What does it take to get AI out of pilot mode? Allianz, Anthropic, AXA XL, New York Life, Tokio Marine HCC and Kin on Day 1 of Insurtech Insights USA 2026

What did the host ask the room to do first?

Kristoffer Lundberg, CEO of Insurtech Insights, opened the show with two asks. First, have a clear vision for how AI will help you serve customers better, not just do today’s work with fewer people. He pointed to a case study released the day before by Travelers and OpenAI, where he said voice agents now handle 90 percent of first notice of loss calls for auto claims. The bigger win, he said, was freeing people for more complex claims.

Second, slow down at the start. Too many teams run three pilots at once and then cancel them when the token bill arrives. Drawing on Bent Flyvbjerg’s work on large infrastructure projects, his advice was to “think slow and act fast.” Plan the people, the data and the setup first, then sprint to production.

What did Allianz and Anthropic say it takes to move past pilots?

The opening keynote paired Christian Freytag, Group CTO of Allianz, with Mike Ram, who leads the insurance team at Anthropic. Ram said that when the partnership began, Anthropic set out to train up to 7,000 Allianz developers across three continents in 30 days.

The big change of the past year, Ram said, is the “harness,” the layer that lets a model act inside real systems.

“If the model is the brain, the harness is really the body.”

Mike Ram, Head of Insurance, Anthropic

A year ago, he said, AI could read a submission and run a guideline check. Now it can pull the submission from email, find missing data, draft a reply to the broker for a human to review, and even rate and quote.

Freytag described AI as a staircase. Allianz started with a claims project called Nemo, with a human in the loop on every decision. Next comes giving the whole workforce good tools; his mantra is for everyone to feel one to two hours more productive each day. He also reminded the room of an old truth: you can run a poor process on good data, but not a good process on poor data.

Asked about pitfalls, Ram said “when you sell intelligence, everything is a use case,” so picking the right ones matters. He sees AI change in three pillars: people, process and new products. Most firms underinvest in the first. The ones that do best build small incubator pods of about ten people, iterate safely, and only go to production once evaluations are strong.

On risk, Freytag said AI is changing the risk picture faster than anyone can model it. He also argued that regulation should be seen as an advantage, because carriers that build for the rules can go where others cannot.

“Trust is the ultimate scaling element.”

Christian Freytag, Group CTO, Allianz

Ram’s closing advice was simple. Run hands-on sessions where the C-suite opens laptops and builds something. That is when leaders truly get it.

What new risks does AXA XL see in the intelligence economy?

Lucy Pilko, CEO, Americas, of AXA XL, spoke with moderator Naveen Agarwal about the risks building up behind AI. She sees data centers in three stages: financing, construction and operations. Construction is closest to business as usual. Operations, she said, is “a whole new ballgame,” and the pain points have not shown up yet because most of the money is still in the build stage.

She said business interruption could dwarf the cost of the physical assets. Her concerns include a cyber attack that takes down many providers at once, and autonomous agents making choices at scale.

“A rogue choice can be everywhere all at once, and so the liability impact that creates is massive.”

Lucy Pilko, CEO, Americas, AXA XL

Risks are also moving from physical to digital, and they overlap. Is a loss cyber, tech E&O, property or liability? That makes contract clarity harder. Pilko said insurers can help clients build governance into their code, since a company cannot hand off AI risk the way it can with a third party. She expects more public-private partnerships, though the legal ground is still unclear.

On talent, she wants people who can work across risk lines that used to be separate, much as cyber cover grew out of silent cover in older policies. “The job of commercial insurers is to underwrite the economy,” she said.

How is New York Life using AI with its advisors?

Deepa Soni, EVP and Chief Information Officer of New York Life, gave a simple example. The company has 12,000 advisors, each with thousands of clients. Who should an advisor call today? Past insights were rule-based and often out of date by the time they arrived. AI can spot real-time life events, like a recent job change that makes a 401(k) rollover worth a call, and it learns when advisors flag a bad lead.

She said advisors may spend up to two weeks preparing for a meaningful client meeting, and AI can help with that. She was clear that New York Life is not moving slowly. “I think it’s a disciplined pace,” she said.

Why does Tokio Marine HCC keep AI on a short leash?

In a fireside chat with Aman Gour, co-founder and CEO of FurtherAI, Arron Lamp, CIO of the Public Risk Group at Tokio Marine HCC, said most firms fail at AI because they skip the groundwork. His team pulled policy and claims data into a single format and moved its business logic into shared code libraries. That work is what lets them move fast now.

“You need deterministic logic if you’re going to transact.”

Arron Lamp, CIO, Public Risk Group, Tokio Marine HCC

Underwriting rules, tax and compliance need 100 percent accuracy, he said, and a model never gets there. But reading a big document and pulling out the facts? A model does that well, often better than a person typing. So he prefers AI microservices that do one job very well, with access controls, an audit log and a human who checks the result, over broad agentic AI. If he cannot show his compliance officer how AI is used, he said, he cannot use it.

What makes a company AI-native, according to Kin?

Kevin Greene, CTO of Kin Insurance, said an AI-native company keeps asking “what if a machine did this?” He named four pillars: clear shared data definitions, programmatic access to both data and actions, careful metrics, and empowered people.

Even “quote” means different things to different teams, he said. Kin’s answer is an internal tool it calls Insight Engine, where staff ask questions in plain words and an AI model works through a shared semantic layer connected to the company’s Databricks lakehouse.

He warned against easy-to-game metrics like AI usage or token counts. Kin checks its automation with redundant reviews, sending some work to both an AI agent and a human, which surfaces both model errors and training gaps. And he wants tools in everyone’s hands, not just a center of excellence. At Kin, actuaries now write code for rate work, a quality assurance director built a sales coaching app, and a fraud investigator built her own detection tools.

Where are P&C margins being squeezed, and can AI help?

A panel moderated by Parul Kaul-Green of Datos Insights looked at cost pressure. Dan Woods, CEO of Socotra, said legacy cores make even simple changes, like launching a product in one more state, slow and costly. He said AI has cut his company’s development costs by about 75 percent. His advice on long migrations: take a hard look at anything still running after two years, and if it is three years, “just stop.”

Kyle Nakatsuji, co-founder, president and CEO of Clearcover, described growth and profit as a string pulled from both ends. He warned of a new kind of cost pressure: a competitor that suddenly runs its claims team with half the people. But AI alone does not cut costs. People, systems and data do.

“If you do not change those inputs but you add AI, you have spent more money, not less.”

Kyle Nakatsuji, Co-founder, President and CEO, Clearcover

Lance Ondrej, EVP and Chief Claims Officer at Germania Insurance, said 2023 and 2024 brought three shocks at once: higher reinsurance costs, high inflation and a very active storm season. The lesson was to get claims, underwriting and actuarial working from shared data, so leading indicators show up sooner and rates move faster. He also urged carriers to get full value from tools they already own before chasing new ones.

Topics