Conversations on the Creek

The episode is only 10 minutes and covers almost entirely surface-level observations - GenAI helps adjusters, hallucinations are a risk, data quality matters. The only moderately non-obvious point is about carriers using smaller pre-trained corpora to reduce hallucinations, but it is not developed with any depth.

“we have seen a lot of carriers think about working with smaller data sets that are pre-trained so they can, A, understand the impact of different changes within that, and also, B, because of the fact that there aren't the same kind of hallucinations”

“first notice of loss, being able to pull all of the different documents in a claims situation together so the adjuster is able to make some sort of a meaningful assessment”

Every idea here - human in the loop, change management resistance, hallucinations, regulatory uncertainty - is a stock 2023-2024 AI conference talking point. Nothing is contrarian, first-principles, or counterintuitive; the guest explicitly echoes the 'phrase of the conference' the host identifies.

“I think, phrase of the conference, is human in the loop”

“not necessarily to take their jobs, but it to help them actually do their jobs in a way that is better”

Andrew Schwartz is an analyst at a research and advisory firm, giving him breadth of industry observation but no direct practitioner experience deploying GenAI at scale inside a carrier. His commentary reflects conference synthesis rather than hard-won operational knowledge.

“Sellent is a research and advisory firm dedicated to helping financial institutions formulate comprehensive business and technology strategies”

“It was a fantastic panel on claims productivity through analytics. Definitely a very receptive audience and a lot of interesting takeaways”

The only concrete data point in the entire episode is that the NAIC model bulletin has been ratified by six state departments of insurance. No carrier names, no metrics, no loss-ratio figures, no implementation timelines or dollar values are provided anywhere.

“the NAIC model bulletin on AI which has been ratified by six departments of insurance now since it was passed at the end of last year”

“loss ratios have been very sticky”

The host asks broad, leading questions ('lowest hanging fruit,' 'regulatory challenges') and never pushes back on a single claim. His follow-ups largely restate what the guest just said, and the conversation ends with filler affirmations rather than any productive tension.

“Sure. So when it does come to generative AI, which part of the claims lifecycle do you think might be the lowest hanging fruit”

“Good stuff. No, I appreciate it. Transparency, traceability, visibility, and what you're looking for in your vendor”