What a difference a less-than-a-year makes!

Just nine months ago, after attending Celent’s 2024 Generative AI Symposium in New York City, I observed that insurers are adopting GenAI like no emerging technology before it. That’s only proven more true; while insurers haven’t been quite as fast on the uptake as verticals like banking, compared to how insurers typically approach a new technology, GenAI’s adoption curve is rocketing up.

Here's where things stand since September: our most recent GenAI-oneers report, which surveyed 55 insurers across both property/casualty and life, found that 44% of insurers have active genAI deployments as of May 2025, putting us near the end of the “early majority” part of the adoption curve. By September that’s projected to increase to 68%, which will firmly place us in the “late majority” part of the curve. Naturally, there may be a self-selection effect in play (i.e., insurers who feel they’re doing a great job with generative AI are probably more disposed to talk to us about it). But considering the benchmarks established in the 2023 and 2024 editions of this study, the rate of increase is still pretty dramatic: 8% deployment in 2023 became 28% deployment in 2024 became 44% deployment in 2025. That’s remarkable uptake.

I think what’s most striking about this expansion is that its happened as we’ve uncovered new use cases for GenAI. Normally insurers like to see established uses for a new technology before they invest. Cloud notoriously was viewed with suspicion, until the benefits became clear. With GenAI, we saw early adopters experimenting with the technology even when the best use cases were suggesting verbiage to customer service and marketing representatives or acting as a natural language query layer on top of internal search engines—cool stuff, but hardly revolutionary.

Underwriting is a great example of how these use cases have flourished. The initial applications for GenAI didn’t lend themselves to underwriting—and in fact, LLMs’ tendency to hallucinate and their inability to produce consistent output from identical input made the use of GenAI for underwriting seem actively hazardous. But then new applications started to emerge, like document summary. As insurers and solution providers experimented, they found that generative AI was really good at absorbing tranches of documentation and highlighting key points. Today, these applications can even point underwriters toward features of a particular risk that are unusual or abnormal, to prompt them to order a report, ask the agent for more info, or take it into account for pricing. From our GenAI-oneers report, we now have a number of case studies that show how insurers are putting GenAI to work helping their underwriters work more efficiently or rate risk better:

• Zurich created an assistant called ZuriChat which can suggest standardized phrasing for policy terms and helps ensure that an underwriter’s decisions align with overall company guidelines

• QBE has deployed an AI-enabled underwriting assistant for its Cyber lines which can read and summarize broker submissions, check for completeness, highlight key risk factors, and determine whether the application is likely to be suitable for QBE’s appetite

• AIG is integrating LLMs into their underwriting process as part of an enterprise transformation and modernization initiative

These applications will continue to grow in complexity as insurers and solution providers start to tap into the potential of agentic AI. I’ve already seen live demos of solutions that can use a natural language prompt to create workflow for a new product, including ingesting documents and calling raters. Given the rate of change, I’m looking forward to Celent’s 2025 GenAI Symposium, which is in Chicago on September 4th—I myself will be leading a session on operationalizing GenAI (not just in underwriting but across the enterprise) and hope you’ll join us to continue the conversation. Insurers and wealth managers can register for free here.