The Leadership in Insurance Podcast

The episode contains a handful of genuinely useful operational insights - win-rate benchmarks, contributory data models for bias detection, STP proof-of-concept mechanics, lifetime value as an underwriting signal - but they are interspersed with extended platitudes about jobs changing, going faster, and 'fail fast' thinking that dilute the overall density.

“we have clients that only, of the quotes they send out, they only bind or win about 2.5%. Some are as high as 15 to 20”

“actuarial math that has really dominated insurance is more m about knowing how, what math was used than actually uh, trying to get an accurate outcome. So it's been about transparency.”

The Milliman franchise-model origin story and the indirect carrier acquisition strategy via MGA paper are genuinely non-obvious angles; however, the AI-won't-take-jobs argument (framed with the 'gray hair' heuristic) and the build-vs-buy discussion are recycled takes that appear on almost every insurtech podcast.

“they said well actually our business model is more like a franchise, uh, model which is you can start your own franchise of Milliman”

“my board was surprised that we sold some carriers because we, I sort of Said to him, they take a long time to make decisions. I'm not trying to do that now. And suddenly we land some carriers.”

Stan Smith is a legitimate multi-exit operator and genuine founder-practitioner who has built a real product with paying enterprise customers and can speak concretely to model validation, MLR impact, and go-to-market mechanics; he is not a career podcaster or pure thought leader, though the conversation rarely draws out the deepest technical or strategic depth he presumably possesses.

“This is my sixth startup. I didn't start them all but uh, I've enjoyed the idea of new ideas, innovation, fast moving companies delivering great value and having exciting results as a result of that. A couple IPOs along the way.”

“Their own internal data science team did the measuring, not us.”

The episode scores above average on specificity with named bind-rate ranges (2.5% - 15 - 20%), ROI figures (single-digit millions to tens of millions, one 10x case), a described A/B test methodology, and STP proof-of-concept framing, but it lacks named clients, precise model accuracy statistics, or verifiable market-size numbers, leaving many claims at the anecdote level.

“we have clients that only, of the quotes they send out, they only bind or win about 2.5%. Some are as high as 15 to 20”

“if it's a small book, the ROI will be in the single digit millions typically. But in large books it can be tens of millions of dollars for ROI per year”

The host asks structurally reasonable questions covering governance, human-in-the-loop, and automated underwriting timelines, but consistently follows interesting guest answers with affirmative validation ('Yeah, absolutely,' 'Yeah, as you said') rather than probing follow-ups, and closes with a predictable retrospective softball that yields no new information.

“Yeah, absolutely. Well, that's, that's another point because, um, I think we're just seeing a bit of a domino effect”

“Yeah, I always like to ask this at the end of uh, an episode, Stan. So looking um, back, uh, gradient, how you built it, if you were to go right the way back to the beginning, knowing what you know now, would there have been anything you would have done differently?”