From submission intake to preliminary indication, without the manual work.
A commercial lines underwriter evaluating 20+ submissions weekly spends 60-70% of time on data gathering, not risk assessment. Each submission involves reading 50+ page broker packages, extracting risk data, entering into the rating system, pulling external data, and compiling for the underwriting decision.
McKinsey found multi-agent AI can handle intake, risk profiling, pricing, and compliance in underwriting. AIG reported AI improved data intake accuracy from approximately 75% to over 90%. By 2026, carriers are scaling AI into production after 2025 pilots.
The bottleneck is data assembly that precedes the decision. Underwriters are skilled risk assessors spending most time on clerical work.
The agent processes broker submissions: applications, loss runs, questionnaires, financials, property schedules. Extracts structured risk data and compiles a unified summary — insured details, coverage requested, loss history, exposure data, risk factors.
Simultaneously pulls external data: property characteristics from data services, loss history from ISO/NISS, financials from D&B. All assembled into a risk profile reviewable in minutes.
Runs through your rating algorithm for a preliminary indication with transparent assumptions. The underwriter reviews, adjusts based on expertise, and decides. Agent handles assembly; underwriter handles judgment.
Document extraction uses LayoutLM fine-tuned on your submission formats, feeding structured data to Claude for interpretation and risk narrative generation. FastAPI handles the orchestration layer, with Celery managing async enrichment calls to external data sources — ISO/NISS loss history, D&B financials, and regulatory status APIs. The agent integrates directly with your policy admin and rating systems via configured API connections, running submissions through your existing algorithm rather than replacing it. Setup takes 4–5 weeks: extraction training on your submission types, API configuration, and integration with your existing tech stack.
Ingests broker packages in whatever format they arrive — ACORD forms, loss run PDFs, schedule spreadsheets, financial statements. LayoutLM extracts structured fields across varied layouts without requiring standardized templates. Output is a normalized risk object ready for enrichment and rating.
Pulls property data, ISO/NISS loss history, D&B financial profiles, and regulatory status automatically during submission review. Each external call is logged with source and timestamp for audit trails. Underwriters get a complete risk profile without running manual lookups across five different systems.
Runs enriched submission data through your existing rating algorithm and returns a preliminary indication with the assumptions made at each step. Claude generates a plain-language summary alongside the numbers so underwriters can verify the logic, not just accept the output. Assumptions are editable before the indication is finalized.
Scans each submission for adverse loss trends, geographic concentration risk, coverage-to-exposure gaps, and internal inconsistencies across the broker package. Flags are prioritized by severity with supporting data — not just a warning, but the specific loss runs or financials that triggered it. Underwriters focus attention on submissions that actually need it.
Anonymized work with hard metrics — NDA-bound, no client names.
72%
Processing Time Reduction
94%
Classification Accuracy
1.2 days
Avg Processing Time
“The classification consistency was the biggest operational win. Adjusters are now working from standardised severity assessments rather than making independent calls on damage they have never seen before.”
— Head of Claims Operations, Motor Insurance Division
We custom-build each agent to fit your data, your rules, and your existing systems.
Free 30-min scoping call
Fordel Studios