Extract data from insurance policies

An insurance policy can run 80 pages, but the page everyone actually reads is the declarations page, the summary at the front that states who is insured, what is covered, and for how much. Brokers, risk managers, and claims teams pull the same fields from it regardless of line of business: the named insured, the policy number, the carrier, the policy term, and the schedule of coverages with each coverage's limit and deductible. A commercial account with general liability, property, auto, and umbrella coverage has a declarations page per line, and a risk manager tracking a portfolio reconciles limits and renewal dates across all of them. Trouble comes from the coverage schedule and the way limits are expressed. A general-liability dec page states a per-occurrence limit and a separate aggregate (such as $1,000,000 per occurrence and $2,000,000 aggregate), and the two are not interchangeable. Property coverage carries a building limit, a contents limit, and a deductible that may be a flat dollar amount or a percentage of value. Endorsements listed by form number modify the base coverage in ways that only matter at claim time. The policy term sets the renewal date a broker has to act on, and a policy effective 2026-01-01 expiring 2027-01-01 drives the renewal calendar. Talonic reads the declarations page and returns the named insured, carrier, policy number, term, and the coverage schedule with each limit, sublimit, and deductible as discrete fields. A broker or risk team loads a portfolio of policies and tracks limits and renewals without retyping each dec page. A Goodwin Retail Co. policy from Hartford Fire Insurance, effective 2026-01-01 to 2027-01-01, carries $1,000,000 per occurrence, a $2,000,000 aggregate, and a $5,000 deductible in USD, which a US risk manager tracks against the renewal date.

What gets extracted from insurance policies

How extraction works for insurance policies

Declarations pages arrive as carrier PDFs, broker-management-system exports, and scans, and each carrier and line of business lays the schedule out differently. Talonic reads the dec page and maps it to the policy schema in the Field Registry, which captures the named insured, carrier, policy number, term, and a coverage schedule. Limits are kept as distinct fields, so a per-occurrence limit and an aggregate limit are never collapsed, and a deductible is captured as either a flat amount or a percentage with its basis. Endorsements are captured by form number and linked to the coverage they modify. The policy term is parsed into effective and expiration dates that drive the renewal calendar. Every value returns with a confidence score and pixel-region provenance under DIN SPEC 91491 conformity, so a broker can verify a limit against the source declarations page before binding or renewing.

Sample extraction

A commercial general liability declarations page

{
  "named_insured": "Goodwin Retail Co.",
  "carrier": "Hartford Fire Insurance Co.",
  "policy_number": "GL-2026-558102",
  "line_of_business": "commercial_general_liability",
  "coverages": [
    {
      "coverage": "each_occurrence",
      "limit": 1000000
    },
    {
      "coverage": "general_aggregate",
      "limit": 2000000
    },
    {
      "coverage": "products_completed_operations",
      "limit": 2000000
    }
  ],
  "deductible": 5000,
  "effective_date": "2026-01-01",
  "expiration_date": "2027-01-01",
  "currency": "USD"
}

Frequently asked