
AI Solutions Beyond the Hype

Artificial intelligence has been touted as a revolution for everything from advancing nuclear physics to identifying mystery plants with just a snap of a picture.
In this Q&A, Epstein and Watkins provide much-needed grounding in the face of the hype and discuss how the insurance industry can take advantage of AI and large language models to standardize and centralize operations. They also discuss how to develop AI use cases that give accurate, trustworthy results by leveraging deep insurance and engineering expertise internally and externally with partners like ReSource Pro, a business and technology services provider focused exclusively on the insurance industry, specializing in workflow optimization and transformation, data and systems migration, and managed services.
DAN EPSTEIN: One of the key takeaways is that it’s very clear agencies see that AI is real. It’s compelling. They believe the promise is significant, but they also believe that the maturity level is low.
When we asked people to look at the Gartner hype cycle, 62% felt that AI was in the earliest stage, which is called the innovation trigger, and only 1% saw it as at a kind of a steady state, more productive layer level. In that sense, it’s still seen as something that is full of promise but not yet ready for prime time. ReSource Pro is offering to share this primary research with readers who are interested in exploring the findings further.
EPSTEIN: One of the key constraints is the sheer fragmentation of the work inside most insurance organizations. We have seen instances where the same insurance organization has thousands of different ways they processed the same task.
We see an opportunity for significant standardization and centralization, and we work with agencies to help them get there.
It’s important for insurance organizations to look across their businesses and understand what level of customization is appropriate and important given the specifics of the line of business and characteristics of the risk to determine what level of variation is what you might think of as over-servicing or over-customization.
EPSTEIN: Our survey was very interesting, because it was clear that our clients saw that AI on its own was not sufficient. We asked, “To achieve value and scale, how much effort should be dedicated to people & processes versus tech & algorithms?” Fifty-two percent said 70% [of effort should be focused on advancing] people and processes. Forty percent said an even split of focus should be applied to people and processes and AI—so 92% that think at least 50% of the focus should be on people and process.
If you’re going to rely on AI agents, you have to have confidence that they are doing insurance work correctly and not with any kind of bias or hallucinations. It’s going to be people with deep process knowledge and nuanced understanding of insurance that are going to be required to oversee the outputs.
WATKINS: ReSource Pro has been on this journey for many years. Take, for example, extracting data from PDFs. There’s a lot of domain expertise and engineering process discipline that goes into putting together a system to not only improve the accuracy of AI-enabled data extraction, but to understand where the inaccuracies are and have confidence in the results. Through close collaboration with insurance experts at ReSource Pro, we’ve been able to improve the accuracy from 80% to 99% with a 95% confidence factor of exactly where any errors might be. That’s important because otherwise you spend too much time auditing for errors.