- Complexity: Advanced
- Deployment: HeyQuinn recommended
The problem
Reinsurance underwriters extract catastrophe loss data from cedant submission packs by hand, one Excel workbook at a time. Every cedant formats its submission differently, so there is no standard sheet layout or column structure to rely on.
- Underwriters manually locate the relevant sheet in each submission pack
- Extraction errors introduce risk into pricing and risk models
- Matching new events against historical catastrophe records is manual and error-prone
- New records must be checked against existing entries before they update the loss database
- There is no consistent audit trail across manual steps
The solution
A four-agent pipeline extracts, matches, populates and compares catastrophe loss data end to end, while underwriters confirm or redirect the process at every handoff.
- Identifies the relevant sheet in each cedant submission pack and extracts event year, name and loss amount
- Fuzzy matches each extracted event against records in the historical catastrophe database
- Populates the cedant loss database with the structured records once matching completes
- Compares new records against existing entries and flags any changes for review
- Pauses at each stage boundary for underwriter confirmation before continuing
Where people stay in control
- An underwriter confirms or redirects execution at every agent handoff
- Each agent uses a narrow tool set to keep its reasoning focused on one task
- Flagged changes are surfaced for review before they are treated as final
Systems and data
The workflow reads cedant submission packs in Excel format and the firm's historical catastrophe database, and writes structured event records into the cedant loss database.
- Excel
- historical catastrophe database
- cedant loss database
Why HeyQuinn is recommended
The workflow coordinates four agents across sequential handoffs with persistent state and regulated financial data, so HeyQuinn is recommended to provide monitoring, audit traceability and controlled handoffs between agents.
Frequently asked questions
Does AI update the loss database without review?
No. An underwriter confirms or redirects the process at every handoff before records are written or changes are treated as final.
How does it handle submission packs with different formats?
A parser agent identifies the relevant sheet in each pack using extraction rather than a fixed template, so it adapts to each cedant's format.
Why use multiple agents instead of one?
Each agent works with a narrow tool set focused on one task, which keeps its reasoning focused and avoids errors from an overloaded context.