Specialty commercial lines carrier
AI Submission Processing
An LLM platform that reads insurance submission emails and attachments, then writes structured accounts into the policy administration system.
A specialty commercial lines carrier was receiving submissions as email, with free-text message bodies and a variable set of attachments: applications, loss runs, schedules, broker notes. Every one had to be read by a person and keyed into the policy administration system before the account could be quoted.
What we built
A platform that reads the whole submission, both the message content and every attachment, using large language models, and interprets it the way an intake analyst would.
- Extracts named insured, exposure, coverage requested, loss history and broker detail
- Reconciles conflicting values across attachments rather than trusting whichever it saw last
- Normalises to the carrier’s own data model, not a generic schema
- Writes the structured account into the policy administration system, quote-ready
- Escalates genuine ambiguity to a human instead of guessing
Why it works
The hard part of submission intake is not text extraction, it is judgement about which document wins when a schedule and an application disagree, and knowing when to stop and ask. We modelled that judgement explicitly, which is why the output is trustworthy enough to write directly into the system of record.
Accounts arrive prepared for quotation without human intervention, and the intake team’s time moves from keying to underwriting support.