Generative AI is accelerating transformation across the commercial insurance industry. Processing risks is a manual and time-consuming process, but GenAI technology and help lift premium per person and provide profitable growth without adding operational costs.
Large language models improve AI performance and scalability, enabling the insurance industry to do more across multiple functions - including writing more risks with the same resources, servicing and adjudicating more claims with the same-sized team, achieving higher levels of accuracy in evaluating risk and improving how individual risk selection decisions support targeting portfolio shape.
Over time, LLMs - tuned to commercial insurance - will enable a move to a model where companies can onboard, renew and service risks at close to zero marginal cost, underpinned with more consistent control over risk selection.
LLMs are trained on vast amounts of data that allow them to understand and generate content to perform various tasks. When tuned to specific domains, they can solve deep problems that unlock massive value.
To fully understand how LLMs work and how they streamline the insurance process, let's break it down.
What are LLMs?
LLMs are a foundation model trained on vast amounts of data, providing the capabilities to drive multiple applications and tasks. They are trained to understand and generate content like a human and create relevant responses for various tasks. LLMS can perform various functions, including classifying, editing, summarizing, interpreting, answering questions and creating content. Within commercial insurance, risk information is heterogeneous and unstructured, LLMs enable risk data to be unified, digitized and standardized so risk decisions can be made in a more streamlined way across the value chain.
See also: How AI Can Keep P&C Insurers Profitable
How LLMs work
LLMs are based on transformer architecture, which consists of multiple layers of neural networks with an encoder and decoder with parameters that can be fine-tuned during training. LLMs learn the relationships between different portions of words (or tokens), which enables them to be effective at generating both structured and unstructured content (including natural language text). Fine-tuning like reinforcement learning with human feedback (RLHF) can remove biases and factual errors. LLMs can be trained on unstructured data, which is one reason why they are powerful in the context of insurance, which generates large amounts of such data. LLMs are also able to create new forms of content efficiently, including text and images, which enables them to perform a wide array of tasks.
How are LLMs used in the insurance industry?
When tuned to insurance, LLMs support underwriting and claim adjudicating capabilities, streamlining risk processing, lifting efficiency and improving broker and customer service at all stages of the insurance process. LLMs enable risks to be digitized, evaluated and turned decision-ready without human intervention and let homogeneous risks be handled via straight-through processing. In more complex risk segments, LLMs create significant capacity for underwriters to write more risks and focus on the aspects of them that are unusual relative to the norm, enabling a more informed decision.
LLMs also let insurers retain their unique view of risk yet achieve high levels of efficiency by digitizing, standardizing and interpreting risk data relative to their specific schema and rules. LLMs enable risk data to be transformed into a format that corresponds with the insurers' unique target schema.
An important precondition of achieving LLM performance in risk digitization is uniting dispersed risk data across many internal and external data sources.
Claims
When tuned to claims processing workflows, LLMs can classify claims documents, digitize claims information, link different claims data fields together across different transactions and enable claims to be turned adjudication-ready without human intervention. This enables claims teams to service more claims faster, with the same resources, resulting in better customer service. Lower-complexity claims can be auto-adjudicated, resulting in progressively higher levels of straight-through processing.
See also: What Makes Insurance Invoicing Different
Risk Analytics
Any sharp learning curve required by staff to analyze complex data is reduced with LLMs, making analysis available to users who may not have relevant technical training. Risk professionals can ask questions about a submission to identify aspects of the risks that require specific attention. Risk professionals can also ask contextual questions and compare a given risk with the entire risk submission flow and in-force portfolio to understand the degrees of homogeneity and difference. This enables economies of scale, where similar homogeneous risks can receive more accelerated processing and allows better, more integrated decisions across different underwriting and claims teams.
Closing
For many years, full digitization of insurance workflows was constrained due to the vast heterogeneous data formats, a lack of standardization and different requirements that each insurer has. Today, LLMs are enabling insurers to digitize end-to-end risk workflows and move to fully scalable business models due to their ability to unify, digitize and understand risk data regardless of their format and level of heterogeneity.
