← Industries/InsurTech
Promethean for InsurTech
AI Act §5(c) explicitly classifies AI for risk-assessment + pricing in life and health insurance as high-risk. IDD (Insurance Distribution Directive) demands product-oversight evidence. EIOPA AI-supervisory expectations push toward per-decision explainability. Promethean is the evidence layer that satisfies all three without restructuring the underwriting platform.
Who this is for
Compliance + product teams in insurtech SMBs (digital insurers · claims-AI · underwriting platforms · MGAs). Typically:
- Seed–Series C insurtech with 10–80 engineers; 1–4 AI features (claims, underwriting, fraud, pricing).
- Operating under MGA or full insurer licence in 1–8 EU member states.
- Decision volume: 10k–500k per month per active product (quotes, claims, fraud screens).
- Reinsurer + regulator demand product-oversight evidence beyond what a standard MGA-platform provides.
- Often selling to larger carriers — those carriers' procurement requires per-decision audit trails.
The regulatory pressure
EU AI Act §5(c) Annex III — risk assessment + pricing (life + health)
AI for risk-assessment and pricing in life and health insurance is high-risk. Full Chapter II applies: record-keeping (Art. 12), human oversight (Art. 14), accuracy/robustness (Art. 15), technical documentation (Annex IV).
Deep-dive →IDD Art. 25 — product oversight + governance (POG)
Insurance distributors + manufacturers must operate a product-oversight + governance process. AI-driven pricing or distribution decisions require traceability per decision.
EIOPA AI Supervisory Statement (2021+)
European Insurance and Occupational Pensions Authority has issued supervisory expectations around AI use including explainability, governance, robustness. Per-decision evidence underpins all three.
GDPR Art. 22 — solely-automated decisions
Claims declined + premium pricing decisions clearly produce legal effects. Solely-automated decisions either need explicit consent or contractual necessity + meaningful human intervention.
Deep-dive →Where LLMs typically run in InsurTech
Claims triage + initial decision
Classifier producing settle / investigate / decline + rationale per claim. Closed-enum output; reviewer gate on schema violations + always-human on high-value claims.
Underwriting risk scoring
Classifier producing risk band + recommended premium adjustment. specHash captures the underwriting model version per decision; supervisor can request reproducibility on demand.
Fraud-flag classification
Per-claim fraud-risk score with closed-enum verdict + confidence. Reviewer gate on low-confidence; the chain shows fraud detection rate and false-positive rate over time.
Quote narrative drafting
Drafter producing the explanation accompanying an automated quote. Always-human reviewer gate on high-value quotes (broker review); reviewer-verdict captured.
How the substrate maps to your audit
| Your LLM feature | What the regulator asks | Promethean evidence |
|---|---|---|
| Claims triage | Per-claim decision + reviewer engagement; AI Act Art. 14 + IDD Art. 25. | L12 entry with reviewerVerdict + specHash + modelIdentity per claim. |
| Underwriting pricing | GDPR Art. 22 + AI Act §5(c) — reproducibility + explainability per decision. | specHash + inputHash bind the configuration + inputs; chain export reproducible. |
| Fraud classification | EIOPA explainability + AML-related explainability for SAR filings. | outputCanonicalHash + schema-valid flag; per-entry rationale committed. |
| Annual product-oversight review | IDD POG — annual report on the product's distribution decisions. | Time-bounded chain export with category + verdict + reviewer statistics. |
| AI Act Annex IV documentation | Technical documentation per high-risk system. | Handoff packet (Enterprise tier) provides the Annex IV-mapped contribution. |
Which Promethean tier fits
Recommended for typical SMBs in InsurTech
Production€499 / month flat
Up to 25 specs · 1M entries / month · hourly OTS anchoring · multi-tenant · federation read-only.
Production tier covers typical insurtech SMB profile (1M entries/month, 25 specs, multi-tenant for carrier customers). When reinsurer or large-carrier customers demand the AI Act Annex IV documentation generator or framework-handoff packets, upgrade to Enterprise (€2,499/mo) for the all-7-framework templates + on-prem option (some traditional carriers require on-prem).
Larger InsurTech operators with multi-tenant or framework-template needs upgrade to Enterprise (€2,499 / month flat).
What this looks like in practice
Hypothetical: an MGA digital-insurance SMB during EIOPA supervisory inspection
An MGA running an AI underwriting + claims-triage stack gets an EIOPA-coordinated national supervisory inspection. The supervisor wants: (1) explainability evidence per claim decided in Q3, (2) demonstration that high-value claims went to human review, (3) per-decision pricing reproducibility under GDPR Art. 22 challenges. The MGA exports the relevant L12 chain slice; the supervisor runs verify.mjs against it with the workspace's public key. The reviewer-verdict distribution shows always-human on high-value, low-confidence on the rest. The specHash sequence shows no silent model swaps. Exit code 0. The inspection moves from 'do you have evidence' to 'does this evidence support your governance claim'.
Frequently asked
We're an MGA, not a full insurer. Does Promethean still apply?
Yes. The IDD applies to insurance distributors (including MGAs) and manufacturers. Your underwriting + claims AI is still subject to product-oversight obligations even if you're not the risk carrier. AI Act §5(c) applies to the AI system regardless of who legally underwrites. Your carrier partners increasingly demand per-decision audit trails because their own supervisor pressure cascades down.
How does this interact with Solvency II requirements?
Solvency II covers the carrier's capital + risk-management framework, not the per-decision AI evidence. But the carrier's ORSA (Own Risk and Solvency Assessment) under Article 45 requires evaluating the risks from outsourced or AI-supported decision-making. Promethean's L12 chain feeds the operational-risk component of the carrier's ORSA. Your MGA is a third-party ICT provider under DORA (Art. 28) for the carrier — you'll be asked for the audit trail eventually.
Does Promethean handle the claims-LLM hallucination risk?
The substrate doesn't prevent hallucinations — that's an LLM-side problem. It bounds the model with closed-enum schemas (the spec says 'output must be one of settle/investigate/decline + a confidence float + a rationale string ≤ 280 chars'). Output that violates the schema triggers the fallback handler (typically 'queue for human review'). The L12 entry records schemaValid: false and fallbackTriggered: true. Over time, the fallback rate is a directly regulatable proxy for model reliability.
Can we use Promethean for AML / suspicious-transaction-report drafting?
Yes. The drafter category handles SAR-narrative generation. Always-human reviewer gate is the appropriate setting (compliance officer signs each filing). The L12 entry captures the spec, the input hash, the output canonical hash, and the reviewer verdict. AML supervisors auditing SAR-filing patterns get a verifiable chain of who decided what, when, under which model.
Procurement at our carrier partners is sometimes 12+ months. Will Promethean's pricing pass procurement?
Production tier (€499/mo = €5,988/year) and Enterprise (€2,499/mo = €30k/year) are deliberately set below most procurement thresholds for a single PO (typically €25-50k). A compliance team or CISO can usually expense Promethean as a controls-strengthening tool without committee review. Larger carriers may insist on the framework-template handoff packets — which are Enterprise tier.
Drill into a specific use case
Each InsurTech use case has its own page with the regulation-specific evidence mapping + hypothetical scenario:
Definitions used on this page
The substrate primitives referenced above (L12 receipt chain, spec hash, reviewer gate, fallback behaviour, OpenTimestamps anchor, tenant ID) all have canonical definitions in the glossary: