Snorkel data // insurance
Building AI for insurance means competing on more than the base model. It means proving agents can complete real insurance workflows.
Snorkel builds the expert-curated benchmarks and evals that measure whether a model can do real insurance work. Our environments test whether an agent can execute full workflows within a realistic insurance company simulation, not just answer a question.
Each dataset is built by Snorkel's research team along with human domain experts, so you can have confidence in its ability to help you deliver results.
Featured insurance dataset
Featured
Commercial P&C Underwriting Environment
2,000+ tasks that put an agent inside a realistic insurance company simulation with submissions, contracts, loss history, endorsement rules, structured tables, and callable tools, all scored deterministically.
Used as RL training data, it has improved model performance by as much as 3.9x.
tasks
2,000+
RL Training lift
3.9x
Submissions
Contracts
Loss history
Endorsement rules
Structured tables
Callable tools
resources
Snorkel in Insurance AI
