Execution and Accountability Are the Same Problem

Scan this year's ITC agenda and you'll notice something: there isn't one conversation happening on agentic AI in insurance. There are two, and one is getting far less stage time than the other.

I'll be spending my time at sessions that ask what happens after an agent goes live, not just what it can do on day one.

That's the split I'm focused on closing: execution and accountability aren't two separate problems. They're one and the same.

The Execution Story

One big conversation happening at ITC is about capabilities. AI solutions are already past the pilot stage, supporting people and systems throughout every corner of the insurance industry.

Interesting sessions to me include:

• "The future of insurance: Powered by AI," opening the AI Powered Innovation Summit, this session frames it plainly: work moving from software that supports people to systems that execute quoting, triage, decisioning, documentation, and servicing directly.
• A panel with a Chief Data Officer, Chief Transformation Officer, and Chief AI Officer will talk through where human judgment still has to sit on an AI-native desk. We think of this as Human-In-the-Loop workflows in regulated industries.

None of these sessions are to hype us up anymore. These are real operators describing real deployments to those interested.

The Accountability Story

The second conversation being had is quieter, led by a different set of people. It's not the underwriting and claims operators demoing agents but the risk, compliance, and governance leaders who have to sign off on them.

• One session, with Arch Capital's SVP of Finance, is about permissions, cost control, and auditability once you have more than a handful of agents running. Figuring out this workflow is essential for a successful implementation.
• Another is about connecting decision engines, human workflows, and compliance checkpoints into one system instead of three.

Notice how few sessions are asking this question compared to the ones about what agents can execute. Which tells you where the industry's attention still is.

Why Accountability and Execution Need to Be One and the Same

Here's my issue with the separation: it doesn’t actually exist.

In a regulated industry, execution and accountability are the same problem, described from two different vantage points. An agent that can execute a claim decision but can't produce an audit trail a regulator would accept hasn't solved anything for insurance — it's just solved a demo.

The industry keeps treating this as a tradeoff: move fast and stay light on governance, or govern carefully and accept you'll be slower to deploy. But there is no trade off. You need, and can have, both execution and accountability. The constraint isn't technical, it's whether anyone designed it from day one.

In my opinion, when a promising pilot fails it's almost never a capability gap. It's an accountability gap. The agent works but nobody can explain why it made a call it made in a way that would satisfy the person who has to defend that decision six months later.

That's the thesis we built Notch around. We ran an MGA before we were a software company. We've sat on the other side of an audit with no way to explain a decision after the fact. AI governance isn't a feature to us. It's table stakes. 

What I'd Ask On the Exhibit Floor

If you're an insurer or independent agent attending ITC this year and sitting through an agentic AI demo, ours or anyone else's, these are the four questions worth asking before talking about ROI:

1. Walk me through what happens when the agent gets it wrong.
2. Can I see the audit trail for a specific decision in a format a regulator would accept?
3. Who signs off when the agent's instructions change and is that logged anywhere?
4. Is this deterministic or probabilistic and does your team actually know the difference in how they talk about it?

If a vendor can't answer these questions cleanly, the conversation about ROI doesn't matter yet.

See you in Vegas.

‍