Build and own a Reinsurance Recovery Specialist Model.
Can the carrier recover this loss from reinsurance, and if so, how much, from which treaty, and with what supporting documentation?
Overview
Faster identification of claims that qualify for reinsurance recovery instead of relying on manual treaty-by-treaty review.
Who owns this decision
Head of Reinsurance, VP Ceded Reinsurance
Why owned AI may fit
Treaty language, attachment points, and exclusions are carrier- and treaty-year-specific, requiring the actual treaty documents as context rather than general reinsurance knowledge.
What this Specialist Model actually does
Decision: Can the carrier recover this loss from reinsurance, and if so, how much, from which treaty, and with what supporting documentation?
This is a larger system with 3 specialist model candidates and 2 supporting or external components.
Who uses this Specialist Model
- Head of Reinsurance
- VP Ceded Reinsurance
- Chief Financial Officer
- Reinsurance Accounting Director
Where the specialist model sits
This is a larger system with 3 specialist model candidates and 2 supporting or external components.
Structured extraction of treaty terms, attachment points, limits, and exclusions from treaty documents
Specialist Model
Structured extraction of claim facts, reserves, and payment history from claims systems
Specialist Model
Model-based reasoning layer
Specialist Model
Model-based reasoning layer to justify recovery eligibility and amount with citations to treaty clauses
More Specialist Models in Insurance →
Matching engine
Supporting component
Matching engine to align claims against applicable treaties and layers
How Oumi connects to your stack →
Ceded claims analyst review workflow
Supporting component
Ceded claims analyst review workflow for flagged or disputed recoveries
How Oumi connects to your stack →
How to evaluate and improve it
Evaluation
- Treaty-match accuracy (correct treaty and clause identified)
- Expected recovery amount accuracy against actual collected amount
- Recovery-eligible classification precision and recall
Feedback signals
- Actual recovery amount collected versus expected recovery amount
- Reinsurer dispute or denial of a recommended recovery
- Ceded claims analyst overrides of treaty matching
- Audit findings on recoverables aging and collectability
- Renegotiated or clarified treaty terms following a dispute
Model candidates to evaluate
- NuExtract 3 (Apache 2.0)
- Qwen3.5-27B (Apache 2.0)
- DeepSeek V4-Flash (MIT)
These are candidates, not performance claims. Run them against your evaluation set.


