A captive auto finance company already meets car buyers at the moment they need insurance, and it wanted to know whether to sell that insurance itself. The decision had three realistic structures: form a captive carrier, operate as a managing general agent, or partner with a fronting carrier. Each structure carries different capital, licensing, and margin consequences in every state, and leadership wanted a recommendation backed by a five-year pro forma within twelve weeks.
The constraint was research volume. Fifty states of insurance regulation, competitor and carrier activity, market sizing, and buyer behavior all had to be read and organized in weeks. We built a multi-agent AI research engine for the study that splits the work into parallel research streams: state regulation, market sizing and segmentation, attach rates, and the competitive landscape. It returns structured findings the strategy team can check and use, and its output fed the decision matrix, the regulatory roadmap, and the inputs to the pro forma.
Stakeholder interviews ran alongside the research, and they surfaced something no filing or dataset contained: the state of the company's relationships with the insurers it already works with, and how those insurers would respond to a new competitor. The first recommendation underweighted that context, and the client asked for the analysis to be rerun with it weighted in. That is the right division of labor for AI research. The engine covers more ground than a team could read in twelve weeks, and the interviews supply the judgment the data cannot.
The study's deliverables are a decision matrix across the three structures, a five-year profitability pro forma, a 50-state regulatory roadmap, and a go-to-market view built on buyer segments and attach rates. The research engine handled the reading, which kept the strategy team's hours on interviews, modeling, and the judgment calls that decide the recommendation.
Client details in this case study are generalized, and in places combined across engagements, to protect confidentiality. The engine came out of our AI implementation and generative AI consulting work, and our guide to market sizing covers the math the research feeds. We are glad to walk through comparable work under NDA.
