Yingju Ma

University of Cambridge · Economics · 2026–27 job market · Postdoc at University of Cambridge

Fields

Finance, Innovation, Industrial Organization, Theory

Job market paper

Generative AI in Financial Services: A Field Experiment in Insurance Advisory

I study a generative-AI tool that offers insurance brokers real-time reply suggestions, using two randomized phases of its rollout, involving 1,070 licensed brokers at China’s largest insurance brokerage. In a ten-week pilot, 253 randomly selected brokers were given access to the AI tool; in a subsequent 20-day experiment, about 165,000 customers were randomly assigned to conversations in which the broker either had or did not have access to AI advice. Access to AI advice raises realized premium by 28 percent within 15 days among customers of non-pilot brokers, and customers buy more life insurance. AI’s benefits depend on broker ability and the product. In life insurance and large policies, where premiums per sale are higher, gains rise with brokers’ baseline ability; in health insurance, the tool narrows performance gaps across brokers. Brokers who adopt suggestions more often do not gain more, suggesting that adoption frequency is an imperfect proxy for AI’s value. Pilot brokers’ advantage emerges over the initial weeks and persists after full roll-out, even when adjusting for their greater effort, consistent with learning.