand context, extract appropriate information and generate human-like responses to user input. Accenture forecasts that Generative AI will potentially automate up to 62% of underwriting and claims processes – enabling human experts to concentrate solely on higher-value tasks. Systems built around insurance-specific Generative AI are currently delivering real benefits.
Already, underwriting teams
In claims, Generative AI-powered systems deliver strong results throughout the entire process, including document handling times reduced from 3-5 days to minutes–at 95%+ accuracy—and a 90% reduction to manual claims setup. Results like these are driving significant reductions in claims overpayment while expanding operational capacity to contend with spikes in demand from natural catastrophes.
Clearing a path to operational excellence with AI
Generative AI has demonstrated significant capability in processing unstructured data, but choosing a system is a complex decision requiring a measured approach to managing new risks and challenges related to data, security, ethics, and regulatory compliance.
Generative AI built around publicly available models offered by Google, OpenAI, Meta and others are not trained and fine-tuned on insurance data. This lack of insurance context can impair their ability to integrate seamlessly with insurers' existing processes and systems, delaying time to production and value.
Building a system in-house requires resources out of reach for all but a handful of global giants. According to an Oliver Wyman/Celent survey, over 20% of insurers in the U.S. are attempting to build vertical Generative AI solutions in-house. However, less than half of these businesses have successfully developed these projects from the proof-of-concept phase into production. This poor performance stems from the dependence on in-house specialized knowledge, infrastructure and data resources to build and manage models with insurance domain expertise.
Neither of these approaches to implementing Generative AI offers a clear path to success.
Factors for positive results
When evaluating AI solutions, insurers should ask the following questions to ensure the best outcomes and scalable results:
- Is the solution trained on insurance data?
- Are the models fine-tuned to the organization's specific insurance domains?
- Does the technology recognize and identify specific data from any source as document volumes rise?
- Can the solution adapt to new document types without additional training and use the organization's data in a secure environment?
- Will the offering integrate with existing systems out of the box for a successful user experience?
- Can the technology be used with minimal human oversight with clear exception-handling routines?
- Is the solution highly secure, reliable and aligned with users' internal policies and industry standards?
- Will the technology enable the organization to be compliant with local and federal data regulations and in-house standards for auditability and bias?
Additionally, insurance IT, underwriting and claims leaders should look at general characteristics in four key areas to determine the suitability of a Generative AI solution for their business needs.
First, the solution should contain large language models (LLMs) for natural language processing. A subset of Generative AI, LLMs comprehend and process human language, making them ideal for handling text-based unstructured data like emails, reports and policies.
The system should also offer contextual understanding of information to enable accurate interpretation of complex documents.
The solution should be data-aware, identifying missing information (e.g., blank cells, empty checkboxes, etc.) and improperly formatted data to more effectively optimize extraction from large volumes of unstructured data.
Lastly is data synthesis, where the LLM can integrate information from various sources to produce insights or summaries, assisting decision-making processes.
Insurance's unstructured data challenge is massive, but so are the opportunities for companies that effectively harness this information. Generative AI offers natural language input processing capability to address the challenge. It empowers insurers to unlock the value hidden in their documents and delivers critical decision data into underwriting and claims management workflows. This, in turn, improves efficiency and allows underwriting and claims experts to concentrate on driving greater value through customer engagement, resulting in mutual success through superior service and enhanced insurance products.
