AI Insurance Research Lab

Open benchmarks, playbooks, and quarterly insights to accelerate safe, effective agent adoption.

Where insurance-domain rigor meets agentic AI engineering. Reproducible benchmarks, governance playbooks, and a readiness index—so you can adopt faster, safer, and audit-ready.
View Benchmarks
Browse Playbooks
Get Full Readiness Report
Open Benchmarks
Intake, coverage, explainability, fairness.
View benchmarks (Coming Soon) →
Playbooks
Adverse action, rationale templates, HITL.
Browse playbooks (Coming Soon) →
Readiness Index

Assess org readiness across dimensions.

7.8 / 10
Request assessment →
Quarterly Insights
Trends, architectures, anonymized learnings.
Read latest issue (Coming Soon) →
Open Benchmarks (Insurance-Native)

Each benchmark ships with datasets/specs, metrics, scoring logic, and governance notes. 

Document Intake & Normalization

Tasks
  • ACORD/email ingestion
  • Exposure extraction
  • SOV parsing
Metrics
  • F1 / Exact-match
  • Latency P50/P95
  • Hallucination rate

Coverage & Exception Analysis

Tasks
  • Coverage mapping
  • Limits & deductibles
  • Exclusions detection
Metrics
  • Precision/Recall
  • Exception recall
  • Rationale faithfulness

Quote Comparison & Explainability

Tasks
  • Multi-carrier delta
  • Price/terms rationale
Metrics
  • Delta accuracy
  • Rationale consistency

Underwriting Signals & Referral Reasoning

Tasks
  • Rule hit explanations
  • Referral packaging
Metrics
  • Evidence citation rate
  • Referral precision

Claims FNOL & Triage

Tasks
  • Photo/document triage
  • Vendor selection suggestions
Metrics
  • Top-k accuracy
  • Time-to-decision
  • Leakage proxy

Fairness & Robustness Checks

Tasks
  • Shift tests
  • Proxy-bias probes
  • Prompt-attack resilience
Metrics
  • Δ-performance
  • Robustness score
  • Red-team pass-rate

Suitability & Adverse Action

Tasks
  • Suitability scoring
  • Adverse language templates
Metrics
  • Appeal overturn rate
  • Clarity/consistency

Policy Service Automation

Tasks
  • Loans
  • Withdrawals
  • 1035 exchanges
Metrics
  • Cycle time
  • STP
  • NIGO

Treaty & Bordereaux QA

Tasks
  • Data conformance
  • Exception surfacing
Metrics
  • Field-level accuracy
  • Exception recall

Cat/Event Analytics

Tasks
  • Exposure roll-ups
  • Driver explanation
Metrics
  • Attribution quality
  • Latency
Readiness Index (Assessment)

Measure your posture and get a gap-aligned plan.

7.8
Overall / 10

Request assessment
Governance
69%
Data & Tools
74%
People & Skills
65%
Change & Adoption
71%
Playbooks (Governance-First)
Production-grade templates you can adopt as-is or tailor.
Governance & Risk
  • NAIC/NYDFS/CO control mapping
  • Vendor risk & ECDIS kit
  • Model inventory & lifecycle
Open →
Explainability & Language
  • Adverse action templates
  • Referral rationale scaffolds
  • Appeals flow
Open →
Operations & HITL
  • Authority matrices
  • HITL patterns
  • Rollback rehearsals
Open →
Data & Privacy
  • Redaction/masking policies
  • PHI/PII SOPs
  • Synthetic data recipes
Open →
Quarterly Insights
Trends, reference architectures, and anonymized field learnings.

Where STP Works, Where Agentic AI Excels, and Where HITL Must Stay

Decision Segmentation and Cost Curves for Document-Heavy Workflows in the Age of Agentic AI
Read →

Reference Architectures for Governed RAG/Agents

Patterns for claims triage and UW explainability packs in regulated stacks.
Read →

Benchmark Trends: Accuracy vs. Latency

How carriers balance P95 latency with explainability and reviewer throughput.

Read →
Methodology & Assurance
  • Standards alignment: NAIC, NYDFS, Colorado SB21-169, NIST AI RMF
  • Artifact Factory: run cards, decision traces, explanation bundles, audit-ready PDFs
  • Reproducibility: deterministic harness, seed control, prompt/version pinning
  • Privacy-by-Design: redaction defaults, role-based access, synthetic data fallbacks
Participate
  • Contribute data specs (we can synthesize if real data can’t be shared)
  • Submit models/agents to the leaderboard (private eval lanes available)
  • Co-publish anonymized findings with Perpendo Research
Join the Lab
Submission Guidelines

Is this model ranking?

It’s use‑case benchmarking with governance signals, not just raw model scores.

Can we run it internally?

Yes—on‑prem private harness with the same metrics and artifact factory.

What about sensitive documents?

Use redaction + synthetic recipes; we provide contracts and SOPs for PHI/PII.

Ready to test, learn, and scale—safely?

Kick off with benchmarks, tailor playbooks, and get your readiness plan in days—not months.
Start Assessment
Talk to an Expert