AI Agents Are Shopping for Insurance. The Purchase Is the Hard Part.

What are agents doing on The Zebra today?

“We’ve mostly seen personal agents doing the time-consuming tedious work that consumers hate, like gathering vehicle and driver information, testing coverage options for prices, and then reporting options back to their users,” Herrington says. In a Nov. 5 follow-up, he said that over the prior few weeks, automated agents were comparing rates and completing applications at “more than twice the rate of traditional desktop users.” The comparison is company-reported and does not quantify the share of all quote traffic coming from agents; completing an application does not mean a policy was bound. He says The Zebra has begun allowing a small amount of traffic through infrastructure that lets a customer’s personal agent communicate with The Zebra’s licensed advisors, but agents are not yet completing transactions.

Why could a high volume of agent-generated quotes raise costs for insurers?

Herrington says more accurate, bindable quotes can require carriers to purchase third-party data such as motor vehicle records, which can cost up to $20 per quote. As a hypothetical, he says an agent requesting 40 quotes at $20 each would mean $800 in quoting expenses for carriers, although only one policy might be bought. The Zebra, he says, monitors its own look-to-book ratio and calls quotes only when a customer has a high propensity to bind.

Who is the producer when an agent helps a consumer choose coverage?

“We don’t quite know yet,” Herrington says. He describes a growing chain involving a person, a personal agent, an insurer’s AI and a licensed advisor. At The Zebra, he says, no transaction is completed without a licensed advisor. That leaves regulators and market participants to work through licensing, consumer permissions, disclosures, privacy and recordkeeping as agent-led shopping develops.

Should independent agents ignore, compete with or connect to agentic shopping?

Herrington says independent agencies may be able to ignore agentic shopping over the next 12 months, but argues that building a proprietary consumer-agent front door would mean competing with large technology companies. His recommendation is for agencies to connect to the platforms consumers choose, rather than build a rival personal AI agent.

What needs to happen before agents can buy insurance?

Herrington says insurance platforms need infrastructure that lets agents interact directly and securely, rather than navigate websites like human users. In his Nov. 5 response, he said The Zebra was developing a Model Context Protocol server to give agents a more structured way to interact with its platform. He says the process must also verify consumer permissions, maintain records for oversight and involve licensed advisors where needed; fully autonomous shopping remains ahead.