Advocate Insurance Desk
What if the most accurate way to run an insurance compliance check is to stop the AI from thinking?In this episode, Katie sits down with David, head of product engineering at Advocate, who built the World Insurance Model (WIM) over roughly five years of R&D. WIM is the deterministic engine underneath the Advocate app, the thing that turns thousands of manual policy checks into consistent, testable results. The Advocate Insurance Desk is a data-driven commercial insurance podcast. Most episodes we use the Advocate Market Terminal, our insurance intelligence platform, to show exactly what is happening inside specific markets: real carrier behavior, real premiums, real pricing by segment. This episode goes one layer deeper, into the engine that powers the platform itself. The core idea: frontier models on their own are not good enough for compliance work. They perform decently and then leave you to clean up the rest. Pair a model with WIM as a tool, and it offloads the reasoning to a deterministic engine that returns the same output for the same input every time. Accuracy roughly doubles while token cost stays flat, because the model stops guessing at requirements and starts asking WIM which fields actually matter.In this conversation we cover the scale of the problem (around 30 million commercial policies and 500 billion dollars in annual premium running through compliance every year), what deterministic actually means and why it matters when one missed check can cost millions, how documents flow through the platform from upload to compliance report, and the Advocate App Labs benchmarks: Sonnet moving from 26 percent of coverage gaps found on its own to 63 percent with WIM, the rule engine alone reaching about 74 percent, and a licensed human reviewer still leading at 96 percent. David also walks through the model harness that mixes engine and frontier models step by step, how hallucinations are handled with citations and a human in the loop, why review time compresses from about 90 minutes to a couple of minutes, and what WIM means for brokers looking to expand into new asset classes.The takeaway: the reading and data-pulling part of a review is already being automated. The judgment calls stay with the human, and the accuracy gap keeps closing.See WIM at work. Create a free account at https://advocate.app/?utm_source=spotify&utm_medium=podcast [https://advocate.app/?utm_source=spotify&utm_medium=podcast]The full benchmark study is on Advocate App Labs.Connect with David: https://www.linkedin.com/in/david-a-haddad/Chapters0:00 Why this episode goes one layer deeper1:16 Meet David, who built WIM2:24 The scale of the problem: 30 million policies3:30 What the World Insurance Model actually is4:52 Deterministic vs probabilistic, explained6:05 Dropping documents into the platform7:56 Benchmarking WIM against the frontier models9:12 How a model uses WIM as a tool12:18 The numbers: accuracy gains and cost15:05 The harness and mixing models16:20 Hallucinations and the human in the loop18:18 Will AI take the reviewer's job?19:19 Trying WIM yourself20:11 What WIM means for brokers22:16 What to watch for next #AdvocateInsuranceDesk #AdvocateTechnologies
21 episodios
Comentarios
0Sé la primera persona en comentar
¡Regístrate ahora y únete a la comunidad de Advocate Insurance Desk!