SUSEP Circular 621Insurance product governance
Defines requirements for product design, distribution and supervision. Dynamic pricing models need non-discriminatory actuarial grounding, with features defensible before SUSEP.
Applied AI for life, insurance.
Underwriting, claims, support and prevention — AI applied end to end across the insurance lifecycle.
Overview
What we deliver in this vertical.
Life · Auto · Health · Property
Insurance
Underwriting, claims, support and prevention — AI applied end to end across the insurance lifecycle.
−41%claim time
+29%fraud detection
+12ppNPS
Applications
- 01Automated underwriting
- 02Computer-vision claims
- 03Claim fraud detection
- 04Dynamic pricing
- 05Omnichannel support
Positioning
Insurance with AI isn't cheaper — it's more precise.
Manual underwriting, slow claims and retention support are gaps AI closes without disfiguring the product. What changes isn't the offer — it's the time between customer want and policy response. When that drops from days to hours, NPS rises and churn falls. Everything else is marginal.
By the numbers
AI's impact on claims, underwriting and service.
−0%auto claim time with computer vision
+0%claim fraud capture with combined model
+0ppNPS in mass-market underwriting
−0%operational cost in low-complexity lines
Under the hood
A stack aligned to the insurance cycle.
Typical stack for mid-size insurers in mass-market lines (auto, life, residential). Each component was chosen to satisfy SUSEP and respect the regulated actuary.
Computer vision
- Detectron2 + ONNXClaim photo analysis.
- Fine-tuned ResNetDamage severity classification.
- Segment Anything (SAM)Automatic damaged-area segmentation.
Actuarial NLP
- Proprietary narrative modelFraud triage by narrative pattern.
- Policyholder-history embeddingRisk similarity.
- Regex + LLM extractionOn reports, contracts and statements.
Underwriting
- Combined model (actuary + ML)Actuary validates features regulatorily.
- Regulatory feature storeTo justify pricing to the regulator.
- Drift by policy cohortReflects portfolio shifts.
Compliance
SUSEP, LGPD and the regulated actuary.
Insurance AI has stricter boundaries than other financial markets. Dynamic pricing requires actuarial justification; automated underwriting needs documented human review. We treat them as design constraints.
SUSEP · AIAI guidelines
SUSEP governance roadmap for AI use in insurance: explainability, model management, bias and human review in adverse decisions. Regulator-format model-cards.
LGPDBrazilian Data Law
Automated underwriting is a decision with legal effect — requires human-review right, documented legal basis and DPIA per product. Default in every onboarding.
CNSP 543National Insurance Council Resolution
Defines governance and risk management obligations for insurers. Critical models enter the operational risk inventory with a contingency plan.
In production
Results in insurance.
Claim time · auto
Computer vision on claim photos + estimation model. Simple claims processed in hours instead of days.
Claim fraud detection
Model combining policyholder history, claim patterns and textual analysis. Catches 29% more fraud without increasing friction on legitimate claims.
Underwriting NPS · life
Automated underwriting on mass-market products reduced friction. NPS rose 12pp with underwriting integrity preserved.
04 · LA AI Method
From hypothesis to operation.
Five steps that prevent eternal pilots. Each stage has deliverables, metrics and decision gates.
M1
Diagnostic
Opportunity map, estimated value, data readiness.
2–3 wksM2
Strategy
Prioritized roadmap, reference architecture, governance.
2 wksM3
Proof of Value
POC with production criteria and business metric defined.
4–6 wksM4
Deployment
Model, integrations, security, observability and UX.
6–12 wksM5
Continuous Operations
Evolution, fine-tuning, eval suite, cost per use, new features.
RecurringFrequently asked
About AI in insurance.
How does the regulator view dynamic AI pricing?
Regulators require actuarial and non-discriminatory justification for pricing factors. We work with the actuarial team so model features are defensible.
Does it work in regulated health insurance?
Health has more restrictions. Common use cases: mass-market underwriting, claim management, support. Pricing is more constrained.
Does claim computer vision require high-quality photos?
Works with standard smartphone photos. The system flags when an additional photo or specific angle is needed, helping the policyholder provide what's required.
Time from POC to production?
Average 4-8 months for mass-market products, longer for complex underwriting.