The adoption of artificial intelligence in insurance is less a sprint to the finish than a scramble to reach the starting line, experts say.
“I think this is sort of a race to the starting line,” said Indico Data CEO Tom Wilde. “Getting your data in order so that AI can take advantage of it is going to prove to be one of the keys to compete into the next decade.”
While agentic and generative AI are still in their infancy within the insurance sector, Wilde believes these technologies are poised to revolutionize decision-making in ways once thought unimaginable.
“I think of insurance as sort of driven by decisioning: Should I underwrite this risk? How do I adjudicate this claim?” he said. “We kind of describe this as the dawn of the decision era where the combination of the rise of cloud computing and data investments and now AI really creates an unprecedented opportunity to drive high-performance decisioning across the enterprise—especially the insurance enterprise.”
Wilde noted that insurers are beginning to realize their long-standing dependence on individual expertise has its limits.
“I absolutely have heard and seen [carriers] moving from this period of individual expertise to institutional expertise,” he said. “And AI really allows us to ring that out and make sure that decision capability is done repeatedly with transparency. So, I think that’s a huge opportunity.”
He added that, for the first time in the insurance industry, it's now possible to turn unstructured data into actionable endpoints—streamlining processes and improving efficiency.
“We can take a 25-page underwriting guideline document and almost make it a programmatic endpoint where we can query it. We can make it part of a workflow,” he said. “Whereas previously, everything would grind to a halt while somebody had to open it and read it and figure out how to apply what was in that document to whatever process it was part of.”
Although AI is transforming key areas of insurance—from underwriting to claims—humans still have a vital role to play. Michael Parcelli, senior vice president at Xceedance, emphasized that successfully integrating AI hinges on striking the right balance between automation and human judgment.
“It’s not going to replace humans wholesale. There’s still going to be an exception management trail there, and you still want an auditor and some oversight. So, I think that you’ll still have a balance,” he said. “But whereas you may have done 80 percent of the work in the past, and maybe 20 percent of the oversight and auditing, you can spend 20 percent on your work and then leave that 80 percent for the tools and enabling technologies to pick that up.”
This approach allows technology to handle routine operational tasks, increasing efficiency, while humans provide oversight to safeguard accuracy and fairness.
“AI still needs some kind of human intervention,” Parcelli said. “It’s not perfect, and it’s not true fact all the time. There are misrepresentations as we work through some of the development and design of our own software and products. We have to correct and edit it, and then there’s an iteration process… So, I would not suggest that you can rely 100 percent on the credibility and the validity of the information that comes through. It has to be tuned to work properly, and that’s why I think it will always need some level of oversight.”
Wilde agreed that as the insurance industry advances toward a more agentic future—where AI systems make decisions and learn autonomously—insurers bear the responsibility of ensuring that every step in the process is traceable through a clear audit trail.
“If you think about how much complexity could be embedded in an agentic workflow—things like what model was used, what version of what model, what prompts were used, who wrote those prompts, who could edit those prompts—all of that has to be trapped in an audit trail to be able to assess an agentic decision and improve it over time,” he said. “As regulators show up and want to know exactly how these decisions are being made, you’re going to have to be prepared to answer that.”
