Risk Management: Brick by Brick

The episode covers genuine tactical insights about AI implementation in insurance - governance frameworks, data quality ownership, model fairness testing, and process-driven transformations - but frequently retreats into abstract discussion about 'culture' and 'mindset' without concrete operational detail. Several segments lack density, particularly around talent retention and executive conversations.

“Most of what we had applied. There are things that are different. So for example, with AI models you don't have hold out data sets that you then test outcomes again.”

“Data quality is everyone's problem. But it has to be jointly owned. It's not a data governance team issue. It's not an IT issue. It's a collective issue that we get the right information to actually feed into the models and get the right answers out.”

The guest rehearses established frameworks (buy vs. build decision criteria, democratization of data, cross-functional silos) that circulate widely in enterprise tech discourse. The most original moment - linking machine learning's introduction to auto insurance decades ago as a precursor to current AI transformation - is good but brief. Most other claims are predictable takes on AI adoption maturity.

“People with AI will replace people without AI.”

“I think companies that are listening to the need from their customers are trying to get product out quickly.”

Kristen Bessette holds legitimate operational authority as CDO at Zurich North America with deep actuarial and analytics background across three major carriers (Liberty Mutual, QBE, Zurich). She has executed real transformations, not theorized about them. However, she remains primarily an internal practitioner speaking to her own execution rather than someone with cross-industry pattern-recognition or market-shifting influence.

“I've been with Zurich, uh, North America for about a year and a half. Prior to that I was at QBE North America, I was Chief Actuary Data and Analytics Officer there and then um, prior to that a lot of actuarial roles at Liberty Mutual.”

“Chief Data Officer at Zurich North America”

The episode includes some concrete examples (data center insurance product launch, visuals for underwriter risk identification, Guidewire/Salesforce embedding) but avoids naming metrics, revenue impact, timeline specifics, or measurable ROI. Many claims remain at the level of 'we're doing things faster and better' without quantification. Risk manager audience deserves dollar figures and outcome data.

“We work with vendors to bring in visuals so the underwriters can see really quickly where a property has risk and where it doesn't.”

“we got our data center product out very quickly into market... Relative to maybe what we would have done a couple years ago.”

Host Jason Reichel asks sharp, layered follow-ups that push the guest beyond platitudes - particularly on talent retention challenges, interconnected risk modeling, and the gap between perceived vs. actual exposure. He contextualizes claims within risk manager reality (San Francisco/Waymo example) and challenges expectations around AI accuracy. However, he occasionally accepts surface-level answers and misses opportunities to push harder on governance trade-offs or competitive differences.

“One of the dirty secrets that I keep running into is we give tools to say a risk manager within a corporate environment, they think they know what their risk appetite is, they think they know what their exposure is. They run these through these models and they get a very different picture than what they believed their exposure was or these things.”

“Did you spot those trends early on and were you able to respond to them or were you able to pull together a product quickly because of the data.”