Published
AI didn't start with ChatGPT, and the opportunity in underwriting goes far beyond cutting costs. In the debut episode of Agentic Insurance, Joe Kleinhenz and Genpact's Yasir Andrabi dig into where agentic AI genuinely moves the needle for property and casualty (P&C) underwriters: better risk selection, tighter loss ratios, and more time for the judgment calls that actually require a person.
Joe brings more than 20 years of machine learning and AI experience across Fortune 100 companies, including scaling a global AI team to over 150 data scientists. His take? We're at a tipping point – and the insurers that redesign their operations around AI now, rather than bolting it onto existing processes, may be better positioned to pull ahead.
This conversation moves beyond the hype to look at what works, what doesn't, and what to do next.
What you'll hear in this episode
The discussion covers the full arc of agentic AI in underwriting – from its historical roots to the strategic decisions leaders face today.
The evolution of underwriting technology
AI in financial services predates the generative AI boom by decades. We reframe the conversation – this isn't a new technology; it's a new chapter for one that's been quietly reshaping the industry for years.
What agentic AI actually does for underwriters
The biggest opportunity isn't speed. It's improving underwriting judgment at scale. Agents gather evidence, flag risk signals, surface inconsistencies, and present recommendations – so underwriters can focus on the calls only humans can make.
When not to use an agentic system
A critical, often-overlooked point: some underwriting decisions require deterministic outputs, not probabilistic ones. Joe explains where agentic AI fits, where it doesn't, and how to blend it with classical machine learning to meet regulatory requirements.
The "blank slate" question
If you were designing underwriting from scratch today, with all these tools available, would you build what you have now? The answer shapes everything about how you should approach AI adoption.
What's blocking scale
Process debt. Data debt. Talent debt. Most insurers struggle to get past proof of concept, and Joe identifies exactly why and what to do about it.
How to stay ahead without burning out
Upskill your people, take an enterprise-wide approach, and stay agile. Joe's advice is direct: don't get locked into one solution, keep learning, and share that learning across the organization.
Key takeaways
Agentic AI augments judgment – it doesn't replace it. One of the most valuable things an agentic system does is free underwriters to focus on decisions that genuinely require human expertise
Rethink the process, don't just accelerate it. Bolting AI onto a broken process gives you a faster broken process. The real gains come from redesigning underwriting with AI as foundational, not supplementary
Not everything should be agentic. Probabilistic systems and regulatory audit trails don't always mix. Know when to use a deterministic model alongside your agentic tools
Data and process debt are the real blockers. The biggest barrier to scaling AI in insurance isn't technology – it's the accumulated weight of poor data quality and outdated workflows
Speed of learning has changed. "Fail fast" no longer means what it did a year ago. The organizations pulling ahead are the ones sharing insights across teams and pivoting quickly when something isn't working
Upskilling isn't optional. AI fluency is becoming a baseline expectation – the same way email and spreadsheets did. The underwriters who embrace these tools can put themselves in a stronger position than those who don't
Frequently asked questions
What is agentic AI in the context of P&C underwriting?
Agentic AI refers to AI systems that don't just automate fixed tasks – they can gather information, evaluate risk signals, identify inconsistencies, and present recommendations with a degree of autonomy. In underwriting, this means an agent can handle evidence collection and pattern recognition while the underwriter focuses on judgment-intensive decisions.
How is agentic AI different from traditional underwriting automation?
Traditional automation follows predefined rules and workflows. Agentic AI can determine what to do next even when a pathway isn't preset. It can accumulate context over time, learn from outcomes, and decide when to act versus when to escalate, making it far more adaptable than rules-driven systems.
Is agentic AI suitable for all underwriting decisions?
Not always. Some regulatory requirements demand deterministic outputs, meaning the same input must produce the same decision every time. Probabilistic AI systems don't always meet that standard. In those cases, blending agentic AI with classical machine learning models can give you the best of both: flexibility and auditability.
What's stopping insurers from scaling AI in underwriting today?
According to Joe Kleinhenz, the main blockers are process debt, data quality issues, and talent gaps – not technology. Many insurers treat AI as a point solution rather than a foundational redesign, which limits how far it can scale.
How should insurers approach AI adoption strategically?
Start by asking whether you'd build the same underwriting process from scratch if you had today's tools available. Most organizations wouldn't. The insurers seeing real results are the ones bringing cross-functional teams together to redesign workflows with AI at the core – not layering it on top of what already exists.
How can underwriters prepare for an agentic future?
Build AI fluency across your team. Address fear and resistance directly. And approach learning as a continuous, organization-wide discipline – not a one-time training exercise. As Joe Kleinhenz puts it, AI proficiency is becoming table stakes, much like knowing how to use a computer or email.
Disclaimer: Views expressed by guest speakers are their own and do not necessarily reflect the views of Genpact.
