Insurance · Global Insurance Broker

Structured data from commercial insurance policies

We ran a head-to-head evaluation of Azure Document Intelligence and Azure Content Understanding for pulling key terms out of long commercial policy documents.

63 / 72
Fields populated
Content Understanding, vs. 56 for Document Intelligence
6
Carriers and lines tested
Policies of up to 79 pages

The challenge

Commercial policies, proposals, and binders run up to 80 pages, and every carrier lays them out differently. The broker wanted to know which approach could reliably extract named insured, limits, premiums, deductibles, and sublimits without training a model per carrier.

What we built

  • ✦A 12-field extraction schema covering parties, policy number, line of business, dates, premium, limits, deductible, and nested sublimits
  • ✦A Python harness running both services on the same documents: Document Intelligence layout with query fields, and a custom Content Understanding analyzer
  • ✦Policies from six carriers and lines, including excess liability, property, cyber, and a 79-page business proposal
  • ✦Compared fields populated, latency, and token cost for each document, plus two analyzer and model versions
  • ✦Only Content Understanding returned sublimits as a structured list rather than free text

Security & compliance

  • ✦Sample documents were obfuscated before processing

Stack

PythonAzure Document IntelligenceAzure Content UnderstandingAzure OpenAI

Building in a regulated space?

Send us an email and we'll walk you through how we'd approach it.

Email us