Use case Insurance · Adolphsplatz · BaFin + EIOPA + PRA

Claims triage and Solvency II reporting with Decision-Layer - Composite, Life, Marine. Lloyd's syndicate-ready.

Solvency II QRT reporting and claims triage with audit trail before BaFin and PRA. Composite, Life, Marine - three regulatory worlds, one Decision-Layer pattern. Lloyd's of London + UK divergence handled.

Workshop at Grindelberg All 7 use cases ←

Chapter 1 — Three regulators, one Decision-Layer

Solvency II + IFRS 17 + EU AI Act simultaneously - but per step deterministically assigned.

Hamburg insurers at Adolphsplatz carry three parallel regulatory worlds: Solvency II (BaFin reporting obligation, EIOPA validation rules per quarter), IFRS 17 (since 2023 for listed and group-consolidated insurers), EU AI Act (Annex III Point 5(c) for Life/Health insurance, plus GDPR Art. 22 for every automated underwriting decision). For UK-headquartered groups with German subsidiaries: add PRA Supervisory Statement SS1/23 on model risk management.

Operational reality: quarterly QRT reporting with over 740 data points, of which 38 manual overrides from claim files. 4-day sprint with 22 actuarial staff overtime at quarter-end. Plus IFRS 17 dual-booking. Plus BaFin AI Orientation Guide (May 2025) for every underwriting decision. Plus the Hamburg data protection authority imposed a EUR 492,000 fine on a financial institution in 2025 - automated rejection without justification generation. Same pattern hits any underwriting rejection.

Decision-Layer split typical for claims triage at a mid-sized composite insurer: 40 percent RULES (limit check, deductible check, contract scope), 35 percent AI AUTONOMOUS (claim file classification, survey triage, language variants on international claims), 25 percent HUMAN (discretionary decision by underwriter, major claims above value threshold, discrimination escalation).

Marine insurance at Adolphsplatz has its own sister chain: P&I Club contracts, IUMI statistics, classification societies (Lloyd's Register, DNV, BV, ClassNK), tanker incidents with 50+ involved parties. For Lloyd's slips on shared placements (London + Hamburg co-leads): Lloyd's binders for delegated authority + Box Working Party coordination. Different anchors, same pattern. Detail in Chapter 3.

Chapter 2 — Decision-Record for composite claims triage

How a motor third-party liability claim notification is triaged in the Decision-Layer.

Anonymised decision-record for an incoming motor third-party liability claim notification at a composite insurer headquartered at Adolphsplatz. Standard claim (no major loss, no personal injury) - 70 percent of all triages run like this.

CLM-MTR-2026-05-15-08:42:13-DE

Motor third-party liability · Received 15.05.2026 08:42 · Claim amount indication EUR 4,200 · No personal injury

Result Triaged · Standard handler pool
  1. 01 REGEL

    Contract eligibility check

    Policy number extracted from claim notification. Contract active, insurance event in scope (Germany, standard cover). Deductible EUR 150. Rule motor_eligibility_v3.1.

    ✓ Notification valid
  2. 02 REGEL

    Claim amount threshold

    Indication EUR 4,200 < major-claim threshold EUR 10,000. Standard triage pool. Above EUR 10,000 or personal injury: direct underwriter routing. Rule claim_amount_v2.4.

    ✓ Standard pool
  3. 03 KI

    Claim file classification

    Incoming documents (loss report, 4 photos, police report) classified. Claim type: rear-end collision, property damage rear. Model motor-claims-classifier-v3.7.

    Confidence 0.94 · threshold 0.85

    ✓ Rear-end collision classified
  4. 04 KI

    Claim amount plausibilisation

    Workshop quote EUR 4,200 mapped against make/model repair cost database. Plausible in corridor EUR 3,800-4,600 for rear damage on this model. Model repair-cost-validator-v2.1.

    Confidence 0.91 · threshold 0.85

    ✓ Plausible
  5. 05 REGEL

    Bias pre-check (GDPR Art. 22 + PRA SS1/23)

    Statistical parity against prior-year triage quota by postcode, age, insurance duration. Deviation > 12% triggers underwriter escalation. Rule bias_pre_v2.0 (based on BaFin Guide May 2025 + PRA SS1/23 for UK divergence).

    ✓ Within tolerance corridor
  6. 06 REGEL

    Handler assignment

    Standard pool, routing by workload. Handler receives structured data set with classification, plausibility range and customer-letter draft. Rule routing_standard_v1.3.

    ✓ Handler B.
  7. 07 KI

    Customer letter draft (justification template)

    Standardised letter with claim type, plausibility corridor and right of challenge. Handler reviews and sends. Model customer-letter-v1.4.

    ▲ Handler reviews
  8. 08 REGEL

    Solvency II reserving

    Claim reserve per plausibility range plus safety margin per Solvency II methodology (Best Estimate + Risk Margin). Booking to SCR account. Rule sii_reserving_v4.2.

    ✓ Reserve EUR 4,620 booked
  9. 09 REGEL

    Annex III compliance log + audit trail

    EU AI Act Art. 12 logging plus GDPR Art. 22 access-right preparation: complete decision-record persisted with model version, input hash, confidence scores, escalation path. Immediately exportable on challenge. Rule audit_v1.4.

    ✓ Audit trail persisted

Chapter 3 — Marine insurance as sister decision chain (Lloyd's syndicate-ready)

P&I Clubs, classification societies, 50+ parties per tanker incident, Lloyd's binders - same pattern.

