PolicyIQ

Commercial insurance comparison runs on PDFs. Coverages, limits, exclusions, scattered across carrier documents in formats that don't match and language that doesn't reconcile. For an independent insurance agency handling submissions across multiple carriers, every comparison used to mean a full human read of every document.

PolicyIQ reads the documents. Carrier proposals and policies go in as PDFs, native text or scanned; coverages, limits, deductibles, and premiums come out, mapped to one structured ontology across sixteen lines of business regardless of which carrier wrote the paper. Policies sit side by side, carrier against carrier, term against term. Every displayed value carries a verbatim citation back to the exact passage it came from, a Worth Noting pass flags discrepancies and gaps a broker would want to see, and Robin, the built-in assistant, answers questions about any policy with the same citations.

Underneath: a two-model extraction ensemble in which two frontier models answer every question independently and must agree, and a stack of deterministic accuracy guards that hold any uncertain value for human review rather than guess. Nothing displays that the document does not literally say. Independently audited on real client policies at 96 to 100 percent displayed-value accuracy with zero invented values, with a frozen golden set that re-scores every release in minutes. Runs on Azure with per-tenant isolation and more than two thousand automated tests. PolicyIQ: live in production.

Built as part of Document & Data Processing.