Because underwriting, selling, and servicing rely so heavily on humans processing large quantities of written or verbal communication, existing tools have struggled to properly automate these services and materially impact loss ratios (losses on written premiums) and expense ratios (underwriting and servicing written premiums). Large language models (LLMs), with their ability to proficiently collect and distill large amounts of data, could change this as they can augment or fully replace the process of a human combing through large amounts of data.
While current machine learning technology allows for improved decisioning on simple products like auto and home insurance, more complex underwriting processes like commercial and life insurance remain challenging. This has less to do with the process of decisioning relevant data and more to do with collecting and synthesizing the relevant data. While traditional ML models have helped dramatically improve more standardized underwriting processes like home and auto, LLMs could potentially help with the more complex group by gathering data to help underwriters make better decisions, especially in more intricate cases like large commercial policies where more context and follow-up questions are required. For example, most large commercial policies cover dozens or more locations, and each location has specific nuances (such as electrical panels, fire doors, sprinkler density/effectiveness, management effectiveness, amount of combustible storage) that must be gathered from the applicant, understood by the underwriter, and evaluated against underwriting guidelines. LLM-powered workflow software for underwriters could drive down underwriting time and cost while increasing accuracy.