Marine insurance at Adolphsplatz has its own anchors: P&I Clubs (Protection & Indemnity, e.g. Skuld, Britannia, Gard, North Standard), IUMI (International Union of Marine Insurance) for statistics and market standards, classification societies (Lloyd's Register, DNV, Bureau Veritas, ClassNK) for technical assessment. For Lloyd's syndicate work (London + Hamburg co-leads): Lloyd's binders for delegated authority arrangements, Box Working Party coordination, Lime Street placement workflow. Plus Solvency II + EU AI Act Annex III Point 5(c) where Life/Health lines are involved.

Decision-Layer pattern identical to composite claims triage. RULES: contract terms check, P&I Club membership verification, classification status of affected ship registration, Lloyd's binder authority limits. AI AUTONOMOUS: survey reports classification (Lloyd's format, DNV format, BV format), multilingual master's reports, claim amount triage. HUMAN: major claims above EUR 1M, tanker casualties with environmental damage, disputes between shipowner and P&I Club, Lloyd's referral committees.

Tanker incident in Hamburg port: A typical container loss or cargo damage incident involves 50+ parties. Current workflow: SharePoint folder with 400+ files, handler searches 6 hours before triage is possible. Decision-Layer extracts triage-ready data set from incoming documents in 4 hours, classifies surveys by claim type, identifies contractual relationships between parties, and triages by underwriter responsibility. For Lloyd's slips: parallel routing to London box, audit trail compatible with PRA reporting requirements.

Workshop at Grindelberg addresses both worlds: composite underwriters in main room with BaFin-relevant decision-chain demos, Marine underwriters in sister session with P&I Club contracts and IUMI statistics. For Lloyd's syndicate clients: dedicated session on Lloyd's binders + UK PRA SS1/23. Engineering on-site, head office Hallerstraße 8 five minutes from Adolphsplatz cluster (Hamburg Insurance Exchange). Source code, prompts and rule sets transferred to insurer by repository handover, contractually.

Related use cases

Frequently asked questions

Does the spoke address composite insurers or only Marine?
Both. Composite insurers (motor, property, liability, accident) are the primary target group - Marine is a specialised case with its own anchors (P&I Clubs, IUMI, classification societies, Lloyd's syndicates). The Decision-Layer architecture carries both: same pattern (eligibility, claim-amount threshold, bias pre-check, human escalation, audit log), different rule sets (motor liability table vs. P&I contract terms). For German subsidiaries of UK insurers, both worlds run on one architecture.
How does Solvency II QRT reporting accelerate through the Decision-Layer?
Solvency II QRT (Quantitative Reporting Templates) have over 740 data points per quarter, typically 38 manual overrides from claim files. The Decision-Layer extracts structured data from surveys, claim notifications, expert reports - and assigns them to EIOPA validation rules. Manual overrides remain human-mandatory (escalation), but every position gets an audit trail with rule version. At BaFin audit (and PRA audit for UK divergence), the path crosses the table, not the result. EIOPA validation rule updates per quarter stored as versioned rule sets.
EU AI Act Annex III for insurance risk assessment - what does it mean concretely?
Annex III Point 5(c) covers AI systems for risk assessment and pricing in Life and Health insurance as high-risk. Composite (motor, property, liability) is NOT explicitly covered - but once underwriting AI carries discrimination risk (e.g. postcode discrimination in motor), GDPR Art. 22 and BaFin AI Orientation Guide (May 2025) apply. UK parallel: PRA Supervisory Statement SS1/23 on model risk management + ICO Article 22. Decision-Layer architecture meets both frames: for Life/Health automatic high-risk handling, for composite risk-graduated with discrimination pre-check per decision.
How does the Decision-Layer solve the IFRS 17 vs Solvency II dual-booking problem?
IFRS 17 (since 2023) requires different valuation of insurance contracts than Solvency II - Building Block Approach for Best Estimate, Risk Adjustment, Contractual Service Margin. Actuarial teams at Hamburg insurers rebuild Excel models quarterly for both standards. The Decision-Layer stores both valuation logics as versioned rule sets in parallel - one input (contract portfolio, claim files, discount rates), two outputs (IFRS 17 position, Solvency II SCR). Consistency technically enforced, audit trail documents per position which assumptions were used. For Lloyd's syndicates: IFRS 17 + Solvency II UK + Lloyd's Performance Management Standards triple-stack handled in same architecture.
How is the major-claim triage problem with 50+ involved parties addressed (Marine focus)?
A tanker incident in Hamburg port typically involves 50+ parties (shipowner, charterer, freight forwarder, P&I Club, classification society, port authority, surveyors, customs, police, plus Lloyd's syndicate underwriters where Lloyd's is on the slip). Current workflow: SharePoint folder with 400+ files per case, claim handler searches 6 hours before triage is possible. Decision-Layer extracts structured claim records from incoming documents, classifies survey reports by claim type and severity, and triages by underwriter responsibility. Human-mandatory at discretionary decisions, AI-triage at classification. Audit trail per document. Same architecture handles Lloyd's binders for delegated authority arrangements.

Schedule workshop at Grindelberg

3-day discovery: Day 1 process analysis, Day 2 Decision-Layer mapping, Day 3 use-case prioritisation. Concrete deliverable.

Schedule meeting

Discovery workshop below EUR 10,000. Pilot fixed price discussed after the workshop.

← Back to overview: AI Agents Hamburg (7 use cases)